A V Balakrishnan Aeroelasticity The Continuum Theory

Aeroelasticity
A V Balakrishnan
Aeroelasticity
The Continuum Theory
123
A V Balakrishnan
Department of Electrical Engineering
Department of Mathematics
University of California
Westwood Blvd.
Los Angeles, California
USA
ISBN 978-1-4614-3608-9 ISBN 978-1-4614-3609-6 (eBook)

DOI 10.1007/978-1-4614-3609-6
Springer New York Heidelberg Dordrecht London
Library of Congress Control Number: 2012938709
Springer Science+Business Media, LLC 2012

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publishers location, in its current version, and permission for use must always be obtained from Springer.
Permissions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations
are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Foreword
The author is a noted applied mathematician and control theorist who has in recent
years turned his attention to the subject of aeroelasticity. The present volume is
the result of his research and provides an illuminating new view of the field.
As he emphasizes in his introduction, his approach is one of continuum models
of the aerodynamic flow interacting with a flexible structure whose behavior
is governed by partial differential equations. Both linear and nonlinear models
are considered although much of the book is concerned with the former while
keeping the latter clearly in view and indeed a complete chapter is devoted to
nonlinear theory. The author has provided new insights into the classical inviscid
aerodynamics described by the celebrated Possios equation and raises novel and
interesting questions on fundamental issues that have too often been neglected or
forgotten in the development of the early history of the subject. The author contrasts
his approach with discrete models that have gained enormous popularity in the
practice of aeroelasticity such as the doublet lattice model of Rodden et al. for
unsteady aerodynamic flow and the finite element model for the structure. Much
of aeroelasticity has been developed with applications firmly in mind because of
its enormous consequences for the safety of aircraft. Aeroelastic instabilities such
as divergence and flutter and aeroelastic responses to gusts can pose a significant
hazard to the aircraft and affect its performance. Yet it is now recognized that
there are many other physical phenomena that have similar characteristics ranging
from flows around flexible tall buildings and long span bridges to flows internal
to the human body such as blood flows through arteries and air flow over the
tongue, and the author touches on these topics as well. For the theorist and applied
mathematician who wishes an introduction to this fascinating subject as well as for
the experienced aeroelastician who is open to new challenges and a fresh viewpoint,
this book and its author have much to offer the reader.
Durham, USA Earl Dowell
v
Preface
Aeroelasticity deals with the dynamics of an elastic structure in airflow with primary
focus on the endemic instability of the structure called flutter that occurs at
high enough speed. This book presents the continuum theory in contrast to
extant literature that is largely computational; where typically one starts with the
basic continuum model, a partial differential equation usually highly nonlinear
but omitting the all important boundary conditions and disregarding the question
of existence of solution; going immediately to the discretized approximation;
presenting charts and figures for a confluence of numerical values for the parameters
and conclusions drawn from them.
Here we stay with the basic continuum model theory until the very end, where
constructive methods are developed for calculating physical quantities of interest,
such as the flutter speed. Indeed this is considered mission impossible because it
is nonlinear and complex.
As in any scientific discipline, continuum theory provides answers to what if
questions which numerical codes cannot. It makes possible precise definitions
such as what is flutter speed. Physical phenomenasuch as transonic dip, for
examplecan be captured by simple closed-form formulae. And above all it can
help develop intuition based on a better understanding of the phenomena of interest.
As with any mathematical theory it enables a degree of generality and qualitative
conclusions, increasing insight.
But the use of continuum models comes with a price: it requires a high
level of abstract mathematics. For a precise statement of the problem, however,
the language of modern analysisdeveloped in the latter half of the twentieth
centuryabstract functional analysis, in particular, the theory of boundary value
problems of partial differential equations, is unavoidable. Indeed the aeroelastic
problem, the structure dynamics in normal air flow-formulates as a nonlinear
convolution/evolution equation in a Hilbert space.
On the other hand the numerical range of the physical parameters plays an
important role in being able to generate constructive solutions otherwise impossible
vii
viii Preface
from the mathematics alone. What we do is indeed applied mathematics in the

sense that we use mathematics to solve todays engineering problems addressed
to engineers as well as mathematicians.
And now for some points of view, points of departure, of this book closer to the
subject matter. Aeroelasticity is concerned with the stability of the structure in air
flow. The air flow per se is of less interest. Thus we are not concerned, for example,
whether there are shocks in the flow or not, in itself a controversial matter. The faith
of the aeroelasticians in shocks, it turns out, is not substantiated by the mathematical
theory (2D or 3D flow). It may be heresy to the clan but shocks may exist that do
not affect the stability (or rather the instability) of the structure. Another and more
significant view concerns the interaction between Lagrangian structure dynamics
and Eulerian fluid dynamics, often the most mysterious part of computational work.
Here we take the simple engineering inputoutput point of view where the
velocity of the structure is the input and the pressure jump across the structure is the
output. The inputoutput relation is the integral equation of Possio that does not get
any mention in as recent a work as [17] which features partial differential equations.
The Possio Integral Equation can be looked as an illustration of the Duhamel
principle and we make systematic use of itlinear and nonlinearthroughout the
book. We show that flutter speed is simply the smallest speed at which the structure
becomes unstable; it is a Hopf bifurcation point determined completely by the
linearized model about the steady state. In turn this means incidentally that the
control for extending the flutter speed need not be nonlinear, contrary to current
wisdom.
The mathematical style of the book is largely imitated/borrowed from that of
R.E. Mayer [14], and ChorinMarsden [4] where they claim to Present basic ideas
in a mathematically attractive manner (which does not mean fully rigorous).
In this sense although we use abstract functional analysis, we try to reduce the
abstraction and sacrifice mathematical generality, preferring to emphasize construc-
tive solutions and basic ideas rather than get lost in Sobolev spaces and weak
solutions. Quoting another pioneer in this style: I shall not be guilty of artificially
complicating simple matters. A phenomenon that sometimes occurs in mathematical
writing. Tricomi in his book Integral Equation, 1957 [11].
We should caution that there are many problems that mathematical theory cannot
currently answer especially in viscous flow and as a result also in aeroelasticity. We
invoke the Prandtl boundary layer theory, for example, with this caveat.
We should also note a price to be paid for mixing the abstract with the concrete,
saying too much or too little at either end.
Acknowledgments
It is a pleasure to acknowledge the generous help of my colleagues and my peers:

Earl Dowell (who was my chief mentor and also opened my eyes to the nonaircraft
applications)
John Edwards
Oddvar Bendiksen
Dewey Hodges
Ciprian Preda
Marianna Shubov
Roberto Triggiani
Amjad Tuffaha
My former students:
Jason Lin
Oscar Alvarez Salazar
Irena Lasiecka (Post Doc)
And finally the NSF monitor Dr. Kishan Baheti for his unfailing belief in the value
of the work and his help and encouragement throughout.
Los Angeles, California, USA A V Balakrishnan
ix
Contents
1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 1
2 Dynamics of Wing Structures . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9
2.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9
2.2 The Goland Beam Model . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 9
2.3 Time Domain Analysis.. . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 11
2.4 Structure Modes and Mode Shapes. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 18
2.5 Robust Feedback Control Theory: Stability Enhancement .. . . . . . . . 26
2.6 Nonfixed Wing Models: Flying Wings . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 35
2.7 Nonlinear Structure Models . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 40
2.8 Beams of Infinite Length.. . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 43
3 The Air Flow Model/Boundary Fluid Structure
Interaction/The Aeroelastic Problem .. . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47
3.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47
3.2 Notation/Physical Constants .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 47
3.3 Nonviscous Flow: The Euler Full Potential Equation . . . . . . . . . . . . . . 49
3.4 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 58
4 The Steady-State (Static) Solution of the Aeroelastic Equation . . . . . . . 65
4.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 65
4.2 Goland Structure Model . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 66
4.3 Linear Aeroelasticity Theory: The Finite Plane Case . . . . . . . . . . . . . . 72
4.4 The General Nonlinear Case . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 80
5 Linear Aeroelasticity Theory/ The Possio Integral Equation .. . . . . . . . . 103
5.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 103
5.2 Power Series Expansion of the 3D Potential . . . .. . . . . . . . . . . . . . . . . . . . 103
5.3 The Linear Possio Integral Equation . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 106
5.4 Linear Possio Equation: Compressible Flow: M > 0 . . . . . . . . . . . . . . 122
5.5 Linear Time Domain Airfoil Dynamics .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 155
5.6 State Space Theory .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 184
xi
xii Contents
5.7 Flutter Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 230

5.9 Appendix: Computer Program for Flutter Speed
in Incompressible Flow: Goland Model .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 260
6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics:
Flutter Instability as an LCO .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . 269
6.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 269
6.2 The Aeroelastic Equations: Linear Structure Model.. . . . . . . . . . . . . . . 270
6.3 The Nonlinear Possio Integral Equation . . . . . . . .. . . . . . . . . . . . . . . . . . . . 296
6.4 Nonlinear Aeroelastic Dynamics . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 298
6.5 Stability .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 315
6.6 Limit Cycle Oscillation . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 318
6.7 The Air Flow Decomposition Theory . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 323
7 Viscous Air flow Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 327
7.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 327
7.2 The Field Equation/Conservation Laws . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 327
8 Optimal Control Theory: Flutter Suppression . . . . . .. . . . . . . . . . . . . . . . . . . . 337
8.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 337
9 Aeroelastic Gust Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 349
9.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 349
9.2 The Turbulence Field. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 349
9.3 The Gust Forced Aeroelastic Equations .. . . . . . . .. . . . . . . . . . . . . . . . . . . . 351
9.4 Structure Response to Turbulence .. . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 362
9.5 Illustrative Example .. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 364
10 Addendum: Axial Air flow TheoryContinuum Models. . . . . . . . . . . . . . 367
10.1 Introduction .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 367
10.2 The Aeroelastic Equations . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 367
10.3 Steady-State Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 370
10.4 Power Series Expansion . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 370
10.5 Linear Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 376
10.6 Stability: Aeroelastic Modes .. . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 379
10.7 Linear Time Domain Theory: The Convolution/Evolution
Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 384
10.8 Nonlinear Stability Theory: Hopf Bifurcation/Flutter LCO .. . . . . . . 386
References .. .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 387
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 393
Chapter 1
Introduction
The subject matter of aeroelasticity is the dynamics of a structure in air flow,

traditionally the wing of an aircraft in flight. There are important nonaircraft
applications of more recent interest as well which we treat in Chap. 10. Thus it
involves both structural dynamics and fluid dynamics without being either. In this
sense it has an inherent identity complex but still has developed as a separate
discipline over the many years since flight began. The central problem is flutter,
a potentially destructive instability that occurs at any altitude as the speed is
increased to the flutter speed, the flutter boundary. Indeed the Federal Aviation
Administration mandates that a margin of 15% be maintained in all flight. And thus
flutter analysis is an ongoing activity in flight centers as well as in aircraft design.
Although in the aircraft context ultimately a flight test would be needed to verify
the flutter phenomena, flight tests are costly and need to be kept at a minimum. Wind
tunnel tests are much less expensive but paperwork is the least expensive. Hence the
importance of analysis! The ultimate objective is of course to minimize the number
of flight tests needed.
Here one has to distinguish between analytical (continuum) theory and (digital)
computation (FEM/CFD) which is predominant in all the current work. Both
structure and aerodynamic models originate as continuum models and digital
computation needs perforce to approximate the continuum: convert the distributed
parameter system to a lumped parameter system and replace the continuum
partial differential equations by ordinary differential equations, however high the
order. And of course digitization discretizes time and amplitude. Furthermore the
system parameters must be specified numerically, numeric computation as opposed
to symbolic computation. This restricts generality and can inhibit qualitative
understanding of phenomena. Perhaps another way of putting it is to say that
in this work we do not truncate the model at the beginning; we work with
the continuum models or symbolic software and approximation as needed only
eventually at the end-level calculation of specific functionals (such as flutter speed
as a function of system parameters). We approximate the solution, rather than
approximate the equations. Phenomena occur at the continuum level that do not
occur in any solution of the discretized equations, however fine the discretization.
A V Balakrishnan, Aeroelasticity: The Continuum Theory, 1

DOI 10.1007/978-1-4614-3609-6 1, Springer Science+Business Media, LLC 2012
2 1 Introduction
In the computational approach the discretization is made without any concern about
whether the continuum equations have a solution and whether are unique. Moreover
there is no attempt to prove that there is a solution to the problem in the language
of the starting continuum formulation. The digital computer solution is accepted as
a good enough approximation to the solution. This can be particularly misleading
when the continuum equations are shown to have more than one solution or no
solution at all. Indeed it is typical in extant literature (see the List of Papers) in
aeroelasticity where complex continuum structure models are introduced without
any consideration of whether they have any solution and are immediately converted
to discretized computational models (FEM). This is also true of the fluid dynamic
models where the partial differential equations are introduced without any precise
statement of the (fluid/structure) boundary conditions.
Thus we emphasize that a distinguishing feature of this treatment is that a precise
statement of the aeroelastic problem is made especially at the LagrangianEuler
fluid structure boundary; the dynamic equations are shown to have a unique solution
and qualitative results are deduced and then only numerical computation is made for
the functionals of interest, without discretizing the equations at any level.
The material has been distributed into 10 chapters in reasonable logical
progression and are listed in the Table of Contents. Here we provide an overall
preview of the various chapters and how they interlink.
We begin in Chap. 2 with the dynamics of structures, where the important
notion of modes and mode shapes is described in the language of function spaces,
specifically L2 -spaces. (It is interesting to note that the need for infinite-dimensional
spaces is underscored by a recent publication Dynamics of Very High Dimensional
Systems [24].) The primary reason for this level of abstraction is that a precise
mathematical statement of the structure dynamics in air flow is not possible
without it.
The first flexible wing modelthe Goland modelwas proposed by Goland
[78] in 1945, a uniform rectangular beam of zero thickness with two degrees of
freedom, plunge and pitch. Similar equations with higher degrees of freedom are
employed in applications to spacecraft structures [54, 61]. It is a fixed wing
model cantilevered to the fuselage with the wing tip (clamped/free end conditions).
This is our canonical model; (all later versions still anchored essentially on this)
including nonlinear models. A drawback is of course that the thickness or the wing
camber is neglected. However, extending it to a plate model turns out to be too
complex to handle and it is questionable whether the results justify the analytical
complexity. There are no existence theorems, neither where the models include
wing camber nor the wing section as a tear drop [18]. Another peculiarity is that
the cantilever beam model is used only for the structure dynamics corresponding
to the aerodynamic loads, and to calculate the latter one takes advantage of the
high aspect ratio of the wing which must needs take into account the flexibility
of the structure, long slender wing. Some authors even further qualify it by very
flexible wing [82]. The flexibility is only along the span; we have thus a beam
instead of a thin plate. Furthermore, this allows the enormous simplification that
we may neglect the dependence on the span axis in describing the aerodynamic field
1 Introduction 3
equation. This is referred to as the strip theory or the typical section theory. We
can also see this as a first approximation of the finite plane theory (cf. Chap. 3).
In other words the structure beam model is consistent with the aerodynamic
strip [6] theory.
The Hilbert space for structure state variables presented is standard and essential
use is made of the notion of elastic energy, including a characterization of the
structure deflections for which the energy can be defined. This defines the modes
and mode shapes as eigenvalues and eigenvectors respectively. In continuum theory
the number of modes is infinite. In practice we know we will only deal with the
first few modes, so that much of the mathematical theory of asymptotic modes
is irrelevant. In the approximated computer model, the number of modes is indeed
finite, but then we can always add one more! Indeed we construct a continuum
model in which we match the first few modes of interest. This is because of the
inherent damping of the structure that eventually increases as the mode number
increases. But truncating the model at the beginning one loses physical phenomena
that only the continuum model can capture.
Control theory is needed for our objective of stabilizing or enhancing the stability
of the structure and here the truncated numerical model is useless. We need the
continuum model to generate feedback control laws. One difficulty is in not knowing
the inherent damping. The few models we have are not faithful enough. Hence in
robust control we assume a model in which there is no damping (as in our models)
but devise a feedback control which is guaranteed not to decrease the inherent
damping.
Current control design for distributed structures uses colocated point controllers
where the sensor and actuator are colocated. An inherent limitation here is that the
damping will decrease (not necessarily monotonically) to zero as the mode number
increases. An alternative is to use the self-straining actuators [70] which can
ideally achieve super stability. So they are attractive to try in the aeroelastic case.
Finally we consider nonlinear structure models (they may be considered exten-
sions of the Goland model) still beam models with zero thickness. The models
suggested [75, 76] unfortunately have not been analyzed in the continuum version
unlike the linear case. There is no longer the notion of modes (eigenvalues) which
makes the determination of stability difficult.
To make a mathematical statement of the structure dynamics we first need to
invoke aerodynamics to calculate the aerodynamic loads. Chapter 3 thus begins
with the aerodynamics part of aeroelasticity, the theory leading to the calculation
of the aerodynamic lift and moment. We limit consideration to nonviscous flow
where the viscosity of the air is neglected. Thermodynamics is involved and the
assumption thruout is that the perfect gas law holds. In the absence of viscosity we
need to assume that the entropy is a constant so that the heat exchange processes
are adiabatic. The assumption of constant entropy enables us to relate the pressure
to density:
p D A
4 1 Introduction
and show that the flow is vortex-free and we have potential flow which can be
then characterized by the Euler full potential equation. However, what distinguishes
the aeroelastic problem from being simply fluid dynamics is the fluidstructure
interaction, the boundary conditions.
The assumption here is flow tangency, the wing velocity normal to the plane is
equal to the fluid velocity normal to the wing plane. In addition we have to impose
the conditions discovered by Joukowsky and the condition named after Kutta which
is necessary for uniqueness of solution.
We are then able to come up with a complete statement of the aeroelastic
equations. Here we also specialize to 2D flow, and zero angle of attack in which
we ignore the dependence on the wing span (abandon the Finite Plane model) and
consider the typical section approximation justified by the highaspect ratio of
the winga raison detrealso for the flexible PDE model. Finally here we also
extend the formulation to include nonlinear structure models.
The study of stability of the wing structure (which for us now depends on the
assumed far-field speed that we consider a parameter) requires that we specify
stability about the state of the system. One usually thinks of the equilibrium state
to which the system is attracted or returns in the absence of any disturbance.
We need to determine the solution to the aeroelastic equation where we set all time
derivatives to zero. We call this the static solution. This is the objective in Chap. 4.
The technique for solution is to exploit the analyticity in terms of the structure state
variables so that we can make a Taylor series expansion about the zero or rest
structure state. We show that the solution is nonunique for a sequence of values of
the far-field speed parameter for which we have two different solutions, even if we
may call one of them trivial.
The main tool is the static version of the Possio integral equation which turns
out to be the finite Hilbert transform solved by Tricomi, who showed the need for
the Kutta condition for uniqueness of solution; see [11, and therein to the work
of Sohngen, a German pioneer in the theory]. See also [31]. Here we are able to
show the convergence of the power series pointwise. There are no discontinuities
except on the boundary, for z D 0. Here we also derive a simple formula for the
Transonic Dip for nonzero angle of attack which in numerical computation has to
be extrapolated from a point on the graph. See [65] as well as the references therein.
The nonzero static solution can be considered as an eigen value problemalbeit
nonlinear. The nonlinearity is primarily in the pitch or torsion variable.
The flow has no discontinuities in the field and the convergence is pointwisein
sharp contrast to the time varying case in Chap. 6.
We conclude the chapter with an example which turns out to be relevant to
viscous flow treated in Chap. 7.
One of the major results of the continuum theory is the characterization
of the smallest speed at which the wing is unstablethe Flutter speed
can be characterized as a Hopf bifurcation pointan illustration of the
Hopf bifurcation theory the successor to the classical stability theory of
Liapunov [42]. Here the stability is determined completely by the linear system
obtained by linearizing the nonlinear system about the steady state. Naturally the
1 Introduction 5
bulk of the aeroelastic theory deals with the Linear case, and there is no exception
here. By the linear case we mean the solution to the equations linearized about
a steady state. The term linear model is often used in the aeroelastic literature
without specifying the steady state about which it is linearized. Of course the
natural steady state would be that of constant flow and zero structure state. The
corresponding linear theory is treated in Chap. 5 the longest chapter in the book,
with eight sections. This provides the bread-and-butter of aeroelasticity and the
reason for its life in the sun. Here we study linear flutter analysis with continuum
models without approximation of any kind. (No Pade approximation or the Rodden
computer algorithm [17, 36]). The main concept is that of aeroelastic modes and
the linear structure dynamics is formulated as a convolution/evolution (semigroup)
equation in a Hilbert space. What is also new here is the development of a state space
theory which in particular shows that the aeroelastic modes are the eigenvalues of
the statespace stability operator. We also show that the case M D 0 is a special
one for which we have second degree noncirculatory terms not present for M > 0.
Also we are able to characterize the mode shapes explicitly and their properties
as eigenfunctions of the stability operator. The central theme is the linear Possio
integral equation [56, 89] which connects the Lagrangian dynamics to the Eulerian.
More specifically it provides the inputoutput relation between the structure velocity
as the input and the pressure jump across the wing as the output, so that we can
express the aerodynamic force and moment in terms of the structure state variables.
The key concept is the Mikhlin Multiplier and the Balakrishnan formula based
on it. The original version of the integral by Possio [89] was in the frequency
domain and is transported here to the Laplace domain. This suffices for calculating
the flutter mode and frequency. The aeroelastic modes are shown to be the zeros
of the determinant of a 3 by 3 matrix function of a complex variable which is
analytic except for the logarithmic singularity along the negative real axis for all
values of M . The aeroelastic modes are shown to be in a bounded vertical strip.
Zero is an aeroelastic mode for all M < 1. The associated flutter speed is known
as the divergence speed and is shown to be finite. The latter result is essential in
showing the existence of the flutter speed. It is discontinuous in M as a function
of the angle of attack and an explicit formula exhibits a transonic dip for a high
enough value of M , hitherto deduced from graphical computation. The modes are
characterized as the eigenvalues of the stability operator in the state space theory.
The uniqueness of solution of the Possio equation requires the Kutta condition
and in fact uniqueness implies existence. Various expressions are deduced for the
solution.
Key to the state space representation is the characterization of the aeroelastic
structure equation as a linear convolution evolution equation in a Hilbert space. The
convolution part plays the essential role in the flutter phenomena. We also derive an
explicit time domain solution for
M D0 and M D 1:
6 1 Introduction
We note that all the results obtained in the classical treatise [6] based on the
Theodorsen approximation can be deduced from the solution to the Possio equation
as in fact we do here. The sonic case .M D 1/ is interesting in that the role of the
Theodorsen function is taken over by the error function.
Essential for the stability theory is the fact that the slope of the stability curve is
negative for small speeds for all modes. The system becomes more stable initially
with airflow and then becomes unstable at some point as we increase speed, which is
the flutter speed. A relatively simple algorithm for it based on the continuum model
is given in the appendix to the chapter. One of the interesting by-products of the
continuum theory is the existence of evanescent states which decay too fast to be
captured in the CFD computer calculations.
All of this is only a prelude to the stability theory for the nonlinear system
developed in Chap. 6. The main result here is that the system is stable for all
values of speed less than the flutter speed and at the flutter speed the asymptotic
instability can be characterized as an LCO, limit cycle oscillation. It is consistent
with the Hopf bifurcation theory but the details of the proof have had to be developed
independently. The analysis turns out to be tedious and complicated involving many
asymptotic estimates. We show that the limiting response is periodic with the period
equal to the inverse of the linear flutter frequency with the preponderant harmonic
being the third harmonic.
We also derive an expression for the amplitude of the LCO rather involved
in terms of the various parameters. It is valid strictly speaking only for small
initial amplitude and is shown to be proportional to it. Although probably of less
importance in practice, the question of what happens for large initial amplitudes
is left open here as in computation. Heavy use is made throughout of Volterra
expansion. In fact in showing that the composite aeroelastic equation is a nonlinear
convolution/evolution equation, the nonlinearity is expressed in terms of a Volterra
series. And we note that in it the torsion variable is more dominant. We go with
the linear structure model, the extension to the nonlinear case being straightforward
based on the prequel. It is possible to examine the role of the structure in contrast to
the aerodynamics although we dont go into it.
As noted, our primary interest is in the structure but we also obtain en route a
revealing decomposition of the airflowthe velocity potentialinto two parts. One
part produces all the lift and can be linearized about the steady state whereas the
second part is continuous across the wing and cannot be so linearized. Interpreted
in the weak sense (which we do not attempt here), it may contain shocks, which is
controversial.
In Chap. 7 we venture even more into controversial territory where we allow the
flow to be viscous.
We assume small viscosity consistent with the fluid now being air. And we now
have to refresh the conservation laws. The basic field equation now becomes the
NavierStokes equation for incompressible flow. This is a much trodden area in fluid
dynamics and yet still with many unresolved questions, which are then reflected
in the structure dynamics as well. For example, there is no proof yet whether we
return to the Euler equation as the viscosity becomes vanishingly small. The big
1 Introduction 7
question here concerns the presence of shocks. There is no mathematical proof

yet of shocks in 2D flow. The boundary conditions now require that the structure
velocity be completely equal to the fluid velocity. To make any progress at all we
have to invoke the still unproven Prandtl boundary layer hypothesis, and even then
our treatment is not as complete as we would like.
The question of whether we can altogether suppress flutter by means of control
actuators (flutter suppression as it is called) is the theme of the eighth chapter. We
do have a model, a state space model, that makes it possible for us to consider
control design. Stabilizability is a central question in optimal control theory but
the application to state spaces of nonfinite dimension essential to exhibit the flutter
phenomena as here poses greater complexity. The main result here is that no actuator
on the structure can stabilize the system; we simply cannot suppress flutter. It will
always occur at some speed. However, we can design controls that enhance the
stability in the sense that we can increase the speed at which flutter occurs. But this
is no longer a crisp optimization problem and comparison of control performance
is probably not possible. One fallout of our theory is that the control need not be
nonlinear. Indeed as we have shown, the flutter speed is completely determined by
the linearized system.
One of the important safety issues for aircraft in flight is the effect of wind gust
which still continues to be of interest and is a familiar topic in flight dynamics [9,33].
Here we consider the problem with the 3D random field Kolmogorov model of air
turbulence [35] as opposed to the deterministic gust models studied in [6, 84]. We
calculate the spectral density of the structure response showing the role of the flutter
speed. Illustrative examples are given for both bending flutter and torsion flutter.
Finally in Chap. 10 we provide an addendum on the latest area of research in
flutter systems, nonaircraft and nonflight applications, for example to Piezoelectric
power generation [106,109] and the biomedical problem of palatal flutter [102,105].
Here the main difference is in the nature of the air flow. It is no longer normal to the
structure but is axial, axial flow. Our treatment is again based on continuum models
as opposed to the regulation computational as in extant literature [50, 102]. This
brings profound differences; the theory is more complicated and so this chapter is in
the nature of an Addendum confined largely to problem formulation, emphasizing
the difference from normal flow. It turns out that there is one simplification. In the
application we need only consider incompressible flow which is then characterized
simply by the linear Laplace equation. The nonlinearity is thus only on the boundary
dynamics. We consider here again the Goland beam model, and in particular the
symmetric case where the cg is on the elastic axis, and as a result the plunge and
pitch dynamics decouple. However, the mathematics is more complex in that the
Hilbert space has to be generalized to a Banach space: from an L2 space to Lp ,
1 < p < 2. We are limited by space alone to problem formulation with details
deferred to a forthcoming second volume.
8 1 Introduction
Notes and Comments
It is interesting to note that in the beginning in aeroelasticity theory only

continuum models were used by the German pioneers: Kushner, Schwarz,
Sohngen, and Wagner among others. The convolution aspect was already shown by
Wagner. But the Allies after WWII decreed a moratorium on any further research
by Germany from 1945 until 1954. This was a death blow to German work in
aeroelasticity from which it never really recovered. The advent of FEM/CFD can be
traced to the late 1970s. A landmark is the appearance in 1982 of Transonic Shock
and Multidimensional Flows: Advances in Scientific Computing edited by Meyer.
It is ironic that this volume contains an article by G Moretti pleading for Closer
Cooperation between Theoretical and Numerical Analysis in Gas Dynamics. The
Aeroelasticians combined the discrete approximation in structures FEM with CFD
for the aerodynamics. But the integration of the two via the boundary conditions was
always an art, often the more mysterious part. The battle between continuum theory
and computation is still enjoined. And as late as 2010 we see one unresolved point:
see [98] where the author writes: verification of numerical solutions is a step in
their scientific acceptability and a formal requirement of many engineering related
enquiries. . . However, these equations are known to have nonunique solutions. If the
true answer is not unique, what does correctness of approximation mean? As we
show in Chap. 7 that in viscous flow even existence of a solution is not proved
compounding this difficulty, because now we can ask: approximation to what? So
the conclusions from the computation have to be taken on faith.
It should also be noted that time is also discretized in the approximation but
there are phenomena, for example, in stochastic process theory, Gaussianness of the
innovation process in nonLinear filtering, which holds only in the continuous time
model (Girsanovs theorem: see, for example, [26]).
Chapter 2
Dynamics of Wing Structures
2.1 Introduction
This chapter deals with the dynamics of elastic structures: unswept fixed-wing and
free-free wings, modeled as beams of zero thickness, linear as well as nonlinear.
The main concern is with spectral analysis, modes and mode shapes as a means to
study stability. The language is that of abstract functional analysis: Hilbert spaces
and semigroup theory of operators for time domain description.
2.2 The Goland Beam Model
This is the earliest continuum model introduced in aeroelasticity by Goland [78] in

1945. See Fig. 2.1. It is a uniform rectangular beam. The co-ordinate axes are chosen
so that X -axis is the (rigid body) airplane axis also called the Chord-axis with the
width or chord length 2b:
b < x < b:
The one-sided wing span is ` along the Y -axis:
0<y<`
and it is assumed that the aspect ratio
`=b
is high, justifying the flexible beam model. It is endowed with two degrees of
freedom: bending displacement h./ in a plane normal to the beam (plunge in
aeroelastic parlance) and the torsion angle ./ in radians (pitch in aeroelastic

10 2 Dynamics of Wing Structures
Fig. 2.1 Wing model x

TRAILING EDGE
U b
CG bx PITCH AXIS
ba
0
SPAN
b LEADING EDGE

SPAN = ; HALFCHORD = b
ANGLE OF ATTACK =
THICKNESS =
BEAM MODEL: = 0
parlance) about an axis parallel to the wing axis at distance bx from the cg as
shown in Fig. 2.1. The bending dynamics is described by an Euler equation:
R y/ C S .t;
mh.t; R y/ C EI h0000 .t; y/ D L.t; y/; t > 0I 0<y<1<1
(2.1)
and the torsion by the (Saint Venant, string, or Timoshenko) equation:
R y/ C S h.t;
I .t; R y/ GJ 00 .t; y/ D M.t; y/; t > 0; 0 < y < `; (2.2)
where the forcing terms:

L.t; y/ is the aerodynamic force per unit length (lift);
M.t; y/ is the applied aerodynamic.
The moment about the elastic axis per unit length, are both determined later from
the airflow model.
m D Mass per unit length
I D Crosssectional moment of inertia
EI D Bending stiffness
GJ D Torsional stiffness
S D Coupling constant D mbx , jx j < 1,
see Fig. 2.1 for x ; all constant, justifying the term uniform.
See [5] for more details on the physical units.
Here x can be positive or negative, and it is required that S 2 < mI .
The convention throughout is that the superdots denote the partial derivatives
with respect to time t and the superprimes the partial derivatives with respect to the
spatial co-ordinate y. Note that the structure variables do not depend on the chord
variable. To complete the dynamics we need to specify the end conditions.
2.3 Time Domain Analysis 11
End Conditions
1. For fixed-wing aircraft where the wing is attached to the fuselage (as in
Fig. 2.1) we add the Clamped-Free or CF end conditions:
h.t; 0/ D 0I h0 .t; 0/ D 0I h00 .t; `/ D 0I h000 .t; `/ D 0I

.t; 0/ D 0I 0 .t; `/ D 0 t > 0: (2.3)
2. FF Free-free condition
This is typical of the recent UAV aircraft that have no fuselage and resemble a
flying wing. Here both ends are free.
h00 .t; 0/ D 0I h000 .t; 0/ D 0I h00 .t; `/ D 0I h000 .t; `/ D 0;

0 .t; 0/ D 0I 0 .t; `/ D 0 t > 0: (2.4)
3. CC Clamped-clamped condition
This is relevant to non-aircraft applicationsee Chap. 10
h.t; 0/ D 0I h0 .t; 0/ D 0I .t; 0/ D 0;

h.t; `/ D 0I h0 .t; `/ D 0I .t; `/ D 0:
2.3 Time Domain Analysis
Our main concern is the stability of the structure. Stability of motion is a time
domain concept even if all the techniques involve Laplace transformation, Laplace
domain.
For an elementary treatment of vibration of beams reference may be made to
the classic treatise of Timoshenko [32] and the textbook [34], and the more recent
treatment relevant to aeroelasticity by Hodges and Pierce [5].
Here of course we use abstract functional analysis, albeit rudimentary, following
[61] where a model with three degrees of freedom with damping is analyzed and
[56] where specifically the Goland model is treated. Both of these follow the general
Hilbert Space formulation in [16, 55]. Not only does it make the presentation
compact and elegant, this level of abstraction is actually necessary for a precise
mathematical formulation of the structure dynamics in airflow, the subject matter of
this book.
Let H denote the Hilbert space:
L2 0; ` L2 0; `
with elements

h./
xD :
./
This choice of space is motivated by the fact that the elastic potential energy is
given by
Z Z !
` `
1 00 0
EI jh .y/j dy C GJ
2
j .y/j dy :
2
(2.5)
2 0 0
We begin with the abstract differential operator denoted A.

The Domain of A (denoted D(A)): where we need to distinguish between CF
clampedfree, FF freefree and CC clampedclamped end conditions:
For CF
h
x j h0000 ./ 2 L2 0; Ì 00 ./ 2 L2 0; `; h .0/ D 0I h0 .0/ D 0I
i
.0/ D 0 h000 .`/ D 0I h00 .`/ D 0I 0 .`/ D 0 :
For FF
h
x j h0000 ./ 2 L2 0; Ì 00 ./ 2 L2 0; `; h000 .0/ D 0I h00 .0/ D 0I
i
0 .0/ D 0 h000 .`/ D 0I h00 .`/ D 0I 0 .`/ D 0 :
For CC
h
x j h0000 ./ 2 L2 0; Ì 00 ./ 2 L2 0; `; h .0/ D 0I h0 .0/ D 0I
i
.0/ D 0 h .`/ D 0I h0 .`/ D 0I .`/ D 0 :
Although strictly speaking we need to distinguish between these operators in

principle, it is less messy to go ahead with one operator in the abstract theory, but
making the necessary changes in the concrete calculations.

EI h0000
Ax D : (2.6)
GJ 00
Thus defined it is standard [16] to verify that the domain is dense in H and that A
is closed on its domain. Note further that for x in D(A):
Ax; y D Ay; x for y in D.A/:
Or, A is self-adjoint. Furthermore we can calculate that

Ax; x D 2 potential energy 0.
Or, A is nonnegative definite.

Suppose for some x in the domain of A, we have
Ax D 0:
Then
Ax; x D 0:
Or the potential energy is zero, and hence
x D 0:
Or, zero is not in the spectrum of A. Any such that
Ax D x; x nonzero
is called an eigenvalue of A and x is the eigenvector corresponding to .

The set of eigenvalues is countable; the eigenvalues fk g of A are positive and
the corresponding eigenfunction spaces are each of dimension 1. The eigenfunctions
fk g are orthogonal, complete, span the space, and are orthonormalized for our
purposes here. The resolvent, denoted R.; A/, is compact, because the beam is
finite. Any complex number not equal to k for any k is in the resolvent set. Also
the resolvent is HilbertSchmidt [16]
1
X 1
X 1
R.; A/ k D2
<1
j k j2
kD1 kD1
and we have the eigenfunction expansion:

1
X x; k
R.; A/x D k : (2.7)
k
kD1
Moreover the eigenfunctions are complete in H ; see [16]. Any element in H has
the modal representation
1
X 1
X
xD x; k k ; jx; k j2 < 1:
kD1 kD1
The square root operator

The potential energy can be defined for a largerpclass of elements in H than
D.A/. For this we need the square root of A denoted A. It can be defined in terms
of the eigenfunctions as
1
X
p
x2D A if k jx; k j2 < 1
kD1
and we define
1 p
X
p
Ax D k x; k k (2.8)
kD1
p
p that zero is not an eigenvalue of A.
from which it follows
The domain of A is thus larger than that of A. For x in D.A/
hp p i
Ax; Ax D Ax; x D 2 Potential energy :
p
We then define potential energy for all elements in D A as:
1 p
jj Axjj2 :
2
This is the largest domain for which p
the potential energy can be defined. We rarely
need to know the precise domain of A (see [55] for details on the domain) or the
precise form; we should note, however, that it may not be a differential operator. See
[63] and below.
The time domain solution: S D 0.
We begin by rewriting the dynamic equation (2.1) and (2.2) in abstract form as

L.t/
MxR C Ax D ; (2.9)
M.t/
where M is the nonsingular nonnegative definite 2 2 matrix

ms
s I
and is the mass/inertia operator; the operator A can be called the stiffness operator,
extending the finite-dimensional definitions.
The Energy Space
Let us introduce next the energy space H consisting of elements of the form:
p
xl
Y D x1 2 D A ; x2 2 H
x2
endowed with the energy inner product:
hp p i
Y; ZE D Ax1 ; Az1 C Mx2 ; z2 ; (2.10)
where

Zl p
ZD Z1 2 D A ; z2 2 H:
Z2
Denote this inner-product space by H.

Theorem 2.1. H is a Hilbert space.
Proof. We need to show that every Cauchy sequence converges to an element in the
space.
Let

xn
Yn D
yn
be a Cauchy sequence in H. Then

p
Axn
p
is a Cauchy sequence in H . Because 0 is not an eigenvalue of A,
p p
R 0; A Axn D xn ;
p p
where R ; A denotes the resolvent of A, converges to an element x in H ,
p p
actually in D A , because A is closed.
hp p i
My; y D M y; M y ;
thus it is seen that yn is a Cauchy sequence in H , by a similar argument. Hence it

follows that Yn converges to an element in H. Hence H is a Hilbert space. t
u
p
p Note that what we have proved is that the domain of A is closed in H because
A has a bounded inverse. This is the largest domain for which the potential energy
can be defined. p
We rarely need to know the precise domain of A (see [63] for details on
the domain) or the precise form; we should note, however, that it may not be a
differential operator. In our case we can see that
p
p h EI h00
A D :
0 0
But the square root for the torsion operator is no longer a differential operator.
For the clampedclamped case, for example, it is given by
p Z
1 ` Sin s 0
A D gI g.s/ D `
./d
` 0 Cos s
`
Cos `
0<s<` .0/ D .`/ D 0:
See [63] for more on this.

Note that the norm in H has a physical significance. jjY jj2 D 2 total energy D 2
(potential energy C kinetic energy).
Hence we refer to it as energy space.
As in finite dimensions, we go on to the state space formulation of the problem.
Let
x.t/
Y .t/ D :
P
x.t/
Then the second-order equation (2.5) goes over into the first-order in time equation:
YP .t/ D AY .t/ C V.t/; (2.11)
where

vl .t/
V.t/ D ;
v2 .t/
where

0
v1 .t/ D ;
0

L .t/
v2 .t/ D ;
M .t/
where L.t/ denotes the element L.t; y/ and M.t/ the element M.t; y/ in H, and
the operator A is defined by:

yl 0 I yl y2
A D 1 D
y2 M A 0 y2 M1 Ay 1
with domain:
y1 2 D.A/;
p
y2 2 D A :
Thus defined it is closed with domain dense in H.

Moreover we can readily verify that the adjoint (in H) A is given by A:
A C A D 0
and hence [16] A generates a C0 semigroup S.t/; t 0. This is not true
incidentally if the space is H H . See [39]. The solution of the homogeneous
equation
YP .t/ D AY .t/
is given by
Y .t/ D S.t/ Y .0/ for Y .0/ in D.A/:
Lemma 2.2. The semigroup S./ is isometric:
jjY .t/jj D jjS.t/Y jj D jjY jj: (2.12)
The energy remains a constant, in other words.

Proof. For Y in H,
m.t/ D S.t/Y; S.t/Y E D jjY .t/jj2 :
Then for Y in D.A/
P
m.t/ D .A C A /S.t/Y; B.t/Y D 0
and hence for Y in D.A/

jjS.t/Y jj D jjY jj:
But the domain of A is dense in H, and hence (2.8) follows. t
u
An equivalent statement is that
S.t/1 D S.t/ :
It should be noted that the differential equation:
YP .t/ D AY .t/
is satisfied only for Y .0/ in the domain of A.
The nonhomogeneous equation we started with:
YP .t/ D AY .t/ C V.t/
has the solution

Z t
Y .t/ D S.t/Y .0/ C S.t /./d t >0 (2.13)
0
for Y .0/ in the domain of A, and may need to be interpreted in the weak sense
depending on the function ./ (as a function of time) [3] but we do not need to go
into this until later.
2.4 Structure Modes and Mode Shapes
Our main concern is the stability of the solution or of the semigroup S./. This
means we need to study the spectrum of A. Because R.; A/ the resolvent of A is
compact, we need only consider the eigenvalues of A.
Unraveling

y1
.I A/Y D 0; Y D ;
y2
we have:
y1 y2 D 0;
y2 C M1 Ay1 D 0:
Hence
y2 D y1 ;
2 My1 C Ay1 D 0I Ay1 D 2 My1 ;
Ay1 ; y1 D 2 My1 ; y1 :
Hence it follows that 2 0I and hence 2 D ! 2 ; ! real. But cannot be zero,

because if it is, so is y1 ; y2 .
The eigenvalues then are defined by
y2 D i!y1 ;
Ay1 D ! 2 My1 ;
y1 is an eigenvector of A but with respect to M rather than the identity. Again the
eigenvectors are countable and complete in the Hilbert space H with an equivalent
inner product:
x; yM D x; My:
See [2].
Hence the eigenvalues are purely imaginary:
k D i!k ; !k > 0:
2.4 Structure Modes and Mode Shapes 19
The !k are thus defined as the modes of the structure (in radians) and because
.1=!k / goes to zero, can be arranged in increasing order of magnitude so that we can
talk about the kth mode without ambiguity. We may note here that the asymptotic
behavior of the modes is of little interest to us in practice, where only the first
few modes play a role. Let k denote an eigenvector of A corresponding to the
eigenvalue k .
Note that k has the form

k
k D :
i !k k
In our particular case we can calculate the modes !k and the mode shapes k by
solving the differential equations
2 mh C EI h0000 D 0; (2.14)
2 I GJ 00 D 0: (2.15)
Note that this is no more than taking Laplace transforms of the time domain
equations, familiar in engineering.
Thus k is of the form:

hk 0
or
0 k
and we can distinguish between the bending modes and the torsion modes.
We have corresponding to the bending motion
!k2 mhk .y/ D EI h0000

k .y/ (2.16)
plus the CF end conditions:
h.0/ D 0 D h0 .0/;
h00 .`/ D 0 D h000 .`/:
The !k then are pure bending modes.

As may be expected, this is a classical result already found in Timoshenko, 1928
[32] and in textbooks [34]. The modes !k are determined by the equations:
1 C Cosh k Cos k D 0 (2.17)
and
1=4
1 EI
!k D k 0 < k " :
` m
Mode shape:
C1 .sinh !k y sin !k y/ C C2 .sinh !k y C sin !k y/; 0 < y < `:
For more see [32, 34].

Torsion modes: These are the zeros of
cos h
!k ` D cos
!k ` D 0;
s
GJ
!k D .2k 1/ ; k D 1; 2: (2.18)
2` I
Mode shape
.2k 1/
const sin y 0 < y < `: (2.19)
2`
An obvious comment here is that the modes decrease linearly in magnitude as the
span length increases. As a rule the structure also increases in flexibility so that the
density of modes also increases. There are more modes to be considered as length
increases.
We note here that for the clampedclamped case the modes are determined as the
roots of:
"p p # "p p #!
p 2 `m1=4 jj 2 `m1=4 jj
EI GJ 1 C cos cosh
EI 1=4 EI 1=4
p
` I
sinh p D0
GJ
yielding an interesting variation of (2.4).

The modes are undamped. The energy remains constant and does not increase or
decrease. The structure model is neutrally stable.
Modal Expansion: Greens Function
Theorem 2.3. Any element Y in H has the modal representation expansion:

1
!
X 1
Y D Y; k k C Y; k k ; (2.20)
kD1
2!k2
where

k
k D I Ak D !k2 k I Ak D i !k k I
i !k k
A k D i !k k ; !k > 0; (2.21)
X1
1
2
< 1I k ; k D !k2 I k ; k D 0 (2.22)
kD1
!k

k
Mk ; j D jk
D k
i !k k
ffk g; fk gg are orthogonal eigenvectors of both A and A .
Then
1
!
X 1 i!t
S.t/Y D Y; k e k C Y; k e
i!t
k ; (2.23)
kD1
2!k2
1
!
X 1 k k
R.; A/Y D Y; k C Y; k : (2.24)
2!k2 i !k C i !k
kD1
Proof. Follows [16, 54]. The modal expansion uses completeness of the fk g.
Otherwise, because, the eigenfunctions are orthogonal, the proof of the expansions
is straightforward. t
u
Greens Function Representation
A useful alternate representation of the semigroup and the resolvent is through the
Greens function; see [44, vol. 1]. Here we develop it using the modal representation.
Thus S.t/Y is the function
1
X1
i!k t
Y; k e i!k t
k .s/ C Y; k e k .s/ D Y; G .t; s; :/; (2.25)
kD1
2!k2
where
1 i!k t
1
X
i!t
G.t; s; / D e k ./k .s/ C e k ./ k .s/ (2.26)
kD1
.2!k2 /
is then the Greens function for S.t/. Similarly the resolvent is simply the Laplace
transform which is analytic in excepting the imaginary axis.
O
R.; A/Y D Y; G.; s; :/ (2.27)
X1
O 1 1 1
G.; s; / D k ./k .s/ C k ./k .s/ ; i !k
2!k2 i !k C i !k
kD1
(2.28)
and now the function is square integrable:

Z ` Z `
O
jG.; s; /j2 dsd < 1: (2.29)
0 0
General Case: Nonzero Coupling

Next we consider the general case allowing for nonzero S . The main question is
how much the coupling changes the modes. We do expect the mode shapes to be
coupled. We limit consideration to the CF case here.
The equations corresponding to (2.1, 2.2) are:
2 mh.y/ C 2 S.y/ C EI h0000 .y/ D 0; (2.30)
2 S h.y/ C 2 I .y/ GJ 00 .y/ D 0 (2.31)
with the same CF end conditions as before.

To solve this set of equations we proceed to consider the state-space version.
Here we follow [56].
Thus let
Y .s/ D Col.h.s/; h0 .s/; h00 .s/; h000 .s/; .s/; 0 .s/:
Then we have
Y 0 .s/ D A./Y .s/ 0 < s < 1;

Y .0/ D Col 0; 0; h00 .0/; h000 .0/; 0; 0 .0/
and A () is the 6 6 matrix

0 1
0 100 0 0
B0 0C
B 010 0 C
B C
B0 001 0 0C
A./ D B C (2.32)
Bw1 000 w2 0C
B C
@0 000 0 1A
w3 000 w4 0
m
w1 D 2 ;
EI
S
w2 D 2 ;
EI
S
w3 D 2 ;
GJ
I
w4 D 2 ;
GJ
where w2 and w3 are the coupling terms.
The eigenvalues in the CF case are the zeros of dc ./ D Det Dc ./.
We define the 3 3 matrix:
Dc ./ D P e À./ Qc ;
0 1
00 0
0 1 B0 0 0C
B C
001000 B C
B1 0 0C
P D @0 0 0 1 0 0A Qc D B C:
B0 1 0C
000001 B C
@0 0 0A
00 1
The eigenvalues in the FF case are the zeros of
df ./ D DetDf ./;

Df ./ D P e À./ Qf ;
where
0 1
10 0
B0 1 0C
B C
B C
B0 0 0C
Qf D B C
B0 0 0C
B C
@0 0 1A
00 0
The eigenvalues in the CC case are the zeros of
dc ./ D Det Pc eA./` Qc
where
0 1
100000
Pc D @0 1 0 0 0 0A :
000010
Thus defined d./ (generic for both dc and df ) is an entire function of finite order
(see [10]).
For S D 0,
Dc ./DP e A./ Qc
0 1
1 sinw11=4 ` C sinhw11=4 `
B .cosw1 1=4
` C coshw1 1=4
`/ C
B 2 2w11=4 C
D B 1 1=4 C
@ w1 .sinw11=4 ` C sinhw11=4 `/ 12 .cosw11=4 ` C coshw11=4 `/ A
2
0 0
this yields:
1
dc ./ D .1 C cos `/cosh `cosh
`; (2.33)
2
where
1=4 1=2
D w1 I
D w4 ;
which agrees with our previous calculation (2.4), (2.5).

To make explicit our interest in the dependence on S, let us modify the
notation to:
A.; S / in place of A.)
d.; S / for d./
so that d.; 0/ is given by d./.
The function is clearly analytic in S.
Let
Ap .; S / D .A.; S / A.; 0//=S:
Note that Ap .; S / is given by

0 1
0 000 0 0
B C
B 0 000 0 0C
B C
B 0 0C
B 0 000 C
B C
B 2 C :
B 0 000 0C
B C
B EI C
B 0 000 0 0C
B C
@ 2 A
000 0 0
GJ
Now
d tA.;S /
e D A.; S /etA.;S /
dt
D A.; 0/etA.;S / C SAp .; 0/etA.;S / :
The familiar method of variation of parameters formula [16] yields:
Z t
etA.;S / D etA.;0/ C S e.t /A.;0/ Ap .; 0/eA.;S /d; t > 0:
0
We can treat this as a matrix Volterra equation for etA.;S / and the solution has the
expansion:
1
X
e1A.;S / D e1A.;0/ C S n Fn .1/;
nD1
where
Z 1
F1 .1/ D eA.;0/.1 / Ap .; 0/eA.;0/ d:
0
For terms of higher order, see [70]. As shown there, d.; S / is a function of S 2 and
using the first term in the series yields
d.; S / D d.; 0/ C S 2 d2 .; 0/; (2.34)
where
1
d.; 0/ D .1 C cos ` cosh `/ cosh
`
2
1 p
d2 .; 0/ D . 2
2 .4. 2
2 / coshl
sinl
2
p p
.. 2 C
2 / sinhl
sinhl
/
p p
C coshl .. 2 C
2 /2 C . 2
2 /2 cos2 l
p
C 4 2
2 .2 cosl coshl
cosh2 l
/
p p p
4
. 2
2 / sinl sinhl
/ C 4
p p
coshl
.2
C . 2 C
2 / sinhl sinhl
/
p p
C cosl .. 2
2 /2 . 2 C
2 /2 cosh2 l
p p
4 2
2 cosh2 l
C 4
. 2 C
2 / sinhl
sinhl
///=.8m. 4 2
4 /2 I /: (2.35)
Let k denote a zero corresponding to S D 0. By applying one step of the Newton

root finding algorithm, we can see that for small S it can be approximated by
k C k ; (2.36)
where
k D S 2 d2 .k ; 0/=d 0 .k; 0/
and the main thing to note is that
Re.k C k / < 0I ! 0 as k ! 1: (2.37)
The modes are damped, and the mode shapes are now coupled. The bending mode-
shape vector now has a nonzero torsion component proportional to S 2 , and similarly
for the torsion-mode shape there is a bending component. But the extra components
being small, we continue to identify them as bending or torsion modes. Calculating
these would take us too far from our main interest.
2.5 Robust Feedback Control Theory: Stability

Enhancement
We have seen in Sect. 2.4 that the structure model with zero coupling is neutrally
stable. The system energy neither decreases or increases whatever the initial
condition. Experience shows that all modes decay to zero and the higher the
damping the higher the mode. However, the lower modes may not decay as fast
as we would like. Hence we need to instrument controls to increase the damping
without, however, destabilizing the structure and without increasing the damping in
any mode. Unfortunately, the inherent damping is difficult to model and we need
to design a controller without knowing the damping and yet without degrading the
inherent damping. Fortunately we can indeed design such a control, a triumph of
control theory of structures. In any real system we can only provide for a finite
number of control inputs. Hence we may assume that the controls denoted u./,
u./ L2 Rm ; 0; T for every 0 < T < 1:
Let B denote the control operator so that (2.5) now includes controls and we have
Mx.t/
R C Bu.t/ C Ax.t/ D v.t/:
2.5 Robust Feedback Control Theory: Stability Enhancement 27
To this we add a damping operator D even though we cannot specify it except to

note that D is linear bounded self-adjoint and nonnegative definite. Thus we have
finally:
MxR C D x.t/
P C Ax.t/ C Bu.t/ D 0: (2.38)
An example of a damping operator D is
p p p p p p
2 M T M; T D M1 A M1 ; jj < 1;
see [61], corresponding to proportional damping, damping proportional to the

velocity, going back to Timoshenko [32].
Here we simply assume it is compact and nonnegative definite, zero not in the
spectrum.
With H as before we go to the state-space form:
YP .t/ D Ad Y .t/ C Bu.t/;
where

0 I
Ad D ;
M1 A M1 D

0
BD ;
Ml B
.Ad C Ad /Y; Y E D Dx 2 ; x2 ;
where

xl
Y D :
x2
Hence Ad now generates a contraction semigroup Sd .t/t 0:
Sd Y; Y Y; Y
with compact resolvent. Our main result is to show that we can design a feedback
controller that is robust in that it is not required to know either D or A and is such
that the structure is strongly stable.
Controllability
For this purpose we introduce the important notion of controllabilitysee [16]

the original notion in finite dimensions extended to infinite dimensions. Let R
denote the infinitesimal generator of a C0 semigroup S.t/; t 0, over the Hilbert

space H. We say that the semigroup is strongly stable if for each Y in H,
jjS.t/Y jj ! 0 as t ! 1:
Let B denote a finite-dimensional operator on RmSinto H.

We say that the pair R B is controllable if t 0 S.t/Bu; uRm is dense in H.
Here we can state a basic result of control theory.
Theorem 2.4. Suppose the semigroup S./ is a contraction, the generator has a
compact resolvent, and .A B/ is controllable. Then the semigroup generated by
A BB
is strongly stable.
Proof. See [16]. t
u
Remark. It is known (see [16]) that a finite-dimensional control cannot guarantee
a uniform decay rate for all modes (exponential stability). The rate of decay will
eventually go to zero as the mode number increases indefinitely. Strong stability
means that the energy in any initial state will eventually decay to zero.
Let us apply this to our case, to the semigroup S./ with generator A and control
operator B.
Theorem 2.5. The pair A; B is controllable if and only if for any nonzero eigen-
vector of A,
B 0:
Proof. Suppose contrarywise B D 0 for some nonzero eigenvector with
eigenvalue i!.
Then
B ; u D ; B u D 0 for every u in Rm :
Now
; S.t/B u D S.t/ ; B u D eit ! ; B u D 0:
Hence the set [
S.t/Bu; u Rm is not dense in H;
t 0
which is a contradiction. t
u
And the argument can be retraced for the only if part readily.
We assume now that A, B is controllable. The resolvent of A is compact and B
is finite-dimensional and hence the semigroup generated by
A BB
is strongly stable.
Let us see what this implies. By modal stability we mean that each mode decays
to zero, actually exponentially, with the rate determined by the real part of the
eigenvalue. In the case where the dimension of the space is not finite this does not
mean strong stability. It only implies that the system is damped but the rate of decay
depends on the element. Exponential or uniform stability is defined in terms of the
semigroup S./ by requiring that the stability index:

1
Inf LogjjS.t/jj D !o < 0;
t
which implies that for C0 semigroups that
logjjS.t/jj
! !o as t ! 1:
t
And given any > 0, we can find M such that
jjS.t/jj Me.! C/t
and hence the name exponential stability.

In our present structure dynamics context we may interpret this as guaranteeing
a uniform exponential decay rate for all initial states. Here we have a negative result
[16] that this cannot be achieved by finite-dimensional controllers, hence not by
means of active controllers.
Let us now return to show that the feedback control
u.t/ D B Y .t/
does not destabilize the structure model whatever the damping operator D is, where
M x.t/ R C Ax.t/ C BB x.t/

R C D x.t/ P D0
so that we have a new damping operator
D C BB * ;
which is
D:
If is any mode of this system we see that the damping
.D C BB /; D; :
On the other hand, B being finite-dimensional

1
X
T rBB D jjB k jj2 < 1;
kD1
where k denote the modes, so that the active damping introduced goes to zero as
the mode number increases. It is usually much smaller than any natural damping.
Another point is that one cannot guarantee a specific value of damping for any mode.
Also, few reliable models of damping are known, so designing controls based on any
damping model can be hazardous. For a successful use of this control law see [103].
Self-Straining Actuators
We have seen that finite-dimensional controls cannot yield exponential stability, so

we would need an infinite-dimensional controldistributed control as opposed to
point controllers on the boundaryto achieve this, which is of course not physically
realizable.
A class of actuators using piezzo strips that are self-sensing and self-straining
is described in (see [79] and the references therein) with the potential to yield
exponential stability.
We can extend our theory to investigate this class. We begin with a beam model.
The displacement of the piezzo-electric strip charges a condenser which is then bled
as a current for actuation. The differentiator circuit is an integral component, and
we have really rate feedback. We begin with the torsion dynamics:
R s/ GJ 00 .t; s/ D 0
I .t; 0 < s < `;
.t; 1/ D 0;
GJ 0 .t; 0/ C g P .t; 0/ D 0;
where g is the actuator gain, and we need to study how the system stability depends
on the gain. We may consider this as a singular mass matrix version of the boundary-
control problem:
R `/ GJ 0 .t; 0/ C g .t;
m.t; P 0/ D 0;
where now m is zero.

The novelty here is the inclusion of the values at the ends as separate from the
functions.
Thus let H1 denote the Hilbert Space:
L2 .0; `/ R1 ;
with elements
f ./
xD :
b
Let A denote the operator with domain in H1 :
0 1
f ./jf 00 ./ L2 .0; `/;
D.A/ D @ f .`/ D 0 A
f .0/
and

GJ f 00 ./
Af D :
GJ f 0 .0/
Then A is self-adjoint and nonnegative definite with

Z `
Af ; f D GJ jf 0 .s/j2 ds;
0
p
which is twice the elastic energy. Denoting by A the positive square root of A, it
can be seen following an argument similar to the one given in [11] that the domain:
" #
p f ./ 0
D. A/ D ; where f .`/ D 0 f ./ L2 .0; `/ ;
f .0/
p
but A is not a differential
p operator. We note that zero is not in the spectrum of A
and hence that of A . Next, as before we define the energy space:
p
HE D D. A/ L2 .0; `/;
which is a Hilbert space with inner product:

hp
x y p i
; D Ax; Ay C I f; g
f g E
and

x x
; D 2 (total energy: kinetic C potential):
f f E
We define next the operator A with domain in HE with domain that is a little
complicated:

y fl ./
D.A/ D Y D with y D.A/; D ;
f2 ./ fl .0/
! #
f2 ./ p
and GJ 0 D. A/ :
f .0/
g l
The condition on f2 ./ is equivalent to:
f2 .`/ D 0;
f20 ./ L2 .0; `/:
This domain is dense in HE . We now define A by:

0 1
f2 ./
B GJ 0 C
AY D B f .0/ C :
@ g 1 A
GJ 00
f
I 1
./
Then A is closed linear with dense domain and compact resolvent. Also
.GJ / 0
AY; Y E D GJ f2 ./; f100 ./ jf1 .0/j2 C GJ f100 ./; f2 ./:
g
Hence
jf10 .0/j2
Re AY; Y D .GJ /2 0:
g
This is enough to prove that A generates a C0 semigroup S.t/; t 0, a contraction
semigroup, but actually exponentially stable, as we show presently. Of greater
interest to us are the eigenvalues and how they depend on the gain g.
Spectrum of A
Let A D which with
0 1
f1 ./
D @ f1 .0/ A
f2 ./
yields
f1 .s/ D c sinh ..` s//; 0 < s < `;
where
I
2 D ;
GJ
where is the root of
GJ I cosh ` C g sinh ` D 0:
Now the function on the left is an entire function of order one and of completely
regular growth [6] with a sequence of zeros given by
p
GJ I
tanh .`/ C D 0:
g
Or
p
g C GJI
e 2`
D p ;
g GJI
k D jk j i !k ;
where the real part is a negative constant given by

s
g C pGJI
1 GJ
jk j D log p ;
2` I g GJI
s
.2k C 1/ GJ p
!k D ; g < GJI
2` I
s
GJ p
Dk ; g > GJI ;
` I
k D nonnegative integers:
Thus all modes are damped at the same rate. And

p
gc D GJI
may be called the critical gain.

A plot of the relative damping constant ` versus the relative gain g=gc is
given in Fig. 2.2.

1 1Cx
log abs :
2 1x
As can be seen from Fig. 2.2, the limiting damping becomes infinite at the critical
gain.
Let us explore further what happens at the critical gain.
We have
d./ D cosh` C sinh` D e`
Fig. 2.2 Damping versus 2.5

(relative) gain
2.0
1.5
1.0
0.5
0.5 1.0 1.5 2.0
and has no zeros at all; no modes at all! The stability index
D 1;
the system is superstable. It is interesting that the implication of superstability

in this case is that the response vanishes in finite time T determined by gc . All
states are evanescent. This has been given the rather naive if picturesque name: a
disappearing solution, in another context, scattering theory. See [80].
But this is of course impossible in practice for structure response, for many
reasons; see [62] for details. Primarily it is the degradation in the operational
amplifier differentiator as the gain increases. The gain gc is not physically attainable.
See [79].
Although our primary interest is in the modal response, we complete the analysis
for any noncritical gain by showing that we have exponential stability, that any initial
state will decay at an exponential rate. The modes are given by
0 1
k ./
k D @ k .0/ A ;
k k ./
where
k .s/ D ak sinhk v.1 s/ 0<s<`
are not orthogonal. But we can create a biorthogonal system; see [1] for more.
Define
0 1
k ./
k D @ k .0/ A ;
k k ./
the superbar denoting complex conjugate.

2.6 Nonfixed Wing Models: Flying Wings 35
Thus defined it is readily verified that
k ; j D 0 for k j:
Hence if
X Y; k
Y D bk k ; then bk D
k ; k
and
X
S.t/Y D e t ei!k t bk k
X
D e t ei!k t bk k ;
which is enough to prove the exponential stability.

The corresponding theory for bendingEuler beams with self-straining
actuatorsis treated in [70], but there is no analagous superstability at the gain
corresponding to maximum damping. This would indicate that we can attain more
damping by torsion actuators, a fact corroborated by experiment.
Our interest is of course in the stabilization of the structure subject to aerody-
namic loading. We would expect that if we have a controller that does well in still
air, it should be a candidate for use in airflow as well. This is studied in Chap. 8.
2.6 Nonfixed Wing Models: Flying Wings
FreeFree Articulated Beam Model
We again consider the basic uniform Goland model, but it is no longer attached at
one end point to the fuselage; it is simply a flying wing. Both ends are free, FF.
Hence we use the matrix Qf in what follows. But it is articulated with discrete
masses placed along points (nodes) on the beam. This is typical of the recent
Helios UAV Flying Wing. See [81]. Such a model for a fixed-wing aircraft was
also considered earlier by Goland and Luke [77]. There is only a single wing
span, 0 < s < `. The masses mi , are at s D si ; i D 0; 1; : : : ; m C 1, with
s0 D 0; smC1 D 1. There are thus .m C 2/ masses maximum. We have only to
set the mass to be zero if there is none at si for any i ! See Fig. 2.3.
40 ft
60 lb
Fig. 2.3 Example of S1 S2 S3 S4 S5

articulated beam
Between nodes: si < s < si C1 , we have the Goland equations (cf. (2.1) and
(2.2)):
R s/ C S .t;
mh.t; R s/ C EI h0000 .t; s/ D L.t; s/; t > 0I si < s < si C1 ; (2.39)
I R .t; s/ C S h.t;
R s/ GJ 00 .t; s/ D M.t; s/; t > 0; si < s < si C1 : (2.40)
To allow for the discontinuities at the nodes we follow the technique in [2]. We
essentially incorporate the nodes into the function space: let
H D L2 .0; `/ RmC2 :
Thus the elements in this space can be denoted:

0 1
f ./
B f .0/ C
B C
B C
B f .s1 / C
B C
f DB f .s / C
B : i C:
B : C
B : C
B C
@ f .sm / A
f .`/
Define the linear operator A with domain in H :
D.A / D jf =f 00 ./ L2 .0; `/j; A f D gI

0 00 1
f ./; si < s < si C1
g D GJ @ f 0 .si C/ f 0 .si / A ; (2.41)
i D 0; 1; : : : ; m; m C 1
where
si C D limit si C ; 0 < ! 0;
si D limit si ; 0<!0
0 D 0I `C D `:
Then it is readily verified by integration by parts that A is self-adjoint with dense

domain, nonnegative definite, and
Z 1
A f; f D GJ jf 0 .s/j2 ds 0; (2.42)
0
which is recognized as the elastic energy. Indeed, the definition of g is designed to

achieve this. The moment-balance end conditions we need at each node are:
R si / GJ . 0 .t; si C/ 0 .t:si // D 0;

ri2 mi C mi ì h.t; (2.43)
where ri is the radius of gyration of the mass mi about the elastic axis at normal
distance ì . We note that if there are no masses at the ends 0 or `, then
0 .t; 0C/ D 0 D 0 .t; `/: (2.44)
We proceed similarly for the bending. We define:
Hh D L2 .0; `/ RmC4
and the closed linear operator Ah with domain in Hh :
0 1
f ./
B f 0 .0/ C
B C
B f .0/ C
B C
B C
B f .si / C
D.Ah / D B C 0000
B :: C with f ./ L2 .0; `/
B : C
B C
B f .sm / C
B C
@ f .`/ A
f 0 .`/
and
0 1
f 0000 ; .s/i < s < si C1
B f 00 .0/ C
B C
B 000 C
Ah f D gI g D EI B f .si C/ f 000 .si /; C (2.45)
B C
@ i D 0; : : : ; m; m C 1 A
f 00 .`/
and it is readily verified by integration by parts that Ah is self-adjoint and

nonnegative definite and
Z 1
Ah f; f D EI jf 00 .s/j2 ds (2.46)
0
the elastic energy in the bending mode. In addition we have to impose the nodal end
conditions
mi .h.t; R si // C EI .h000 .t; si C/ h.t; si // D 0;

R si / C ì .t; (2.47)
at the nodes s D si , and if there are no masses at the ends, the freefree conditions:
h000 .t; `/ D 0 D h000 .t; 0/;

h00 .t; `/ D 0 D h00 .t; 0/:
Next let H denote the cross-product Hilbert space:
H D Hh H
and define the operator:

Ah 0
As D
0 A
on H with domain:
D.As / D D.Ah A /
and

Ah h h
As x D for x D in D.As /:
A
Then As is self-adjoint and nonnegative definite with

Z ` Z `
As x; x D EI jh00 .s/j2 ds C GJ j 0 .s/j2 ds;
0 0
which is twice the potential energy stored

p in the structure. As before we note that
As has apositive square root denoted As , and the elastic energy is defined on
p
D As as
hp p i
E.x/ D As x; As x ;
where we use the same inner-product notation for all three spaces we have
introduced.
Finally the structure dynamic equations can now be expressed:
R C As x.t/ C Bu.t/ D 0
M x.t/ (2.48)
allowing for an actuator on the structure and control input u./ as in the fixed-wing
model, and M as before.
Structure Modes
We now calculate the structure modes for the uncoupled case S D 0. We use the
same notation as before in Sect. 2.4, including A() but we need to introduce more
to account for the fact that now we have several interconnected sections. Thus let
Ei denote the 6 6 matrix
0 1
1 000 0 0
B 0 100 0 0C
B C
B 0C
B 0 010 0 C
B C
B 2 mi mi ì C
2
B 0 0 1 0 C: (2.49)
B EI C
B EI C
B 0 0 0 0 1 0 C
B C
@ 2 m ` 2 mi ri2 A
i i
000 1
GJ GJ
Then the modes are the zeros of
d./ D Det:.PE mC1 eA./.`sm / Em eA./.sm sm1 / E1 eA./s1 E0 Qf /: (2.50)
If mi ; ri , and ì are zero, then we revert to d./ in Sect. 2.4. An important case is
the symmetric case where all the masses are on the elastic axis so that ri ; ì are all
zero.
The Symmetric Case
1. One point mass with s1 D `=2. Here
d./ D P e .A./`/=2 E1 e.A./`/=2 Qf ; (2.51)
which can be expressed:

F1 . ;
/sin
`;
where
1 2 m 1
F1 . ;
/ D 4
.1 C cosh ` cos `/ .1 C cosh ` cos `/
2 EI
!
1Ci .1 C i / ` .1 C i / `

sin sinh :
4 2 2
As we expect, the mass at the center does not affect the torsion modes. Hence we
call the roots of F1 . ;
/ D 0 the bending modes which now depend on
as
well. Obviously we get back the beam modes for large mass m1 .
2. Several Point Masses.
This result readily generalizes to the case of several point masses, but none at the
end points. Following the Helios model [81], we consider (N 1) masses mi at si ,
placed symmetrically:
`
si D i ; i D 1; : : : ; N 1;
N
where N is evenso that we have a mass at the center at `=2. Here

d./ D Det P eA./.`=N / EN 1 : : : : eA./.`=N / E N : : : eA./.`=N / Q which again
2
factors as
FN . ;
/ sin
`;
where the beam modes are the roots of
FN . ;
/ D 0:
2.7 Nonlinear Structure Models
So far we have only considered linear models, where the dynamics are characterized
by a linear equation. One may argue that real life structures are nonlinear. One
can of course come up with nonlinear models but the question of whether they are
well formulated in terms of existence and uniqueness of solutions is often ignored.
See for example [75, 81, 82] where in fact the elastic part is eventually truncated
lumpedto yield ordinary equations in place of partial differential equations. One
immediate difficulty with the nonlinear model is that there is no notion of modes; we
have no concept of spectrum of the nonlinear operator. And even if one could define
the notion, it is not clear what role it plays in the stability of the system, which is
our primary concern; see below.
Also we can no longer define an inner product based on the energy. We consider
two models that are essentially extensions of the generic Goland beam model.
The first one is due to Beran et al. [75]. We call it the BeranStraganac model.
This has only two degrees of freedom but is highly nonlinear.
2.7 Nonlinear Structure Models 41
The BeranStraganac Model
We state this in our notation where we replace their w./ by h./ and ./ by ./.
We have, including the discrete masses, in their delta-function notation:

mhR C S R C EI h0000 C Ms hR xs .t;
R y/ .y ys / D Dx .h0 .h0 h00 /0 /0 C L.t/
(2.52)
plus cross-product terms involving time derivatives which we omit.
I R C S hR GJ 00 C ./.y ys / D CDx .h00 /2

C cross-product terms involving time derivatives which we omit
C M.t/: (2.53)
For the derivation and details we have omitted see [75]. Our main point here is that
the equation is nonlinear so that there is no notion of eigenvalues, of the spectrum
of the differential operator.
Of course the authors do not consider the question of whether these continuum
equations have a unique solution, and all calculations are based on discretized
models. Indeed, it would be a major task to establish this. We return to this model
in the succeeding chapters.
The second example we consider is the DowellHodges model [76].
The DowellHodges Model
This is a nonlinear model with three degrees of freedom as in [61]: the torsion angle
./ about the elastic axis and two bending variables, the plunge h./ in the xz plane
and an additional in-plane bending in the structure xy plane. We show existence and
uniqueness by a constructive method of solution; our emphasis is on the differences
from the linear Goland model.
The continuum equations (consistent with our notation) are:
R y/ C EI 1 h0000 .t; y/ C .EI 2 EI 1 /..t; y/v.t; y/00 /00

mh.t;
D mg sin' C L.t; y/; (2.54)
mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /..t; y/h.t; y/00 /00 D mg cos'; (2.55)
R y/GJ 00 .t; y/C.EI 2 EI 1 /h.t; y/00 v.t; y/00 D M.t; y/; 0 < tI 0 < y < `;
I .t;
(2.56)
where we have omitted the tip masses in [76] but retained the gravity terms as part
of the forcing terms on the right. The angle ' is the angle between the x-axis and
the gravity vector. For the derivation and other details reference should be made to
the original paper [76].
The in-plane (in the xy-plane) displacement is denoted v./ and does not impact
the aerodynamics directly. This is a nonlinear system of equations (including the
airflow, based on the nonlinear aerodynamics described in Chap. 3).
We shall first consider the case where the airspeed is zero so that there is
no aerodynamic loading, as in the linear case. There is no notion of modes that
determines stability any more. We include the gravity terms in the aerodynamic
loading and hence omit them for the pure structure case. Then we have:
R y/ C EI 1 h0000 .t; y/ C .EI 2 EI 1 /..t; y/v.t; y/00 /00 D 0;

mh.t;
mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /..t; y/h.t; y/00 /00 D 0;
R y/ GJ 00 .t; y/ C .EI 2 EI 1 /v.t; y/00 h.t; y/00 D 0;
I .t;
0 < t; 0 < y < `; (2.57)
plus CF end conditions, with the notation:
0 1
h .t; y/
x.t; y/ D @ v t; y/ A t > 0; 0 < y < `:
.t; y/
One surprising feature of the model although still nonlinear, and not typical, is that
we do have modes; in fact those of the linear system.
h.t; y/ D 0;
v.t; y/ D 0;
I R .t; y/ GJ 00 .t; y/ D 0
plus CF or FF end conditions, yields all the pitching modes (cf. (2.5)). In similar
fashion, we also get all the bending modes. Thus we have:
.t; y/ D 0;
v.t; y/ D 0;
R y/ C EI 1 h0000 .t; y/ D 0
mh.t;
2.8 Beams of Infinite Length 43
yields all the plunge modes and
.t; y/ D 0;
h.t; y/ D 0;
mRv.t; y/ C EI 2 v0000 .t; y/ D 0
yields all the in-plane bending modes.

But of course we do not have the modal superposition property, except for the
class of functions for which only one co-ordinate is nonzero for all functions. Hence
it does not tell us much about the stability of the system. We return to this model in
succeeding chapters.
Wing Camber Model
So far we have not taken in-thickness into account. Ideally of course one would
want a plate model, however thin. For aircraft wings we have to model the tear-drop
shapethe camber; see [18].
This brings considerable complication in the continuum model, explaining why
all current works immediately discretize to finite dimensions. An approach to
include camber is presented in [107] for flutter analysis.
2.8 Beams of Infinite Length
It is certainly of mathematical interest to examine the case of a uniform Goland

beam, where the beam length is no longer finite. The main thing here is that we
need to continue with the same notion of energy:
Z 1 Z 1
1 1
ED EI jh00 .s/j2 ds C GJ j 0 .s/j2 ds:
2 0 2 0
An interesting consequence of this is that if the beam is clamped at one end s D 0,

then it would need to satisfy the same conditions at the end s D 1. Thus let H D
L2 0; 1 and let h./ and h0 ./; h00 ./, h000 be in H with
h.0/ D 0;
h0 .0/ D 0:
Then
d 0
jh.s/j2 D h0 .s/h.s/ C h.s/ h.s/
ds
and hence
Z L Z L
d
jh.s/j2 ds D jh.L/j2 D .h0 .s/h.s/ C h.s/.h.s//0 /ds;
0 ds 0
which converges as L ! 1 to
2 Reh; h0 :
Hence h.L/ converges as

L!1
but the limit must be zero because h.0/ is in L2 0; 1. In as much as h00 is also in
L2 0; 1, we have that h and h0 are continuous with
h.1/ D 0 D h0 .1/ D 0:
In a similar way for the torsion,
.0/ D 0 D .1/:
Thus C at one end implies C at the other end also. Thus we have the case CC.
The modes are inversely proportional to the length, therefore we see that the
modes are not defined. Or, the point spectrum of A is empty.
Our interest is in the stability of the structure. Thus we can proceed as we did in
Sect. 2.3 with CC end conditions, and ` replaced by 1. The first problem then is the
definition of the square root of A. Here we need to invoke Fourier transforms. Thus
Q as the Fourier transform of h; Q ./ as the Fourier transform of ./.
we denote h./

h hQ
x./ D I x./
Q D Q ;

Z 1 Z 1
Q
h./ D Q
e2 i s h.s/dsI ./ D e2 i .s/ .s/ds; 1 < < 1:
0 0
Q and ./
Thus defined, we have that h./ Q are in L2 .1; 1/ with the properties
Z 1 Z 1
jh.s/j2 ds D Q
jh./j 2
d:
0 1
Let F denote the operator on L2 0; 1 into L2 1; 1
hQ D F h:
2.8 Beams of Infinite Length 45
Then by the Plancheral theorem (see, e.g., [10, 51])
Q g:
h; g D h; Q (2.58)
F maps L2 0; 1 into a proper subspace of L2 .1; 1/ characterized by the

PaleyWiener theorem [51]:
Z 1 Q
jlogjjh.!/jjj
d!:
1 1 C !2
The operator A becomes a multiplier [22, 28]:
.Ax/ D mx;
Q
where m is the matrix:

!
EI .2 i /4 0
D m; (2.59)
0 GJ .2 i /2
which makes the definition of the square root as a multiplier

p !
p p EI .2v/4 0
A m D p : (2.60)
0 GJ .2v/2
With A defined as before, it can be identified with the 4 4 matrix multiplier

0 I2
(2.61)
M l m 0
and the semigroup S.t/ t 0 by the multiplier

0 I2
MatrixExp t ; (2.62)
M 1 m 0
which then yields an isometric group for 1 < t < 1. This should help answer
the question of what happens as the beam length is infinite.
Notes and Comments
The wing model we use is perhaps the simplest. It represents an unswept wing so
that it is rectangular; we neglect wing camber (see [18] for a detailed wing-shape
description) and assume zero thickness.
The use of functional analysis is now common in applied mathematics but has
yet to reach engineering and there is a disconnect here at present. Indeed, because of
software packages the level of analytical skill is actually decreasing! It is interesting
that the need for infinite-dimensional spaces is underscored by a recent publication,
Dynamics of Very High Dimensional Systems [76].
For the articulated FF case the derivation of the nodal conditions is novel in that it
is derived simply on the basis that the differential (stiffness) operator be nonnegative
definite, and not from physical principles. It also requires the inclusion of the nodal
values in the definition of the structure state. This technique was employed for the
first time in [54]. This may be considered as a generalization of the GolandLuke
model [77].
It should be noted that feedback controllers are all rate controllers well known in
classical control for stabilization.
For a detailed description of self-straining controllers, reference should be made
to [79] where the performance limitation due to operational amplifiers is included.
There are, of course, a great many nonlinear structures too numerous for
inclusion here even restricted to aircraft wings, neglecting thickness. Here we have
chosen two that relate closely to the Goland model. There is no general theory
for nonlinear models such as that for the linear case. The notion of elastic energy
does not seem adequate. We are unable to provide a time-domain solution at the
level of the Goland model. As we have noted, a generic feature here is that the
need to show that they have unique solutions seems not to bother the originators
who go on immediately to discretization of the model after the elaborate effort
for constructing the model; see, for example, [76]. Of course the problem here is
much less complicated than the viscous flow case (Chap. 7) and we only need the
linearized model (linearized about the steady state) for flutter analysis, as we show
in Chap. 6. Hence we resort to a perturbation technique leading to a Volterra integral
equation bootstrapping on the linear equation; see Chaps. 46.
All the beam models we consider have zero thickness. However, we can include
wing camber and we do so briefly.
Finally we consider the case where the beam length is allowed to be infinite and
this case is of mathematical interest in that there are no discrete modes any more, and
the Fourier transform theory and the notion of multipliers provide the appropriate
techniques to this case.
Chapter 3
The Air Flow Model/Boundary Fluid Structure
Interaction/The Aeroelastic Problem
3.1 Introduction
In this chapter we make a precise mathematical statement of the aeroelastic problem

that we wish to solve. Having described the structure model, we turn to the air flow
model simplifying it to the most used case where we neglect viscosity and consider
nonviscous flow but more importantly assume that the entropy is constant. This
makes the flow vortex free so that the flow can be described in terms of the potential.
Our concern is again more the structure response in air flowaeroelasticityand
hence the fluidstructure boundary conditions play the dominant role in determining
the aerodynamic loading on the wing structure.
Starting with the three basic conservation laws, we derive the fundamental field
equation describing the air flow, The Eulerfull potential equation with the Kutta
Joukowsky boundary conditions. We present a complete statement of the aeroelastic
problem at the end of the chapter for nonviscous flow and nonlinear structure
models, including the simplification to Strip theory, the typical section theory.
3.2 Notation/Physical Constants
We begin with the notation we use for the basic parameters necessary to describe the
flow generally throughout the book from now on. We also list the relevant physical
constants we need in the process:
Density
q Fluid velocity vector
q1 Far field velocity
U1 D jq1 j This is a free parameter, the far field air speed
a1 D Speed of sound
p Pressure
Viscosity

48 3 The Air Flow Model/Boundary Fluid Structure Interaction
S Entropy
T Temperature
e Energy per unit volume
Subscript 1 denotes far field values
Perfect Gas Law

p D RT
R D cp cv
D cp =cv Ratio of specific heats
Thermodynamic Relation:
p es=cv D Const.
h enthalpy per unit mass D cp T
E internal energy per unit mass D cv T
All are functions of time t and the spatial coordinates x; y; z:
f D f .t; x; y; z/
The scalar functions are all positive.
Physical Constants:
All at standard air 59 F
0.00238 slug/ft3
1.23 kg=m3
0:372 106 slug=ftsec
17:8 106 kg=msec
p 2; 116; lb=ft2
1:0312 105 N=m2
a1 336 m=sec
Re Reynolds Number U 1= 106
R Gas Constant 287 kg=msecunits
cv 717
cp 1,004
1. 4
k Diffusivity D cp
17:8 103
Prandtl no. 0:7 (here taken as 1)
k cp
D D
R1 R1 1 1
3.3 Nonviscous Flow: The Euler Full Potential Equation 49
3.3 Nonviscous Flow: The Euler Full Potential Equation
Throughout this chapter we set D 0.

The field equation of fluid flow in 3D space (R3 , orthogonal coordinates x; y; z)
is derived from three basic laws of conservation which we state here in differential
form (as opposed to the integral form) with t representing the time co-ordinate.
We begin with the premordial.
Conservation Laws
1. Conservation of mass: Continuity equation
@
C r .q/ D 0: (3.1)
@t
2. Conservation of momentum. The Euler momentum equation (nonviscous flow,
no heat conduction)
@q
C .q r/q C rp D 0 (3.2)
@t
in the usual notation [4, 12], where .q r/q is the vector:
i q r.q i / C j q r.q j / C kq r.q k/ q D i q 1 C j q2 C kq3 ;
where i; i; k are the orthogonal unit vectors.

Note that second spatial derivatives of q are not involved, unlike the viscous
case (see Chap. 7).
3. Conservation of energy: First law of thermodynamics in Eulerian form for perfect
gas [14]
1 p
ED ;
1
DE
C pr q D 0: (3.3)
Dt
Or
@E
C q rE C p rq D 0: (3.4)
@t
Note that we need three equations: conservation of mass, conservation of
momentum, and conservation of energy. All are expressed in differential form
rather than in integral form. This makes it possible to state the dynamic boundary
conditions crucial for aeroelasticity.
This equation (3.4) implies [12, 14] that the total derivative of the entropy S is
zero:
DS
D 0; (3.5)
Dt
or the flow is homentropic [14].
Note that:
pT S
are thermodynamic state variables any one of which is determined by the other
three. Thus, under our assumption that the specific heats are constants, we have
[14]:
p D constant es=cv : (3.6)
Isentropic Flow
A flow is said to be isentropic if
S D constant in t; x; y; z:
Homentropy does not imply isentropy. But we now assume that the flow is
isentropic. In particular then the energy equation is satisfied. And from (3.6) we have
the important conclusion that the pressure is a function of density. More specifically:
p D A ; (3.7)
where A is a constant determined from

p1 1 D A:
Next
rp A
D r

D A 1 rlog
D A 1 rlog

DA 1 rlog 1
1

DA r 1 : (3.8)
1
Hence the Euler equation can be purged of the pressure variable p to yield:
@q
C .q r/q C A r 1 D 0: (3.9)
@t 1
This is what enables us to deduce that the flow is curl or vortex free. Thus let
D r q:
Then taking the curl on both sides of (3.9), we obtain:
@q
r C r .q r/q D 0
@t
because r r 1 D the curl of a gradient is zero.

Hence
@
C r .q r/q D 0:
@t
But
1
.q r/q D r.q : q/ q : (3.10)
2
Hence finally:
@
C r .q / D 0: (3.11)
@t
Consistent with our assumption that the far field is q1 is that
q.0; x; y; z/ D q1 D r1
and hence
.0; x; y; z/ D 0:
Now for any (smooth enough) solution q.t; x; y; z/, we may consider ./ as a
solution of (3.9) which is a linear equation with zero initial condition and has the
identically zero solution
.t; x; y; z/ D 0 D r q.t; x; y; z/:
Or the velocity q ./ is curl-free. Hence it is expressible as
q D r:
And .t; x; y; z/ is the velocity potential such that the far field potential
p
.t; x; y; z/ where r D z2 C y 2 C x 2 ! 1;
D x.q1 i / C y.q1 j / C z.q1 k/
D 1 .0; x; y; z/:
The Euler Full Potential Equation

Assuming entropic flow, we are now ready to derive the field equation for the
velocity potential ./ using the continuity equation. First though we rewrite the
Euler equation using (3.7) and (3.8) as:
@r 1 A
C rr; r C r 1 D 0: (3.12)
@t 2 1
Hence
@ 1
C r; r C A 1 D constant D far field values
@t 2 1
1
D q1 ; q1 C A 1 :
2 1 1
Now because p is a function of ,
dp
D A 1 D a2 ;
d
where a is the local speed of sound with the far field value a1 , so that
2 1
a1 D A 1 ;
or
2
a1
A D 1
: (3.13)
1
Hence we have
!
@ 1 2 a2 1
C 2
U U1 D 1 1 1 ;
@t 2 1 1
where we use U for the flow speed. Hence we have for the density:

1 1 1 1 2 2
D 1 1 2
U U1 C @t : (3.14)
a1 2
Now using the continuity equation:
@ 1 @
D . 1/ 2
@t @t
D . 1/ 2 r .r/
D . 1/ 2 . C r r/
D . 1/ 1 r r 1
using
r r 1
. 1/ D 1 :

From (3.14) we have:

@ 1 1 @2 1
D 1 C r @tr/
2
@t @t 2 a1

1 1 1 2
r 1 r D 1 2
r r C @tr r:
a1 2
Hence

1 1 @2 @r
1 2
C r
a1 @t 2 @t

1 1 1 2 2
@
D . 1/1 1 U U C
2
a1 2 1 @t

1 1 1 2 @r
1 2
rjjrjj C r:
a1 2 @t
Hence finally:

@2 @r 2 1 1 2 2
@
C r D a1 1 U U 1 C
@t 2 @t 2
a1 2 @t
1 @r
r r.jjrjj2 / r;
2 @t
which we can rewrite in the form:
2
@2 @ 2 2 1 U1 jjrjj2 @
C jjrjj D a1 1 C 2
@t 2 @t a1 2 2 @t

jjrjj2
r r : (3.15)
2
This is the 3D Euler full potential equation valid except on the boundary specified
later.
Acceleration Potential and Pressure
We can now derive an explicit expression for the pressure in terms of the fluid
velocity. The acceleration is defined by the total derivative of the velocity:
Dq @q
a.t/ D D C .q r/q
Dt @t
and for potential flow by (3.10)
1
.q r/q D r.q q/:
2
Hence

1
a.t/ D r @t C jrj2 D r .t/;
2
where the acceleration potential
jrj2
.t/ D @t C (3.16)
2
and the far field value
2
U1
1 D :
2
The flow being isentropic
p D A
and by (3.14)
1=. 1/
1
D 1 1 . 1/ 2
(3.17)
a1
and hence =. 1/
. 1/
pD A1 1 . 1/ 2
: (3.18)
a1
And a very reasonable (and universal) approximation here is to take:

. 1/
p D A1 1 2
(3.19)
a1
consistent with
U1
0<M D < 1;
a1
where M is the Mach number, and that the perturbation of the flow by the airplane
is small compared to the far field speed. In any event we use (3.19) for p throughout
for isentropic flow.
The main relations we have are thus (3.19) and (3.15).
What distinguishes aeroelasticity from aerodynamics is of course the interaction
with the structure dynamics on the boundary, a singular boundary that complicates
the problem further.
Boundary Conditions
The far field condition
.t; 1/ D U1 .x cos C y sin /;
where is the specified angle of attack.

As with any field equation, the conditions on the boundary determine the
solution. Here it is further complicated by the fact we have taken into account the
structure dynamics as well.
Flow-Structure Interaction:
Hence we first need to specify the structure model.
The Simplest Wing Structure Model
The simplest wing model is a slender thin plate whose thickness is then taken to
be zero, rectangular in shape (unswept wing) uniform, with ` denoting the half wing
span and 2b the width or chord length, so that b is the halfchord.
We only consider wings of high-aspect ratio:
` b;
which would justify the flexible model. However, we do consider a finite rectangular
plane (Finite plane) as the boundary for the aerodynamic equations. Thus we have
3D aerodynamics and a 2D wing boundary.
We choose the spatial co-ordinate system consistent with the aircraft rigid body
dynamics. Thus the x-axis is along the airplane axis, the y-axis is the span axis and
the negative z-axis is the plunge axis. Thus the boundary for the field equation is
the rectangle in the xy-plane described by:
D fb x bI 0 y `g:
For the structure dynamics, however, we specialize to a beam model, ignoring the
dependence on the chord variable. Such a model was described by Goland, as we
have seen in Chap. 2 where the structure is endowed with two degrees of freedom:
plunge and pitch.
The plunge displacement denoted h .t; y/ is then the displacement of the wing
along the negative z-axis. It is uniform along the chord. The pitch #.t; y/ is the twist
angle in radians about an axis parallel to the y-axis at distance ab from the x-axis
and again does not depend on the chord variable x. Hence h.t; y/ and #.t; y/ are
defined on:
0 tI 0 y `:
The wing is fixed to the fusilage. Thus we have a clampedfree (CF) model with:
h.t; 0/ D 0 D h0 .t; 0/I #.t; 0/ D 0:

The wing is free at the other end, and hence
h000 .t; `/ D 0 D h00 .t; `/ D 0I # 0 .t; `/ D 0:
These are then the end conditions to be satisfied in addition to the

force/momentum balance equations.
The resulting structure dynamic equations are described in Chap. 2. Here we shall
describe the airwing interaction dynamics.
Thus we have two sets of conditions.
I. The Attached Flow Condition

The normal velocity of the air along the wing is equal to the normal velocity of
the wing. We may need to distinguish between the top and bottom of the wing if
we allow for discontinuity in the fluid velocity across the wing, even though the
thickness is zero.
The total displacement of the wing is given by:
!

k z.t; x; y/ where z.t; x; y/ D .h.t; y/ C .x ab/ # .t; y//
and hence the normal velocity of the wing is given by:

!
Dz.t/ ! @z.t/
k D k C q.t; x; y; 0/ rz.t/
D.t/ @t
P
! P y/ C .q.t; x; y; 0/ !

D k h.t; y/ C .x a/#.t; i /#.t; y/

! @
C q.t; x; y; 0/ j .h.t; y/ C .x a/#.t; y// :
@y
Hence allowing for discontinuity in the flow, the flow tangency conditions become:

@.t; x; y; 0C/ @1 P y/ C .x a/#.t;
P y/
D C .1/ h.t;
@z @z
!
!
C .q.t; x; y; 0C/ i /#.t; y/ C .q.t; x; y; 0C/ j /

@
.h.t; y/ C .x a/#.t; y// ; x; y
; (3.20)
@y
@.t; x; y; 0/ @1 h

D P y/ C .x a/#.t;
C .1/ h.t; P y/.q.t; x; y; 0/ !

i/
@z @z

! @
#.t; y/C .q.t; x; y; 0/ j / .h.t; y/C.xa/#.t; y// ;
@y
x; y
: (3.21)
II. KuttaJoukowski Conditions

These conditions are peculiar to aeroelasticity and are given in terms of the pressure
which is discontinuous across the wing. We define the pressure Jump by p:
p.t; x; y/ D p.t; x; y; 0C/ p.t; x; y; 0/: (3.22)
Defining
.t; x; y/ D .t; x; y; 0C/ .t; x; y; 0/;
we have from (3.20) that

p D A1 D 1 : (3.23)
a21
II. The pressure jump is zero off the wing
p.t; x; y/ D 0 for x; y not in : (3.24)

Or
.t; x; y/ D 0 for x; y not in : (3.25)
II. Kutta condition
p.t; x; y/ D 0 as x ! b; x; y in : (3.26)

Or
.t; x; y/ D 0 as x ! b; x; y in : (3.27)
Structure Dynamics
We are now ready to complete the structure dynamics models begun in

Chap. 2 by including the aerodynamic lift and moment which are expressed in
terms of the structure state variables.
Goland Model
Z b
R y/ C S #.t;
mh.t; R y/ C EI h0000 .t; y/ D L.t; y/ D p.t; x; y/dx;
b
0 < y < `; (3.28)
Z b
R y/ C S h.t;
I# #.t; R y/ GJ # 00 .t; y/ D M.t; y/ D .x ab/p.t; x; y/dx
b
0 < y < `: (3.29)
DowellHodges Model
R y/ C EI 1 h0000 .t; y/ C .EI 2 EI 1 /.#.t; y/v.t; y/00 /00 D mg sin' C L.t; y/;
mh.t;
mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /.#.t; y/h.t; y/00 /00 D mgcos ';
R y/ GJ # 00 .t; y/ C .EI 2 EI 1 /h.t; y/00 v.t; y/00 D M.t; y/;
I# #.t;
0 < tI 0 < y < `:
3.4 Problem Statement
The Aeroelastic Problem: 3D/Nonzero Angle of Attack/2D

Boundary
We can now present a complete statement of the combined 3D aeroelastic problem

by the following equations.
Field equation
2
@2 2 2 1 U1 jjrjj2
C @t jjrjj D a1 1 C @t
@t 2 a12 2 2

jjrjj2
r r : (3.30)
2
3.4 Problem Statement 59
Far field
.t; 1/ D xU 1 cos C yU 1 sin:
Boundary conditions
D ffx; yg; b < x < bI 0 < y < `g

@.t; x; y; 0C/ @1 h
D P y/ C .x a/#.t;
C .1/ h.t; P y/
@z @z
!
!
C .q.t; x; y; 0C/ i /#.t; y/ C .q.t; x; y; 0C/ j /

@
.h.t; y/ C .x a/#.t; y// ; x; y
: (3.31)
@y
And
@.t; x; y; 0/ @1 h

D P y/ C .x a/#.t;
C .1/ h.t; P y/
@z @z
!
!
.q.t; x; y; 0/ i /#.t; y/ .q.t; x; y; 0/ j /

@
.h.t; y/ C .x a/#.t; y// ; x; y
: (3.32)
@y
KuttaJoukowski conditions
@ jjrjj2
.t/ D C ; (3.33)
@t 2
.t; x; y/ D 0 for x; y not in ; (3.34)
.t; x; y/ D 0 as x ! b; x; y in : (3.35)
Structure Dynamics: Linear (Goland)

Z b
R y/ C S #.t;
h.t; R y/ C EI h0000 .t; y/ D 1 .t; x; y/ dx; 0 < y < `;
b
(3.36)
Z b
R y/ C S h.t;
I# #.t; R y/ GJ # 00 .t; y/ D 1 .x ab/ .t; x; y/dx 0 < y < `;
b
(3.37)
plus CF end conditions:
h.t; 0/ D 0 D h0 .t; 0/I #.t; 0/ D 0;

h000 .t; `/ D 0 D h00 .t; `/ D 0I # 0 .t; `/ D 0;
or FF end conditions:
h000 .t; 0/ D 0 D h00 .t; 0/ D 0I # 0 .t; 0/ D 0;

h000 .t; `/ D 0 D h00 .t; 1/ D 0I # 0 .t; `/ D 0:
Structure Dynamics Nonlinear
Beran Straganac
The state variables are the same as in the Goland. For the nonlinearities see Chap. 2,
Sect. 2.7. The end conditions are also the same as in the Goland model.
DowellHodges
Here there is an extra state variable v.t; /:
R y/ C EI 1 h0000 .t; y/ C .EI 2 EI 1 /.#.t; y/v.t; y/00 /00

mh.t;
Z b
D mg sin' 1 .t; x; y/dx; 0 < y < `; (3.38)
b
mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /.#.t; y/h.t; y/00 /00 D mg cos'; (3.39)
R y/ GJ # 00 .t; y/ C .EI 2 EI 1 /h.t; y/00 v.t; y/00
I# #.t;
Z b
D 1 .x ab/ .t; x; y/dx0 < tI 0 < y < `: (3.40)
b
The end conditions for v.t; :/ are the same as for h.t; :/.
Typical Section (Strip) Theory
A universally invoked simplification is to neglect the dependence on the y-co-

ordinate in consequence of the assumed high-aspect ratio `=b of the wing.
Problem Statement
Typical Section Theory
2D Aerodynamics 1D Structure
The field equations: with the y-co-ordinate omitted:

2 2 !
@2 @ @ @
C C
@t 2 @t @x @z
0 0 2 2 11
@ @
B C
2 B 1 B
BU
2
@x @z @ C
CC
C
D a1 B1 C 2 B 1 CC
@ a1 @ 2 2 @t AA
0 2 2 1
@ @
B C C
B @x @z C
r r B C 1 < x; z < 1: (3.41)
@ 2 A
Far field potential
.t; 1/ D xU 1 cos C zU 1 sin:
Boundary conditions
is now just the Chord: b < x < b and the flow tangency condition becomes:
!
Total Displacement D k z.t; x/ where
z.t; x/ D .h.t; y/ C .x ab/#.t; y//

and hence the normal velocity of the wing is given by

Dz.t/ !
! @z.t/
k D k C q.t; x; y; 0/ rz.t/ ;
D.t/ @t
P
!
!
D k h .t; y/ C .x ab/#P .t; y/ C q .t; x; y; 0/ i #.t; y/ :
Hence allowing for discontinuity in the flow, the flow tangency conditions become:
@.t; x; 0C/ @1 h

D C.1/ h.t;P y/C.x a/#.t;
P y/
@z @z

! i
C q.t; x; 0C/ i #.t; y/ : (3.42)
And
@.t; x; 0/ @1 h

D C .1/ h.t;P y/ C .x a/#.t;
P y/
@z @z

! i
C q.t; x; 0/ i #.t; y/ x
: (3.43)
KuttaJoukowski Conditions
p.t; x/ D p.t; x; 0C/ p.t; x; 0/;

.t; x/ D .t; x; 0C/ .t; x; 0/;
p D 1 :
The pressure jump is zero off the wing.
p.t; x/ D 0 for jxj > b;
Or
.t; x/ D 0 for jxj > b:
Kutta condition
p.t; x/ ! 0 as x ! b; in:
Or,
.t; x/j ! 0asx ! b; jxj < b:
The structure equations with the indicated aerodynamic loading remain the same.
This is then the precise statement of the aeroelastic problem continuum equa-
tions. The objective is to determine the stability of the structure state as a function
of U1 , the air speed.
To anticipate the theory that follows, the main conclusion is that for a given value
of M (equivalently, speed of sound, equivalently altitude) there is a speed, called
flutter speed, denoted UF , for U1 < UF , the structure is stable (see later for precise
definition of stability), and for U > UF the structure is unstable.
Notes and Comments
It is interesting to note that none of the books on aeroelasticity, including Dowell

et al. [17], Hodges and Pierce [5] or Bisplinghof, Ashley, and Halfman [6] care
to make a precise statement of the aeroelastic problem as we do in this chapter.
Indeed without such a statement it is not clear what it is that the computer codes
used universally (see the many recent papers on aeroelasticity, for example [75, 81
83, 93]) are providing the (approximate) solution to, even omitting the cases where
the solvability of the problem cannot be established.
Indeed without such a formulation it is not possible to define the Flutter Speed
calculating which is a main objective of the theory.
We should note that most progress has been made for the typical section case (2D
aerodynamics) and it is fortunate that flexibility is consistent with high-aspect ratio
so that the typical section approximation is reasonable (without necessarily being
very high).
Regarding the foundational conservation laws, following [4,12] we have invoked
three of them rather than the first two as in [4,17], for example. The triad is essential
for aeroelasticity. As Meyer [14] notes the third is the Euler version of the first law
of thermodynamics.
Chapter 4
The Steady-State (Static) Solution
of the Aeroelastic Equation
4.1 Introduction
In this chapter we specialize to the time-invariant version of the aeroelastic

equations of Chap. 3 where we set all the time derivatives to zero and there is no
input. It is called the static solution in that there is no change with time. It is of
interest on its ownit is in fact central to the study of stabilitybut it also serves
to illustrate the solution techniques used for the general case in Chap. 6.
Often this solution is also referred to as the steady-state solution, and so we need
to clarify the terminology. The term steady-state response is used to indicate the
response of a linear system to a steady input, such as a sinusoid input which is then
not dependent on the initial condition. Here we have the case of zero input. So we
take an arbitrary initial condition (and far field speed) and let time march (literally
in computational programs) until there is no longer any change with time; in other
words we consider the asymptotic response in time. Physically we are considering
a pointwise limit of the potential as well as the structure state. Of course depending
on the assumed speed parameter and the initial conditions there may not be such a
solution. We largely follow [67]. To begin with, we can state the following.
Theorem 4.1. Every steady-state solution is a static solution.
Proof. If a steady-state limit exists, and there is no change in time, then it must
satisfy the static equation. t
u
In our case in addition to the initial condition we have a parameter to specify the
far field speed. And we show that for a sequence of speeds there is indeed more than
one solution to the static aeroelastic equation. However, as we show below, there
is one static solution for all speeds which is also a steady-state solution as well.
This provides us a simple example where the continuum equations need not have a
unique solution depending on the far field speed.
The steady-state solution depends obviously on the structure model used.
We begin with the workhorse model, the linear Goland model.

66 4 The Steady-State (Static) Solution of the Aeroelastic Equation
4.2 Goland Structure Model
We first consider the Goland linear structure model. We set all time derivatives in
the field and boundary equations to zero. And all functions are independent of time
so we drop the time variable. Thus we begin with the velocity potential .x; y; z/
which satisfies the 3D time-invariant Euler full potential equation, purging the time
derivatives in (3.15), yielding now:

1 1
a1 C
2
.jr1 j jrj / r r.jrj2 / D 0;
2 2
2 2
which we may expand as

" 2 2 2 !#
. 1/ @ @ @
0D 2
a11C 2
U1
2
.4/
2a1 @x @y @z

1 @ @ 2 @ @ 2 @ @ 2
r C r C r : (4.1)
2 @x @x @y @y @z @z
Angle of Attack
We assume that the far field velocity vector is in the XZ plane making an angle
with the X -axis. This angle in flight control rigid body model [9] is called the angle
of attack. Hence
1 D U1 .x cos C z sin /;
q1 D U1 .i cos C k sin /:
We next specialize to the time-invariant structure dynamics.
Time Invariant Structure Dynamics
@4 h.y/
EI D L.y/; (4.2)
@y 4
@2 .y/
GJ D M.y/ (4.3)
@y 2
4.2 Goland Structure Model 67
with the CF boundary conditions:
h.0/ D 0I h0 .0/ D 0I .0/ D 0;

00
h .1/ D 0I h000 .1/ D 0I 0 .1/ D 0:
Next the fluid-structure boundary conditions.

I. Flow Tangency Condition
@
.x; y; 0C/ D U1 sin .q.x; y; 0C/ i /.y/ .q.x; y; 0C/ j /.h0 .y/
@z
C.x ab/ 0 .y//; (4.4)
@
.x; y; 0/ D U1 sin .q.x; y; 0/ i /.y/ .q.x; y; 0/ j /.h0 .y/
@z
C.x ab/ `0 .y//: (4.5)
II. The KuttaJoukowsky conditions
p.x; y/ D 0 x; y 62 p.b; y/ D 0: (4.6)
Theorem 4.2. The time-invariant equations are satisfied by the rest or

equilibrium solution:
.x; y; z/ D 1 .x; y; z/:
Structure at rest:
h.y/ D 0 and .y/ D 0 0y1

for all far field speeds.
In this case the fluid velocity and pressure are continuous across the wing.
Proof. Verified by direct substitution into the equations. The solution holds for
every value of the far field speed. u
t
We now show the existence of a nonzero static solution.
The Nonzero Static Solution
To find the nonzero static solution we need to solve the nonlinear static Euler
equation. We show that because of the nature of the boundary conditions, we can
obtain a series expansion, a solution technique that works also in the dynamic case,
but is of course less complex here.
In the structure equations we need to calculate L.y/ and M.y/ for each
y; 0 < y < `. If we fix y, however, the structure variables h.y/ and .y/ may
be looked upon as parameters which enter linearly in fact in (4.4) and (4.5) and
we may postulate that the solution is analytic in these parameters, at least locally.
The solution in other words can be expanded in a power series (for some nonzero
radius of convergence) about
h.y/ D 0I .y/ D 0:
This is conveniently done by introducing the complex number parameter and
considering the response to
h.y/; .y/
for given h.y/; .y/. More specifically, let the solution be denoted .; x; y; z/
corresponding to h.y/; .y/, so that
.0; x; y; z/ D 1 .x; y; z/:
And we have the expansion

1
X k
.; x; y; z/ D 1 .x; y; z/ C k .x; y; z/; (4.7)
k
kD1
where
@k
k .x; y; z/ D .0; x; y; z/: (4.8)
@k
Before discussing the sense in which the series converges, let us first see how we
can calculate the derivatives for each point .x; y; z/. We go back to the field equation
(4.1) and differentiating it once with respect to and setting D 0, we obtain for
the first term therein:
2
a1 1
and for the second term:
@2 1 2
2 @ 1 @2
U1
2
cos2 U 2
1 sin 2U 2
1 cos sin :
@x 2 @z2 @z@x
Hence 1 satisfies the field equation:

.1 / D 0;
where
@2 @2 @2
./ D .1 M 2 cos2 / C C .1 M 2
sin 2
/
@x 2 @y 2 @z2
@2
2M 2 sin cosff ; (4.9)
@x@z
where the Mach number

U1
M D < 1:
a1
Thus 1 satisfies a linear homogeneous equation, with boundary conditions de-

scribed presently. Note that the thermodynamic constant does not appear.
For the higher-order potentials k we need to work a little harder. Let us fix the
spatial coordinates and let:

1 2
f ./ D a1 2
C .U1 jr./j2 / ; (4.10)
2
g./ D ./: (4.11)
Then we need to expand the first term in (4.1) the product
f ./g./
in a Taylor series. Now g./ has the expansion

1
X k
g./ D 4k :
k
kD1
Hence
1
X " X 1
!
k 1 k
f ./g./ D 2
a1 4k C U1 q1 C
2
rk
k 2 k
kD1 kD1
1
!# 1
!
X k X k
q1 C rk 4k
k k
kD1 kD1
1 1
!" 1
X k X k X k
D a1
2
4k C 4k .1 / q1 qk
k k k
kD1 kD1 kD1
1
! 1
!#
1 X k X k
C qk qk
2 k k
kD1 kD1
0 1
X 1 k X1 X 1 j k

D a1
2
4k C .1 / @ 4j q1 qk A
k j D1
j k
kD1 kD1
1 1 1
1 X X X i j k
C qi qj 4k ; (4.12)
2 i D1 j D1 i j k
kD1
where we have used the notation
qk D rk :
Next the second term in (4.1) is best handled using x1 ; x2 ; x3 , respectively, in place
of x; y; z. Thus it can be expressed:
X 3
@ X @ @2
3

i D1
@xi j D1 @xj @xi @xj
X
3 X 1 X
3 X 1 X
1
1 @n @m @2 p
D .nCmCp/ ;
i D1 j D1 nD0 mD0 pD1
nmp @xi @xj @xi @xj
where
0 D 1
so that
@k 0
D 0 k > 1:
@x k
Hence we have the expansion for (4.1):
0 1
X1 k X1 X 1 j k

2
a1 4k C .1 / @ 4j q1 qk A
k j D1
j k
kD1 kD1
1 1 1
1 X X X i j k
C qi qj 4k
2 i D1 j D1 i j k
kD1
X
3 X 1 X
3 X 1 X
1
1 @n @m @2 p
.nCmCp/ D 0; (4.13)
nmp @xi @xj @xi @xj
where the third term:

X
3 X 1 X
3 X 1 X
1
1 @n @m @2 p
.nCmCp/
nmp @xi @xj @xi @xj
X 3 X 3 X 1
1 p @0 @0 @2 p
D
i D1 j D1 pD1
p @xi @xj @xi @xj
X
3 X 1
3 X
1 mCp @0 @m @2 p
2
i D1 j D1 mD1
mp @xi @xj @xi @xj
X
3 X 1 X
3 X 1 X
1
1 @n @m @2 p
.nCmCp/ :
nmp @xi @xj @xi @xj
Next let us examine the boundary conditions:
@z .; x; y; 0C/ D U1 sin .@x .; x; y; 0C/.y/

@h @
C @y .; x; y; 0C/ C .x ab/ :
@y @y
Hence
@1 .x; y; 0C/ @1 .x; y; 0/
D U1 cos .y/ D (4.14)
@z @z
and for k 2:

@k .x; y; 0C/ @k1 .x; y; 0C/
D k .y/ C .@k1 .x; y; 0C//
@z @x
. @h @

@y C .x ab/ : (4.15)
@y @y
KuttaJoukowsky Condition
0 1
1 X
X 1 1
X
1@ 2 kCj k
D U1 C rk rj C 2 r1 rk A : (4.16)
2 kj k
kD1 j D1 kD1
From which we can derive the power series expansion:

2 X k 1
U1
.; x; y; z/ D C k .x; y; z/;
2 k
kD1
1 .x; y; z/ D r1 r1 : (4.17)
And hence
1 D 1 .x; y; 0C/ 1 .x; y; 0/
D U1 Cos ..@1 .x; y; 0C//=@x .@1 .x; y; 0//=@x/ (4.18)
using (4.15) and (4.16), and as we have noted D 0 off the wing.
We pause here at this point to examine in detail the linear case:
k D 1:
4.3 Linear Aeroelasticity Theory: The Finite Plane Case
Linear theory has an importance all its own and in addition we actually bootstrap
about it for the nonlinear case. And we can carry the solution some distance for
the general 3D aerodynamics. We do not discretize it now (as in most work in
aeroelasticity, as we have noted already). The field equation is, as we have seen:
.1 / D 0:
We rewrite this as an equation in z because z D 0 contains the boundary.
@2 1 @2 1
.1 M 2 sin2 / 2M 2
sin cos
@z2 @x@z
@2 1 @2 1
D .1 M 2 cos2 / : (4.19)
@x 2 @y 2
The first step is to choose the function space for the solution.
For each z we seek a solution in
Lp R2 ; 1 < p < 2:
We define the Lp Lq Fourier transform:

Z 1 Z 1
O1 .z; !1 ; !2 / D 1 .x; y; z/e.i!1 Ci !2 / dxdy !1 ; !2 2 R2 (4.20)
1 1
and note that the Fourier transform of the right side of (4.21) yields
.!12 .1 M 2 Cos2 / C !22 /O1 .z; !1 ; !2 /:
Hence in terms of the Fourier transforms, (4.20) becomes an equation in Lq R2 :
@2 O1 @2 O 1
.1 M 2 sin2 / C 2M 2
sin cos
@z2 @x@z
D .!12 .1 M 2 cos2 / C !22 /O1 .z; !1 ; !2 /: (4.21)
Let
@ O
vO .i !1 ; i !2 / D 1 .0; i !1 ; i !2 /:
@z
Then we can show that the only solution which goes to zero (in the Lp norm) as
jzj ! 1 is given by
1
O1 .z; !1 ; !2 / D ezr1 vO .i !1 ; i !2 / for z > 0 (4.22)
r1
1 zr2
D e vO .i !1 ; i !2 / for z < 0; (4.23)
r2
4.3 Linear Aeroelasticity Theory: The Finite Plane Case 73
where
p 2
r1 D .M 2 .i !1 /sin cos .!2 .1 M 2 sin2 /
C !12 .1 M 2 ///=.1 M 2 Sin2 /; (4.24)
p 2
r2 D .M 2 .i !1 /sin cos C .!2 .1 M 2 sin2 /
C !12 .1 M 2 ///=.1 M 2 sin2 / D r1 : (4.25)
We only need to calculate the pressure jump across the wing:
p D 1 :
Let
1
A1 .x; y/ D (4.26)
U1
called the Kushner doublet function; see [6] which by (4.15)
D cos 1 :
Hence the L1 Fourier transform

Z b Z `
AO1 .i !1 ; i !2 / D eix!1 iy!2 A.x; y/dxdy
b 0

1 1
D i !1 cos vO 1 .i !1 ; i !2 / ;
r1 r2
or,
O !1 ; i !2 /
1 A.i
vO 1 .i !1 ; i !2 / D .!12 .1 M 2 cos2 /
2 i !1 cos
p
C !22 /=. .!12 .1 M 2 / C !22 //
and by (4.15)
O 2 /;
D U1 cos .i! (4.27)
where
Z `
O
.i!/ D `.y/eiy! dy:
0
Hence we have that:

O !1 ; i !2 / D U1 cos O .i !2 /;
PO .i !1 ; i !2 /A.i
where
1 1 h
PO .i !1 ; i !2 / D !12 .1 M 2 cos2
2 i !1 cos
. i
p 2
C !22 .!1 .1 M 2 C !22 (4.28)
Z 1 Z 1
PO .i !1 ; i !2 / D P .x; y/eix!1 iy!2 dxdy:
1 1
This is the static version of the 2D integral equation of Possio which we describe
later and which plays a crucial role in our theory.
Z b Z `
P .x
; y /A.
; /d
d D U1 .y/; 0 < y < 1; jxj < b; (4.29)
b 0
where the Kutta condition becomes
A.b; y/ D 0 0 < y < 1: (4.30)
Let us assume that this equation has a solution in Lp b; b. Then
p D 1 U1 A.x; y/: (4.31)
We expect the solution to be of the form:

Z `
A.x; y/ D L.x; y; /./d; x; y on the wing (4.32)
0
and hence, getting back to the structure equations (4.2) and (4.3), we have:
Z b Z `
GJ 00 .y/ D .x ab/ L.x; y; /./ddx
b 0
Z `
D M.y; /./d: (4.33)
0
And with the CF end conditions we have thus a linear eigenvalue problem that would
have a solution at most for a sequence of values of the far field speeds which we may
call divergence speeds. But not being able to calculate the kernel M (., .), we cannot
do much more. So we proceed to simplify the problem where we can say more.
The first and most important simplification is to consider the following.
Typical Section: Airfoil Theory
We assume that the aspect ratio `=b is high enough so we may consider the air flow
on the wing to be uniform so that there is no dependence on the span variable y,
also called the strip theory. In this case the linear field equation (4.20) simplifies to:
@2 1 @2 1 @2 1
.1 M 2 sin2 / 2
2M 2 sin cos D .1 M 2 cos2 / 2 (4.34)
@z @x@z @x
and we define, modifying (4.21)
Z 1
O 1 .z; i !/ D 1 .z; x/eix! dx
1
O !/ D 1 .0; i !/:
.i
1 .z; x/ is the solution of (4.35) subject to the boundary condition:

1
@1 .0; x/ D U1 cos .y/; jxj < b;
@z
where y is fixed.
We want
p.x/ D 1 .x/
and
1 .x/ D U1 1 D 0 jxj > b:
We define:
1
A1 .x/ D ; b < x < b
U1
so that
p D 1 U1 A1 :
We now turn to Fourier transforms which are obtained by simply setting in the
previous expressions:
!1 D !I !2 D 0:
Then

1 1
AO1 .i !/ D i ! O !/:
.i
r1 r2
Hence
1 A.iO !/ h .
p
i
O !/ D
.i ! 2 .1 M 2 cos2 1 M 2 !2 :
2 i ! cos
Let
h . i
1 1 p
PO .i !/ D ! 2 1 M 2 cos2 1 M 2 !2 I
2 i ! cos
Z 1
PO .i !/ D P .x/eix! dx; 1 < x < 1
1
and we have:
Z b
P .x
/A.
/d D U1 .y/; jxj < b
b
A.b/ D 0: (4.35)
The Airfoil Equation
We should note that (4.36) is a special case of the Possio integral equation which
we go into in far more detail in Chap. 5, where it is now the airfoil equation or the
finite Hilbert transform equation. PO .i !/ can be expressed as

1 j!j
PO .i !/ D const ; (4.36)
2 i!
where
1 M 2 cos2
const D p sec
1 M2
and

j!j
i!
Fig. 4.1 Pressure

distribution along chord
4
1.0 0.5 0.5 1.0
is the multiplier corresponding to the Hilbert transform. Thus (4.36) becomes

Z b
1 1
const A.s/ds D U .y/; A.b/ D 0; jxj < b: (4.37)
2 b xs
And we have the explicit solution, due to Tricomi (see [11], and the more general
treatment in [31]).
s s
Z
1 1 bx b
bC
1
A.x/ D U1 .y/ d
; jxj < b:
Const bCx b b

x
In as much as
Z s
1 b
bC
1
d
D b;
b b

x
we thus have the explicit solution:

s p
b x 2 1 M2
A.x/ D bU 1 .y/ cos ; jxj < b: (4.38)
b C x 1 M 2 cos2
Note that A.b/ D 0. The function A.:/ is in Lp b; b for 1 < p < 2 but Not for
p D 2. In other words this is Not an L2 theory, as opposed to structure dynamics,
where it is and we have a notion of energy.
See Fig. 4.1 for the canonical static pressure distribution along the chord for
b D 1.
Note the discontinuity with respect to the angle of attack. For D 0, the pressure
distribution at midchord increases without bound as M increases to 1. But for small
nonzero , the pressure peaks at some point and then decreases to zero. We have a
Fig. 4.2 Pressure

distribution at midchord
D 0 and D 1 degree 70
60
50
40
30
0.995 0.996 0.997 0.998 0.999 1.000
transonic peak. Here the word transonic simply means that M is close to 1. See
Fig. 4.2.
Structure Response: Divergence Speed
Because the pressure jump depends only on the pitch angle the structure response
equations reduce to
h p i
GJ 00 .y/ DU1
2
.y/ 2 1 M 2 = 1 M 2 cos2
s
Z b
bx
1 cos .x ab/ dx.0/ D 0 0 .1/ D 0; (4.39)
b b Cx
where the integral:

s r
Z b Z 1
bx 1x
2 .x ab/ dx D 2b 2
.x a/ dx D b 2 .1 C 2a/ ;
b b C x 1 1 C x
where we are assuming that

1
jaj < :
2
The structure equation for the torsion angle .y/,
Z b
GJ 00 .y/ D .x ab/p.x/dx;
b
becomes .y/ D .y/; 0 < y < 1, with .0/ D 0 .0/ D 0 and has the
00 2
solution .y/ D B siny, where B is an arbitrary constant including zero and we

must have
B cos` D 0:
Thus B D 0 is always a solution, whatever U1 .
If we want a nonzero solutionB is not zerowe have an eigenvalue problem
and thus we have a nonzero solution only for

D .2n 1/ ; n1
2`
and
h y i
.y/ D B sin .2n 1/ 0 < y < `;
2`
where:
.
2 D U1
2
1 cos .1 M 2 /1=2 .1 M 2 cos2 /b 2 .1 C 2a/;
which means that U1 cannot be arbitrary and must equal

. p 1=2
U1 D .2n 1/ .1 M 2 cos2 / 1 M2
1 p .
p
. GJ / . 1 .1 C 2a// sec ;
2b1
n positive integer.
The smallest value is called the divergence Speed:
.p 1=2
Ud D 1 M 2 cos2 1 M2
1 p .
p

GJ . 1 .1 C 2a// sec : (4.40)
2b1
It is interesting to see the dependence on M , and the difference between
D0 and 0:
For D 0, we have
1 p .
p

Ud D .1 M 2 /1=4 GJ . 1 .1 C 2a// (4.41)
2b`
and Ud decreases monotonically to zero.
On the other hand for nonzero we have a nonzero minimum and Ud increases
to infinity at M D 1. We have thus what is referred to as the transonic dip which
occurs at
p
M D .1 tan2 /;
our main interest being small =180. See Fig. 4.3.
Ud =Ud .0/:
It is interesting to contrast this with the way in which it is done in CFD

computation; see [99] where it has to be extrapolated from literally a few points
on the graph.
Fig. 4.3 Transonic dip 0.31

D 1o
0.30
0.29
0.28
0.9985 0.9990 0.9995 1.0000
The bending or plunge shape at the divergence speed is determined by
EI h0000 .y/ D L.y/

s
p . Z
bx b
D 1 Ud2 2.y/ 1 M2 .1 M 2 cos2 /
b b Cx
p . h y i
D 2 b 1 M 2 .1 M 2 cos2 / 1 Ud2 sin ; 0 < y < `;
2`
a nonhomogeneous equation with the end conditions:
h.0/ D h0 .0/ D 0 D h00 .1/ D h000 .1/;
which is determined by the solution to the pitch equation.
4.4 The General Nonlinear Case
Getting back now to the general case, let D denote the derivative with respect to .
Then with f ./ given by (4.11) and g./ by (4.12), we have:
X
k1
D k .f ./g.// D f ./D k g./ C g./D k f ./ C Cjk D kj g./D j f ./;
j D1
where Cmn are the binomial coefficients, which at D 0 yields
1 X k X
k1 j
@j m @m
k C cj kj Cmj ;
2 j D1 mD0
@xi @xi
4.4 The General Nonlinear Case 81
where the second term involves j only for j < k.

Next in (4.14), the coefficient of k is given by
X
3 X X
k2 X 3 X
@2 k1 X 1 @1m @m
3 3 1
@2 k @0 @0
c ;
j D1 i D1
@xi @xj @xi @xj i D1 j D1
@xi @xj mD0 m @xi @xj
lD1
(4.42)
where the second term again involves j ; j k 1.

Hence we obtain the nonhomogeneous linear equation for k :
.k / D gk1 ; (4.43)
where
1 XX k X
3 k1 j
@j m @m
gk1 D C kj Cmj
2 i D1 j D1 j mD1
@xi @xi
X
k2 X 3 X 3
@2 k1 X 1 @1m @m
1
C C
i D1 j D1
lD1
X
3 X
3 X
k1
@0 @m @2 km
C2 Cmk
i D1 j D1 mD1
@xi @xi @xi @xj
X
k1
C .1 / j q1 qkj
j D1
and we note that

g0 D 0:
And in particular
X 3 X 3
@0 @1 @2 1
g1 D 2.1 /1 q1 q1 C 4 ;
i D1 j D1
@xi @xi @xi @xj
illustrating the nature of the functions gk . Note that the derivatives are well defined
for any k.
Next we need to go deeper into the boundary conditions. We begin with the typical
section case where the flow is 2D and brings considerable simplification well
justified for high-aspect ratio wings and is the best developed part of the theory.
We specialize to typical section theory, but keep a nonzero angle of attack.
Typical Section Theory: Nonzero
Flow Tangency Condition: Specializing (4.15), (4.16) we have:
@k .x; 0C/ @.0; x; 0C/

D C1k D k1 .y/
@z @x
@k1 .x; 0C/
D k .y/ (4.44)
@x
@k .x; 0/ @.0; x; 0/

D C1k D k1 .y/
@z @x
@k1 .x; 0/
D k .y/: (4.45)
@x
A crucial next step is as follows.
Flow Decomposition
To solve (4.43), we decompose the flow field as the sum of two interdependent
terms: for each k
k .x; z/ D kL .x; z/ C k0 .x; z/; (4.46)
where kL satisfies the homogeneous equation:
.kL / D 0
subject to the boundary conditions (4.44) and (4.45) and k0 satisfies the
nonhomogeneous equation
.k0 / D gk1
with no boundary conditions imposed. In particular
.rk / D .rkL / 1 D 1L : (4.47)

We define
1
X k
0 D 0 C k0 ; (4.48)
k
kD2
1
X k
L D kL (4.49)
k
kD1
so that
D 0 C L ; (4.50)
where 0 has no discontinuities on z D 0 and hence produces no lift, and L
produces all the lift.
Our main concern is of course the pressure jump across the wing, or equivalently
the acceleration potential jump. From (4.7)
2 2 !
@ @
D C D 0 z D 0; jxj > b:
@x @z
By analyticity therefore
k D 0 for k 1 off the wing:
Again our main concern is to evaluate this on the structure. We begin with
Theorem 4.3. Let ' denote the disturbance potential
' D 0 (4.51)
Suppose jr'j2 D 0; jxj < b.

Then

@k @k
k D U1 cos C U1 sin ; b < x < b (4.52)
@x @z
the same as for k D 1.

Proof. We calculate that:
jrj2 D U1
2
C 2r0 r' C jr'j2 : (4.53)
Hence
jrj2 D 2r0 r' C jr'j2 D 2r0 r'

1
D jrj2 D r0 r':
2
Hence

@k @k
k D r0 rk D U1 cos C sin : t
u
@x @z
Note that ' is the disturbance potential and we expect that jr'j2 is small
2
compared to U1 so that we may neglect it in (4.53) anyway. The angle of attack
is usually less than a degree and can be neglected in the first look, simplifying the
analysis. And sensitivity to the angle of attack is analyzed subsequently.
Zero Angle of Attack
Let us now specialize to the case where the angle of attack is zero.
Then we have the following remarkable result.
Theorem 4.4. Suppose D 0. Then
jr'j2 D 0; jxj < b: (4.54)
The speed is continuous across the wing, in other words.

Proof. First we note that
1
X k
r' D rk :
k
kD1
From
1 .x; z/ D 1 .x; z/
we have:
1 .x; 0C/ C 1 .x; 0/ D 0:
Hence
@1 .x; 0C/ @1 .x; 0/
C D 0:
@x @x
Hence
2
@1
D 0:
@x
Also from

@1
D0
@z
we have
2
@1
D 0:
@z
Hence:
jr1 j2 D 0: t
u
To show that this holds for every k, we use an induction argument 5.

Lemma 4.5. Suppose
jrj j2 D 0; j k 1:
Then
jrk j2 D 0:
Proof. First from the flow tangency condition:

@k @k1
D k.y/ jxj < b
@z @x
and hence
2 2
@k @k1
D k 2 .y/2 jxj < b:
@z @x
Hence
2
@k
D0 jxj < b:
@z
Let us use the notation

@j
j D
@x
along with
@j
j for
@z
and the Fourier transforms by Oj ; O j , respectively. We are only concerned with j L .
t
u
Following the proof for j D 1, we take Fourier transforms and obtain
i!
Oj .i !; 0C/ D O j .i !; 0C/; (4.55)
r1
i!
Oj .i !; 0/ D O j .i !; 0/; (4.56)
r1
where
p
r1 D j!j 1 M2 :
Hence, subtracting, we get:

r1
O j .i !; 0C/ C O j .i !; 0/ D O j (4.57)
i!
and adding, we get
r1
.i !/ D .O j .i !; 0C/ O j .i !; 0//; (4.58)
i! j
where we use the notation j D j .x; 0C/ C j .x; 0/; the super bar does Not
denote complex conjugate.
Next we use the boundary condition (4.44) and (4.45)
j .x; 0C/ C j .x; 0/ D j `.y/ j 1 .x/ jxj < b:
By Theorem 4.2,
k .x/ D Uk .x/; jxj < b
and hence
k .x/
Ak .x/ D D k .x/; jxj < b:
U
Now
j .x; 0C/ C j .x; 0/ D j.y/ j 1 .x/; jxj < b:
But
r1
O j .i !; 0C/ C O j .i !; 0/ D Oj .i !/:
i!
Hence
r1 O
Aj .!/ D j.y/O j 1 .i !/: (4.59)
i!
Next
j .x; 0C/ j .x; 0/ D j.y/j 1 .x/; jxj b
and
i!
j .i !/ D .O j .i !; 0C/ O j .i !; 0//;
r1
hence
i! O
Aj 1 .i !/j.y/ D j .i !/;
r1
or
j!j
p AOj 1 .i !/j.y/ D j .i !/: (4.60)
i! 1 M 2
In (4.60) and (4.59) we have a pair of Possio equations for the time-invariant case,
similar to (4.38). This time it is convenient to define the Tricomi transform:
s r
Z
1 bx b
b C s f .s/
T f D hP; h.x/ D ds; jxj < b;
bCx b bs sx
which we discuss in more detail in Chap. 5. Then we have similar to (4.40):

p
1 M 2 Aj D j.y/ T P j 1
j!j
p AOj 2 .i !/.j 1/.y/ D O j 1 .i !/:
i! 1 M 2
Hence
p
Aj 2 .j 1/.y/ D 1 M 2 T P j 1 ;
where we are taking advantage of the fact that
1 r i! p
p D 1 M 2:
1 M 2 i! r
Hence
p
j.y/Aj 2 .j 1/.y/ D j.y/ 1 M 2 T P j 1 D .1 M 2 /Aj :
Thus we have our recursive relation
.1 M 2 /Aj D j.j 1/.y/2 Aj 2 : (4.61)
An immediate conclusion is that
Aj D 0 for j even
and hence
j D 0 for j odd:
And for k odd:

Ak
D .y/k1 .1 M 2 /..k1/=2/ A1
k
and
A2k D 0:
Hence
1
X 1
X
Ak 1
AD D .y/2k2 .1 M 2 /..2k1/=2/ A1 D .y/2
A1 : (4.62)
k 1C
kD1 kD1 1M 2
It is important to note that the series converges for small enough .y/, but the
closed-form solution holds without any smallness assumption on .y/. Also the
convergence is uniform in x and z.
We can now consider the convergence of the power series expansion:
1
X 2kL .x; z/
1
X .
C ..2kC1/L .x; z// .2k C 1/ (4.63)
.2k/
kD0 kD0
For k even, combining (4.56) and (4.57), we have
1 O .k 1/
kL .i !; z/ D ejzjr1 .y/k1 .1 M 2 /..k2/=2/ AO1 .i !/:
k k
Hence for k even:
kL .x; z/ 1
D 1 .x; z/ .k 1/.y/k1 .1 M 2 /..k2/=2/ (4.64)
k k
and we see that the convergence of the first series in (4.57) is uniform in x and z.
For k odd similarly we obtain using (4.61):
kL .x; z/ D 1 .x; z/.y/k1 .1 M 2 /..k1/=2/ :

Again we see that the convergence of the second series in (4.57) is uniform in x and
z. And we have an explicit solution that we can now describe. First of all we see that
for each y it is of the form:
L .x; z/ D N.M; U1 ; .y//1 .x; z/; (4.65)
where
1
X .2k/.y/k1
N.M; U1 ; .y// .1 M 2 /.k1/
.2k/
kD0
1
X 1
C `.y/2k .1 M 2 /k ;
.2k C 1/
kD0
where .y/ is determined by the structure response.

Thus in particular the differentiability properties of L are those of 1 .x; z/ and
there are no discontinuities in the velocity off the wing.
Steady-State Structure Response
Let us next calculate the steady-state structure response. We have for the aerody-
namic moment:
Z b
1
M.y/ D .y/2
U1 .x ab/A1 .x/dx;
1C 1M 2
b
where the integral

1
D .y/ b 2 U1
2
.1 C 2a/ p :
1 M2
Hence
.y/ 1
M.y/ D .y/2
b 2 U1
2
.1 C 2a/ p :
1C 1M 2
1 M2
Let
.1 C 2a/ 1
2 D b 2 p :
GJ 1 M2
Then the pitch equation (4.3) becomes:
.y/
00 .y/ D 2 U1
2
.y/2
0 < y < 1; (4.66)
1C 1M 2
with the end conditions:

.0/ D 0I 0 .1/ D 0:
We have the trivial solution
.y/ D 0I and h.y/ D 0;
which holds for all values of U1 .

To find the nonzero solution we see that we have to solve a nonlinear eigenvalue
problem. Making a change of variable
.y/
.y/ D p ;
1 M2
this becomes
.y/
00 .y/ D 2 U1
2
0<y<1 (4.67)
1 C .y/2
with the end conditions

.0/ D 0 0 .1/ D 0:
We now establish that (4.67) has a solution only for a countable values of U1 .
Multiplying (4.67) by 0 (y) on both sides we obtain:
2 d log.1 C .y/ /
2
d 0
.y/ D 2 U1 :
d d 2
Or
log.1 C .y/2 /
0 .y/ C 2 U1
2
D const D 0 .0/: (4.68)
2
Putting y D 1 we get:
log.1 C .1/2 /
2 U1
2
D 0 .0/; (4.69)
2
which is an homogeneous boundary condition. Now the solution of the first-order
differential equation (4.68) is analytic in the coefficient
2 U1
2
and in (4.69) .1/ and 0 .0/ are entire functions of 2 U1

2
. Hence the zeros of
log.1 C .1/2 /
2 U1
2
0 .0/ D 0
2
are at most countable and cannot have a finite accumulation point. Thus we have
at most a sequence of values of U1 or which (4.67) holds. An approximation is
provided by the divergence speeds.
We could also solve the nonlinear Volterra equation:
Z y
log.1 C .s/2 /
.y/ C 2 U1
2
ds D 0 .0/y; 0 < y < 1;
0 2
which follows from (4.68). But the sequence of speeds does not play a role in
stability analysis, so we dont pursue this any more.
Next the lift

Z b
.y/
L.y/ D U1 A.x/dx D 2 bU 21 p
b 1 M2
and is determined solely by the beam torsion.

Next the plunge equation is Eulerian:
@4 h.y/
EI D 2 bU 21 .y/ 0<y<1 (4.70)
@y 4
and is thus determined completely by the pitch, and we do not continue the
calculation because the solution is of little interest.
The Velocity Potential
To complete the determination of the velocity potential we go on to calculate the

part 0 in (4.50), even though it plays no role in the stability of the structure.
We first need to solve the nonhomogeneous equation (4.43) for D 0:
@2 k @2 k
.1 M 2 / 2
C D gk1 :
@x @z2
Let us begin with k D 2. We have specializing (4.44)
@1
g1 .x; z/ D 2. 1/1 ;
@x
where
Z b p .
@1
D z 1 M 2 ..x
/2 C z2 .1 M 2 //A.
/d
@x b
s
2 bx
A.x/ D U1 `.y/ p ;
1 M2 bCx
which is bounded continuous in R2 , and

Z 1 Z 1
20 .x; z/ D L.x
; z /g1 .
; /d
d;
1 1
where
p

1 z 1 M
2
L.x; z/ D :
x 2 C .1 M 2 /z2
It follows that
1 Z
X 1 Z 1
0 .x; z/ D 0 .x/ C L.x
; z /gk1 d
d: (4.71)
kD2 1 1
We stop here because as before our interest is again only in the structure dynamics
which does not depend on this part of the flow.
Note, however, we have deduced a constructive solution to the Euler potential
equation in the form of a convergent power series.
It is expressible as the sum of two terms neither of which satisfies the equation.
But only one is required for the structure response. Hence there is no point in
calculating the second part as in CFD codes.
A Variation on the Problem
An interesting (and actually useful, as we show in Chap. 7) variation on this problem

is to consider a virtual structure with no dynamics, but we prescribe a normal
velocity on the boundary and seek the solution to the steady state 2D solution for
zero angle of attack aerodynamic field equations. Thus we specify, with the same
notation as before:
@.x; 0/
D w.x/; jxj < b (4.72)
@z
and the KuttaJoukowsky conditions in addition. Proceeding as before we denote
the solution corresponding to w./ by .; x; z/, assumed analytic in with a
nonzero radius of convergence uniformly in x; z.
.0; x; z/ D x U1 ;
@.; x; 0/
D w.x/; jxj < b;
@z
and we define k .x; z/; k .x; z/ as before.

For k D 1, we obtain as before the Fourier transform
ejzjr1
O 1 .i !; z/ D O 1 .i !; 0C/; z>0
r1
ejzjr1
D O 1 .i !; 0/; z < 0: (4.73)
r1
In particular then
O 1 D 2O 1 .i !; 0/:
We have also that the linear velocity potential

@1
1 .x; z/ D U1 .x; z/:
@x
Hence with
1
A.x/ D ;
U1
to satisfy KuttaJoukowski, we need:
A.x/ D 0 jxj > b;

A .b/ D 0:
Hence the Fourier Transform:
O !/ D 2 1 i ! O 1 .i !/;
A.i
r1
or
1 O j!j p
O 1 .i !/ D A.i!/ 1 M 2:
2 i!
Hence using the Tricomi operator we have:
s
Z s
2 1 bx b bC
A.x/ D p w.
/d
; jxj < b; (4.74)
1 M 2 b C x b b
where to satisfy the Kutta condition we assume that w ./ satisfies a Lipschitz
condition of order ; 12 < 1.
The corresponding solution for the velocity potential is then given by:
jzjr1
e
O1 .i !; z/ D O !/
A.i z>0
2i !
ejzjr1 O
DC A.i !/ z < 0; (4.75)
2i !
where
p Z 1 p .
jzjr1 j!jjzj 1M 2
e De D ei!x 2jzj 1 M 2 .z2 .1 M 2 / C x 2 /dx;
1
z 0:
Hence
p Z b
@1 .x; z/
D z 1 M2 .A .
/d
/=z2 .1 M 2 / C .x
/2 ; jxj < 1:
@x b
(4.76)
Next
@O1 .i !; z/ ejzjr1 O 1 j!j p O !/:

D r1 A.i !/ D ejzjr1 1 M 2 A.i
@z 2i ! 2 i!
Hence taking the inverse Fourier transform,

p Z 1
@1 .x; z/ 1 M2
Dz .B.
//=.z2 .1 M 2 / C .x
/2 /d
; (4.77)
@z 2 1
where
Z b
1 A.
/
B D HA./I B.x/ D d
; jxj < 1;
b x
H being the Hilbert transform. This is an Lp Lq transform, 1 < p < 2, A./ being
in Lp ; 1 p < 2.
A canonical example is:
1
w.x/ D p ; jxj < b; in which case; (4.78)
bCx
s s
Z
2 bx b
1 1
A.x/ D p d
; jxj < b: (4.79)
1 M2 bCx b b

x
The integral does not have a closed form and has to be evaluated numerically.
Fig. 4.4 Pressure

distribution along chord
10
1.0 0.5 0.5
Figure 4.4 shows the plot for b D 1 (within a gain factor). For k 2, abbreviating
the notation for this section k D k0 satisfies the nonhomogeneous equation
@2 k @2 k
.1 M 2 / 2
C D gk1 : (4.80)
@x @z2
For k D 2,
@1
g1 D U1 . 1/1 (4.81)
@x
and is a bounded continuous function in R2 . So in fact is gk , for every k.

The nonhomogeneous equation (4.80) is readily solved by taking Fourier trans-
forms. We have
Z 1Z 1
k .x; z/ D L.x
; z /gk1 .
; /d
d; x; z in R2 ; (4.82)
1 1
where
1 p .
L.
; / D 2jj 1 M 2 .2 .1 M 2 / C
2 /: (4.83)
2
Hence it follows that k ./ is a bounded continuous function in R2 , with continuous

derivative as well, no discontinuities in flow velocity. Hence the series:
1
X 1
0 .x; z/ D 0 .x; z/ C k;0 .x; z/ (4.84)
k
kD2
converges pointwise and has a continuous derivative. Hence so does
D 0 C L :
The flow thus has no discontinuities. And hence there are no shocks in the flow.
The BeranStraganac Model
However complex the nonlinearity we do have the zero or rest solution as the static
solution. In turn the linearization gives us back the Goland equation.
The DowellHodges Model
Here the static aeroelastic system of equations is:

Z b
EI 1 h0000 .y/ C .EI 2 EI 1 /..y/v.y/00 /00 D mg cos ' .x; y/dx; (4.85)
b
EI 2 v0000 .y/ C .EI 2 EI 1 /..y/h.y/00 /00 D mg sin ' (4.86)

Z b
GJ 00 .y/ C .EI 2 EI1 /v.y/00 h00 .y/ D .x ab/ .x; y/dx; (4.87)
b
plus CF end conditions.

Field Equation
" 2 2 !#
. 1/ @ @
0D 1C
2
a1 2
U1
2
./
2a1 @x @z

1 @ @ @ @
jrj C
2
jrj 2
(4.88)
2 @x @x @z @z
jrj2
D :
2
FluidStructure Boundary Conditions
@.x; 0/ @.x; 0/
D .y/:
@z @x
Plus KuttaJoukowsky Conditions
The gravity terms on the rightmg cos '; mg sin 'contribute to the
complexity.
If we set g in the gravity term to be zero, we see that we have the zero solution
for the structure and constant flow for the aerodynamic equation is a static solution
for all U1 . In particular we need to set g to be zero for determining the ground
modes corresponding to U D 0, and we see that we go back to the linear Goland
model.
If the difference term
EI 2 EI 1
is set to zero, then of course, the equations become linear, and we go back to the
solution for the Goland model we started with, except for the addition of the v./
equation:
EI 2 v0000 .y/ D 0;
which stands alone, and has only the zero solution.

Let us examine then the dependence on
.EI 2 EI 1 /:
Let us consider the solution for
.EI 2 EI 1 /;
where is a small parameter so that we have the solution:
h.; y/ v.; y/ and .; y/
satisfying:
EI 1 h0000 .; y/ C .EI 2 EI 1 /..; y/v.; y/00/00

.; y/
D mg cos' 2 bU 21 p
1 M2
EI 2 v0000 .; y/ C .EI 2 EI 1 /..; y/h.; y/00 /00 D mg sin '
.; y/
GJ 00 .; y/ C .EI 2 EI 1 /v.; y/00 h00 .; y/ D ;
.; y/2
1C
1 M2
where
1
D b 2 U1
2
.1 C 2a/ p :
1 M2
Let
dk

hk .y/ D h.; y/ D 0
dk
and the power series expansion:

1
X hk .y/k
h.; y/ D
k
kD0
with a similar notation for the others.

We can calculate that
.0; y/
EI 1 h0000 00 00
1 .y/ C .EI 2 EI 1 /..0; y/v.0; y/ / D 2 bU 1 p
2
1 M2
EI 2 v0000 00 00
1 .y/ C .EI 2 EI 1 /..0; y/h.0; y/ / D 0
GJ 100 .y/ C .EI 2 EI 1 /v.0; y/00 h00 .0; y/ D 1 .y/;
where we have used the fact that .0/ D 0.

More generally we have for k 2:
GJ k00 .y/ C k .y/ D r; k1 .y/
EI 1 h0000
k .y/ D rh; k1 .y/
EI 2 v0000
k .; y/ D rv; k1 .y/
with zero end conditions, because the derivatives with respect to are zero.
The functions
;k1 .y/ h;k1 .y/ v;k1 .y/
depend only on the functions with index .k 1/.

Theorem 4.6. The pitch equation
GJ k00 .y/ C `k .y/ D ;k1 .y/ 0<y<`
with end conditions
CF W k .0/ D 0I k0 .`/ D 0

FF W k0 .0/ D 0I k0 .`/ D 0
has the solution given by:

Z `
k .y/ D G .y; s/;k1 .s/ds; 0 < y < `;
0
except for a sequence of divergence speeds Un for which
GJ k00 .y/ C n k .y/ D 0; 0<y<`
has a nonzero solution.

Proof. r

Let ! D :
GJ
Then
Z ! Z
1 ` s
sin !.s /
k0 .s/ D cos !.` /;k1 ./d C `;k1 ./d
cos !` 0 0 !
excepting ! such that cos !` D 0, which is the sequence Un . t

u
The solution procedure is similar for the other two equations.
Finally then we have the series expansion for the solution which we can express
in the form
X1
xQ k .y/
x.y/ D .EI 2 EI 1 /k ; 0 < y < `; (4.89)
k
kD0
where
xk .y/
xQ k .y/ D ;
.EI 2 EI 1 /k
0 1
h .0; y/
x.y/ D @ v .0; y/ A
.0; y/
is then the static solution to the problem (CF end conditions)
Z `
xk .y/ D G.y; s/xk1 .s/ds; 0 < y < `;
0
0 1
Gh .y; s/
G.y; s/ D @ Gv .y; s/ A :
G .y; s/
Let us calculate x1 .:/. We have
EI 1 h1 .y/0000 D 0;
EI 2 v0000
1 .y/ D 0;
GJ 100 .y/ C 1 .y/ D .EI 2 EI 1 /v.0; y/00 h00 .0; y/

2 !
2 2 y2
D .EI 2 EI l / m g cos ' sin ' 2`y C .6` 3`2 /
4
D .y/
and
Z ! Z
1 ` s
sin !.s /
10 .y/ D k0 .s/ D cos !.` /./d C ./d:
cos !` 0 0 !
And
h1 .y/ D h.0; y/I v1 .y/ D v.0; y/;
which gives some idea about the nature of the solution.
Linearization
Having evaluated the static solution, we now go on to linearize the solution to the
aeroelastic equations about this solution for U nonzero so that g has to be nonzero.
We anticipate that it is more complex than for the Goland model.
Thus let
h.; t; y/ D h.0; y/ C h1 .t; y/;

v.; t; y/ D v.0; y/ C v1 .t; y/;
.; t; y/ D .0; y/ C 1 .t; y/;
.; t; x; z/ D .0; x; z/ C 1 .t; x; z/;
@.; t; x; 0/ @.0; x; 0C/ @1 .t; x; 0C/
D C ;
@z @z @z
where
0 1
h .0; y/
@ v .0; y/ A D x.y/;
.0; y/
L.; t; y/ D L.y/ C L1 .t; y/;

M.; t; y/ D M.y/ C M1 .t; y/;
R t; y/ C EI1 h000 .; t; y/ C .EI 2 EI1 /..; t; y/v.; t; y/00 /00

mh.;
D m g sin ' C L.; t; y/; (4.90)
mRv.; t; y/ C EI 2 v0000 .; t; y/ C .EI 2 EI 1 /..; t; y/h.; t; y/00 /00 D mg cos '
(4.91)
R
I .; t; y/ GJ `00 .; t; y/ C .EI 2 EI 1 /h.; t; y/00 v.; t; y/00
D M.; t; y/; 0 < tI 0 < y < `: (4.92)
Taking derivative with respect to at zero, we have:
mhR 1 .t; y/ C EI 1 h0000

1 .t; y/ C .EI 2 EI 1 /.1 .t; y/v.0; y/
00
C .0; y/v1 .t; y/00 /00 D L1 .t; y/; (4.93)
mRv1 .t; y/ C EI 1 v0000

1 .t; y/ C .EI 2 EI 1 /.1 .t; y/h.0; y/
00
C .0; y/h1 .t; y/00 /00 D 0; (4.94)
I R1 .t; y/ GJ 100 .t; y/ C .EI 2 EI 1 /.h1 .t; y/00 v.0; y/00
C h.0; y/00 v1 .t; y// D M1 .t; y/; (4.95)
where the linearized lift and moment

d
L1 .t; y/ D L.0; t; y/;
d
d
M1 .t; y/ D M.0; t; y/:
d
This is the linearized equation that will determine system stability and is now more
complex than the Goland case because the coefficients in the equation are no longer
constants. It also then requires an existence and uniqueness proof. We consider this
as part of the convolution/evolution equation in Chap. 5.
Notes and Comments
The crucial result used here is the explicit solution to the finite Hilbert transform
given by Tricomi [11] who notes also the work of Sohngen, the German pioneer in
aeroelasticity theory, and shows the Kutta condition as necessary for uniqueness of
solution. There is also a large body of related Russian work (mainly that of Gohberg;
see [31]) which seems to ignore the Tricomi work.
Note that the aerodynamics is replaced by the Hilbert transform, which we look
upon as the static case of the Possio integral equation. A feature of our entire theory
is the systematic use of the Possio equation. As here and more generally, the best-
developed results are for the typical section case where the air flow is 2D and the
Possio equation is 1D. The angle of attack becomes important in that the functionals
turn out to be discontinuous at zero for large M . The point here is that for large
enough M there may be regions in the interior for which the local flow is supersonic,
in which case the flow is termed transonic. The precise value of M for which this
happens is not clear. Meyer [14, p 146] gives an inequality for the interior Mach
number. One author [48] suggests M 0:8 as transonic.
We have in this chapter a simple instance of nonunique solutions for the static
solution for a sequence of values of the far field speed determined by a nonlinear
eigenvalue problem. We could argue that we wont see this nonuniqueness in time
marching because we will not have the exact values for the sequence. In any case
it does raise a question for CFD solutions. Our main result concerning the air flow
would appear to be new: that the potential can be expressed as the sum of two,
one which produces lift and the other that does not, consistent with our theme that
we are concerned only with the structure dynamics whereas the second part holds
little interest for us.
Chapter 5
Linear Aeroelasticity Theory/
The Possio Integral Equation
5.1 Introduction
In this chapter we present the linear theory which plays an essential role in the
nonlinear Hopf bifurcation stability theory to come in Chap. 6. This is the most
studied part of the theory beginning with the pioneers: Kussner, Sohngen, Wagner,
Garrick, and Theodorsen; see the classic treatise [6]. This is also the longest chapter
in the book and covers the theory buttressing a wide variety of topics loosely referred
to as Flutter Analysis: the every day bread-and-butter part of the subject.
5.2 Power Series Expansion of the 3D Potential
We begin with a power series expansion of the solution of (3.15) with the associated
boundary dynamics about the rest or equilibrium aeroelastic state solution (Chap. 4):
0 .t; x; y; z/ D U1 .x Cos C z Sin /;

h.t; y/ D 0I .t; y/ D 0:
We call this an Attractor and note that it is unique in that it holds for every value
of U1 unlike the Divergence speed (Chap. 4).
To determine the Linear potential field equation, we consider the
structure variables as parameters and expand the solution in a power series
about zero. Thus with denoting a complex variable we consider the solution
.; t; x; y; z/ corresponding to h.t; y/; .t; y/:
1
X k
.; t; x; y; z/ D 0 .t; x; y; z/ C k .t; x; y; z/; (5.1)
k
kD1
which satisfies the field equation (3.15).

104 5 Linear Aeroelasticity Theory/ The Possio Integral Equation
Leaving for later the sense in which the convergence is taken, as in the steady
state case in Chap. 4, we first calculate the kth order potentials k given by

k .t; x; y; z/ D @k .0; t; x; y; z/ =@k :
Differentiating (3.15) with respect to we get the field equation for each k:
.k / D gk1 ; (5.2)
a linear nonhomogeneous equation, where

@2 @2 @2
./ D C 2U 1 Cos C Sin ff
@t 2 @t@x @t@z

@2 @2
a1 2
1 M 2 Cos2 2
C 1 M 2 Sin2
@x @z2

@2 @2
C 2M 2
Cos Sin (5.3)
@y 2 @x@z
and gk involves the potentials up to the kth order, and g0 D 0. We begin with
1 the Linear potential. The boundary dynamics are determined from (3.19): Flow
Tangency:
"
@0 P y/ C .x a/.t;
P y/
.@.; t; x; y; 0//=@z D C .1/ h.t;
@z
C ...@.; t; x; y; 0//=@x/.t; y//

C ..@.; t; x; y; 0//=@y/
#
@
.h.t; y/ C .x a/.t; y// ; x; y 2 :
@y
(5.4)
Hence taking the derivative with respect to at D 0, we obtain:

h
P y/ C .x a/.t;
.@1 .t; x; y; 0//=@z D .1/ h.t; P y/
i
C U1 Cos .t; y/ ; x; y 2 : (5.5)
Let us denote the right-hand side by wa .t; x; y/this is referred to as the

downwash in the aeroelasticity parlance.
In particular we see that the normal derivative of the linear potential is continuous
across the wing. Also no spanwise derivatives are involved.
5.2 Power Series Expansion of the 3D Potential 105
The pressure jump
p.; t; x; y/ D 1 .; t; x; y/;
where
2
1

.; t; x; y; z/ D .@.; t; x; y; z//=@t C r.; t; x; y; z/ : (5.6)
2
Hence in the power series expansion for .; t; x; y; z/:

1
X k
1 2
.; t; x; y; z/ D U1 C k .t; x; y; z/; (5.7)
2 k
kD1
we need to calculate
k
@ .0; t; x; y; z/ =@k D k .t; x; y; z/:
For k D 1, this yields
1 .x; y; z/ D .@1 .t; x; y; z//=@t C U1 Cos .@1 .t; x; y; z/=@x/

C U1 Sin .@1 .t; x; y; z/=@z/:
Hence defining (Kussner Doublet Function)
A1 .t; x; y/ D 1 .t; x; y/=U1 ;
we have:
@ @
A1 .t; x; y/ D 1 .t; x; y/ C U1 Cos 1 .t; x; y/: (5.8)
@t @x
And the KuttaJoukowsky conditions are:
A1 .t; x; y/ D 0 for x; y NOT in ; (5.9)
A1 .t; b; y/ D 0; 0 < y < `; (5.10)
which are then the boundary conditions for the linear equation (5.3), along with
(5.5). Thus we can state the
Linear Aeroelasticity Problem: 3D Flow/Finite Plane
Field Equation

@2 @2 @2 @2
C 2U 1 Cos C Sin a1
2
1 M 2
Cos2

@t 2 @t@x @t@z @x 2

@2 @2 @2
C 1 M 2 Sin2 C 2M 2
Cos Sin D 0:
@z2 @y 2 @x@z
Boundary Conditions: Far field
.t; 1/ D U1 .x Cos C z Sin /:
Flow Tangency:
.@1 .t; x; y; 0//=@z

h i
P y/ C .x a/.t;
D .1/ h.t; P y/ C U1 Cosff .t; y/ ; x; y 2 :
KuttaJoukowsky:
.x; y; z/ D .@1 .t; x; y; z//=@t C U1 Cos .@1 .t; x; y; z//=@x;
C U1 Sin .@1 .t; x; y; z//=@z;
A.t; x; y/ D .t; x; y/=U1 ;
A.t; x; y/ D 0 for x; y NOT in ;

A.t; b; y/ D 0; 0 < y < `: (5.11)
Our technique of solution is to convert it to an integral equation: The Possio integral

equation. This is not too surprising because basically we have a Neumann-type
boundary value problem which is usually treated this way (cf [1]).
5.3 The Linear Possio Integral Equation
See [55, 56, 72, 96]. The Possio Integral equation plays a central role in our theory
and we shall revisit it in many places in many forms, in both Laplace Domain and
Time Domain.
5.3 The Linear Possio Integral Equation 107
The Input Output Relation
We may look at what we need from the air flow model as an inputoutput problem
where the downwash is the Input and the pressure jump is the Output. The
InputOutput Relation is then the Possio Integral Equation.
The Connecting Glue: Lagrangian to Eulerian
We may also consider it as providing the link between Lagrangian Dynamics and
the Fluid Eulerian Dynamics. This gluing is usually the most mysterious part in
computational aeroelasticity.
We take the Laplace Transform of (5.3) in the time domain and the Lp Lq
Fourier Transforms, 1 < p < 2, in the spatial x; y coordinates. We use super to
denote Laplace Transform and super ^ to denote the Fourier Transform:
Z 1 Z 1 Z 1
QO
.; !1 ; !2 ; z/ D et eix!1 eiy!2 .t; x; y; z/dx dy dt
1 1 0
Re > a 0I !1 ; !2 " R2
with a similar definition for other functions, such as wa . : /; . : /.

It is convenient from now on to use the reduced frequency
b
kD
U1

D for b D 1:
U1
Then taking the Laplace Transform of the field equation (5.3), we obtain:
Z 1
O @
.; x; y; z/ D et .t; x; y; z/dt I .0; x; y; z/ D 0I .0; x; y; z/ D 0
0 @t
..t; x; y; z/ may be called the disturbance potential) we have:

! "
@O @O @2 O
0 D O C 2U1
2
Cos C Sin
1 M 2 Cos2 2
a1
@x @z @x 2
#
@2 O @2 O @2 O
C .1 M Sin / 2 C 2 2M Cos Sin
2 2 2
: (5.12)
@z @y @x@z
QO
Using '.; !1 ; !2 ; z/ in place of .; !1 ; !2 ; z/; we have, taking Fourier
2
Transforms in (5.12) and dividing by a1 :
@2 '
@'
1 M 2 Sin2 2 M 2 i !1 Cos Sin C M 2 k Sin
@z2 @z
2 2
M k C 2M 2 k Cos i !1 C !12 .1 M 2 Cos2 / C !22 ' D 0: (5.13)
We note that
' D c erz ;
where c is not a function of z, is a solution of this equation provided r satisfies the
algebraic equation
ar 2 2br c D 0;
where
a D .1 M 2 Sin2 /;
b D M 2 .k C i !1 Cos /Sin ;
c D k 2 M 2 C 2M 2i !1 Cos C !12 .1 M 2 Cos2 / C !22 :
The equation has exactly two roots:

1 2
r1 D M .k C i !1 Cos /Sin
.1 M Sin /
2 2

p 2 2
k M C 2kM 2 i !1 Cos C !12 .1 M 2 / C !22 .1 M 2 Sin2 / ;
(5.14)

1 2
r2 D M .k C i !1 Cos ff/Sin
1 M Sin
2 2

p 2 2
C k M C 2kM 2 i !1 Cos C !12 .1 M 2 / C !22 .1 M 2 Sin2 / :
(5.15)
Before we proceed further we need to clarify the definition of the square root in
(5.14) and (5.15).
Lemma 5.1.
k 2 M 2 C 2kM 2 i !1 Cos C !12 .1 M 2 / C !22 .1 M 2 Sin2 /
is never equal to a negative number for any complex number k which is not pure
imaginary.
Proof. If M D 0 or k D 0
k 2 M 2 C 2kM 2 i !1 Cos C !12 .1 M 2 / C !22 .1 M 2 Sin2 /
D !12 .1 M 2 / C !22 .1 M 2 Sin2 / 0:
Next let us consider M 0. Let k D C i . We can calculate that the Im part D

2M 2 . C !1 Cos / and the real part D . 2 2 /M 2 2M 2 !1 Cos C !12 .1
M 2 / C !22 .1 M 2 Sin2 /:
We need to consider the case
. C !1 Cos / D 0;
and is not zero. But if . C !1 Cos / D 0, then the real part is
2 M 2 CM 2 !12 Cos2 C!12 .1M 2 /C!22 .1M 2 Sin2 / 0: t

u
Hence the usual principal value of the square root will be taken throughout.
As noted in Chap. 4, we can again show that the only solution that vanishes as jzj
goes to zero is given by:
QO
.; i !1 ; i !2 ; z/ D AC er1 ; z > 0;

O r1 z
D e ; z > 0;
r1
D A er2 z ; z < 0;

O r2 z
D e ; z < 0;
r2
where we define
@QO

O D z D 0;
@z
which as we have noted is continuous at z D 0.
Now by (5.8)

QO i ! ; i ! / D . C U i ! Cos /
O .; i ! ; i ! /
A.;
1

1
: (5.16)
1 2 1 1 1 2
r1 r2
Hence

O

.; QO i ! ; i ! /
i !1 ; i !2 / D A.;
r1 r2 1
; (5.17)
1 2
r2 r1 . C U1 i !1 Cos /
where

r1 r2 1 1
D k 2 M 2 C 2kM 2 i !1 Cos
r2 r1 2 . C U1 i !1 Cos /
. p 2 2
C!12 .1 M 2 Cos2 / C !22 k M C 2kM 2 i !1 Cos

C!12 .1 M 2 / C .1 M 2 Sin2 /!22 : (5.18)
Let PO .; x; y/ denote the inverse Fourier Transform of
1 1
D k 2 M 2 C 2kM 2 i !1 Cos
2 . C U1 i !1 Cos /
. p
C!12 .1 M 2 Cos2 / C !22 k 2 M 2 C 2kM 2 i !1 Cos

C!12 .1 M 2 / C .1 M 2 Sin2 /!22 :
Then taking inverse Fourier Transforms in (5.18) we obtain

Z b Z `
O

.; x; y/ D PO .; x ; y /A.;
O ; /dd; 1 < x; y < 1
b 0
with the Kutta condition
O ; / ! 0 as ! b;
A.;
where the kernel is specified in terms of its Lp Lq spatial double Fourier Transform
Z 1 Z 1
1 1
PO .; x; y/eix!1 iy!2 dxdy D
1 1 2 . C U1 i !1 Cos /
.
p 2 2
k 2 M 2 C 2kM 2 i !1 Cos C !12 1 M 2 Cos2 C !22 k M

C 2kM 2 i !1 Cos C !12 .1 M 2 / C .1 M 2 Sin2 /!22 ; (5.19)
which we denote by PQO .; i !1 ; i !2 /.

In particular, specializing to , and this is Possios simple yet profound idea, we
get the integral equation named after him:
The Linear Possio Integral Equation
Z b Z `
O

.; x; y/ D PO .; x ; y /A.;
O ; /d d; x; y ; (5.20)
b 0
where
O ; / ! 0 as ! b;
A.;
and where
Z 1Z 1
1 1
PO .; x; y/eix!1 iy!2 dx dy D
1 1 2 C U1 i !1 Cos ff
.
p 2 2
k 2 M 2 C 2kM 2 i !1 Cos C !12 .1 M 2 Cos2 / C !22 k M

C 2kM 2 i !1 Cos C !12 .1 M 2 / C .1 M 2 Sin2 /!22 :
O :; :/.
Or, in other words this is enough to determine A.;
In particular we note that if the Linear Aeroelastic problem has a solution, then
the Laplace Transform of the Kussner Doublet Function defined in (5.8) will satisfy
the Possio Equation (5.20).
And conversely, if the Possio equation has a solution, then the Laplace/
Fourier Transform of the solution of the potential equation (5.13) is given by:
QO
.; i !1 ; i !2 ; z/ D PQO .; i !1 ; i !2 /A.;
QO i ! ; i ! / 1 er1 z
1 2 z > 0; (5.21)
r1
Q QO 1
PO .; i !1 ; i !2 /A.; i !1 ; i !2 / er2 z z < 0: (5.22)
r2
Of course we need next to determine the solution as a function of the time/space

coordinates that we started with. The Laplace/Fourier Transformation has to be in-
verted. But the Transforms by themselvesthe Laplace Transforms in particular
are of independent interest, specifically in studying system stability assuming the
Transform is invertible!that it is the Transform of a time domain function which
satisfies (5.3).
Existence and Uniqueness of Solution
The existence and uniqueness of the Possio equation (5.20) is of primary concern to
us and we shall need to spread this over many stages.
First let us note that if the LDPLaplace Domain Possio equation (5.20)has two
solutions and thus also two time-domain solutions of (5.3), then the difference will
satisfy the equation with
wa .t; x; y/ D 0:
But then the difference will also yield a nonzero solution of the potential flow
equation with zero pressure jump across the wing. Hence existence and uniqueness
of solution of the Linear Euler equation with the KuttaJoukowsky boundary
condition is equivalent to the existence and uniqueness of solution to the LDP
Equation with the extra condition that the solution is a Laplace Transform:
Z 1
O x; y/ D
A.; et A.t; x; y/dt Re: > a :
0
The Typical Section Case
Let us specialize now to the Typical Section or Airfoil case, where we set the y-
coordinate to zero simplifying to 2D aerodynamicswhich is equivalent to setting
!2 to zero in the Fourier Transforms. Thus the 1D Possio Equation becomes:
Z b
O a .; x/ D
w PO .; x /A.;
O /d; jxj < b; (5.23)
b
with the Kutta condition:

O b/ D 0;
A.; (5.24)
O x/ goes to a finite limit
which may be weakened to A.;
as x goes to b; (5.25)
where the Lp Lq Fourier Transform of the kernel is given by:

2 2
Q 1 1 M k C 2kM 2 i !Cos C ! 2 .1 M 2 Cos /
O
P .; i !/ D ;
2 .k C i !Cos / p .M 2 k 2 C 2kM 2 i !Cos C ! 2 .1 M 2 //
1 < ! < 1: (5.26)
And an immediate observation from (5.23) is that if A.; O :/ is the solution corre-
O
sponding to w.; O O
:/, then f ./A.; :/ is the solution corresponding to f ./w.; :/.
In particular this implies that the solution has as many time derivatives as wa .t; :/
has.
Typical Section/Zero Angle of Attack
If we set in addition
D0
the kernel P becomes
1 1 p 2 2
PQO .; i !/ D M k C 2kM 2 i ! C ! 2 .1 M 2 / : (5.27)
2 k C i!
The next result is crucial and is the point of departure from the extant aeroelasticity
literature.
Balakrishnan Formula
Theorem 5.2. (Balakrishnan 2003)

Let D 0.
Then in (5.27) PNO .; !/ can be expressed as
1 j!j p
PNO .; !/ D NO i !/ ;
1 M 2 1 C B.k; (5.28)
2 i!
NO i !/ is the Fourier Transform
where B.k;
Z 1
NO i !/ D
B.k; O x/ei !x dx;
B.k; 1 < ! < 1;
1
where
kx Z 1
O x/ D k p e
B.k; k ekxs a.M; s/ds x > 0;
1 M2 0
Z 2
k ekxs a.M; s/ds; x < 0;
0
where
M M
1 D I 2 D I
1CM 1M
1 p
a.M; s/ D ..1 s/.2 C s// =.1 s/; 2 < s < 1 ;

O :/ is in Lp .R1 / for p > 1.

and B.k;
O :/ is in Lp .R1 / for p > 1.

Proof. First let us prove that B.k;
We note that ja.M; s/j is bounded in 2 s 1 for each M . Let c.M /
denote this bound. Since we have:
O x/ D k B.1;
B.k; O kx/ (5.29)
it is enough to consider k D 1. Then

Z 1 1 ex1

exs a.M; s/ds c.M / ; x>0
0 x
Z 2 .

exs a.M; s/ds c.M / 1 ejxj2 .jxj/ x<0
0
and Z 1
1 ex1 P
dx < 1 for p > 1:
0 x
The representation (5.28) was discovered by the author [56] by contour
integration (see [71]) but proved by direct integration, which we follow here.
First let us note that by (5.33) we may take
k D 1:
Let
1 < Re: z < 2 z 0:
Then
Z 1 Z
O x/dx D p 1 1 1 1 1 p
ezx B.1; ..1 s/
1 1 M2 1 C z 0 s C z
Z
1 2 1 p
.2 C s/// .1 s/ds C ..2 s/
0 sz

.1 C s/// .1 C s/ds;
which by changing variables in the integrals (see [4] for details) can be
expressed as:
z p 1
D .z C 1 /.z 2 / 1
zC1 z2
and for
z D i!
this is
j!j 1 p
D ..i ! C 1 /.2 i !// 1:
i! 1 C i!
Hence
NO i !/ D j!j 1 p..ki ! C /. ki !// 1

B.k; 1 2
i! k C i!
j!j 1 1 p 2 2
D p .M k C 2kM 2 i ! C ! 2 .1 M 2 // 1
i! k C i! 1 M 2
showing that(5.28) =(5.29) as required.

Finally let us note that for Typical Section and zero angle of attack, (5.22) and (5.23)
specialize to
QO
.; i !; z/ D
1 QO i !/erz z > 0
A.;
k C i!
DC
1 QO i !/erz z < 0;
A.;
k C i!
where
p
r D .k 2 M 2 C 2kM 2 i ! C ! 2 .1 M 2 //
so that, in particular
.t; x; z/ D .t; x; z/;

.@.t; x; z//=@z D .@.t; x; z//=@z;
.@.t; x; z//=@x D .@.t; x; z//=@z:
Next we specialize to the most studied case M D 0 See [6].

The case M D 0 corresponds to Incompressible Flow for reasons explained
below. The importance for us is that we can obtain an explicit solution for the Possio
equation and a systematic use of which yields an alternate theory as compared with
the classical theory in [6]. M D 0 is an outlier in that it is atypical in comparison
to the solution for nonzero M. t
u
Incompressible Flow
First we have:
Theorem 5.3. For each and M; 0 M 1, (5.27) defines an Lp Lq
Mikhlin multiplier on Lp .R1 / and hence a linear bounded operator on Lp .R1 /
into itself. Let .; M; k/ denote the operator. It is convenient to regard Lp .b; b/
as a subspace of Lp .b; b/ Let P denote the projection operator on Lp .R1 / into
Lp .b; b/. Then the Possio Equation can be expressed in operator form as an
equation in Lp .b; b/, 1 p < 2:
O :/
wO a .; :/ D P.; M; k/A.; (5.30)
with the Kutta condition

O b/ D 0:
A.;
Proof. By definition (see [28, 41, 56]), a scalar valued function ./ de-
fined on R1 is called a Mikhlin multiplier if g.!/ Q D .!/fQ.!/;
1 < ! < 1; excepting ! D 0 is the Fourier Transform of a function in
Lp .R1 / whenever fQ. : / is. A sufficient condition for this due to Mikhlin [1] is that
. : / be continuously differentiable in .1; 1/ and
j.!/j C j!0 .!/j c < 1; omitting ! D 0:
This condition is readily verified for (5.27). The statement of the theorem follows
from Mikhlins theory [28] who shows that if T denotes the corresponding operator,
then T is linear bounded with norm:
jjT jj cMp where Mp depends only on p:
In our structure dynamics, the downwash is actually C1 in b; b.

That the space has to be Lp , 1 < p < 2 can be seen by considering the
special case where k D 0, which corresponds to the steady state case treated
in Chap. 4. Let us look at some special cases where an explicit solution can be
constructed. t
u
Note that for zero angle of attack, the kernel PQO in (5.27) becomes:
1p 2 2
.M k C 2kM 2 i ! C ! 2 .1 M 2 //=.k C i !/: (5.31)
2
Recall that Re k > 0 in what follows.

Theorem 5.4. Let D 0 and M D 0 (Incompressible Case). Then the Possio
Equation in this case:
O :/
wO a .; :/ D P.0; 0; k/A.; (5.32)
has a unique solution in Lp which we can express explicitly. Here we follow [4].
Proof. Let .0; 0; k/ be simplified to .k/. For D 0 D M the multiplier is
1 j!j
;
2 k C i!
which can be expressed as:

1 k j!j
1 :
2 k C i! i!
Let R.k/ denote the operator corresponding to the multiplier
1
:
k C i!
Then R.k/ is given by:

Z x
R.k/f D gI g.x/ D ek.xs/ f .s/ds; 1 < x < 1: (5.33)
1
j!j
Next the multiplier i! is recognized as representing the Hilbert Transform, denoted
H, given by
Z 1
1 f ./
Hf D gI g.x/ D d; 1 < x < 1: (5.34)
1 x
And we note that:

HR.k/ D R.k/H:
Then P.k/ can be expressed:
1
P.k/Pf D P.I kR.k//HPf for f in Lp .R1 /:
2
Hence (5.33) becomes:
O :/ kPR.k/HP A.;
O a .; :/ D PHP A.;
2w O :/
O :/ kPHR.k/P A.;
D PHP A.; O :/
O :/ kPH.P C I P/R.k/P A.;

D PHP A.; O :/
O :/ kPH.I P/R.k/P A.;

D PHP.I kR.k/P/A.; O :/: (5.35)
Let us evaluate the function
O :/:
PH.I P/R.k/P A.;
Denoting it by q. : /, we have:
Z 1 Z b
1 1 O :/dsd;
q.x/ D ek.s/ A.; jxj < b
b x b
Z 1
1 1 O :/ ;
D ek d L k; A.;
0 xb
where Z b
L.k; f / D ek.bs/ f .s/ds:
b
Hence we have
O
2w.; O :/ kq:
:/ D PHP.I kR.k/P/A.;
Hence
O
2T w.; O :/ kT q;
:/ D .I kL.k//A.; (5.36)
where L.k/ for each k is a Volterra operator on Lp .b; b/ into itself, defined by:
Z x
L.k/f D gI g.x/ D ek.xs/ f .s/ds; jxj < b
b
and kT q is the function

O ://h;
kL.k; A.;
where
s " r #
Z Z 1
1 bx 1 b
bCs 1 1 k
h.x/ D ds e d ;
bCx b bs sb sx 0
where we can calculate that the integral

Z r
1 b
bCs p 2b C 1
1=..s b /.s x//ds D : (5.37)
b bs xb
Here T is recognized as the Tricomi Operator which we treat more fully in the
sequel.
Hence
s
Z
1 b x 1 k p 2b C 1
h.x/ D e d jxj < b: (5.38)
bCx 0 xb
(Note that . : / is NOT defined for k negative. An important point for us in the
sequel, in Sect. 5.5.) Hence (5.37) becomes:
O
2T w.; O :/ kL.; A.;
:/ D .I kL.k//A.; O ://h: (5.39)
t
u
Lemma 5.5.
.I kL.k//1 D .I C kL.0//: (5.40)
Proof. Either by the Volterra Expansion
1
X
.I kL.k//1 D .kL .k//n :
nD0
Or, solving
.I kL.k//1 f D g;
Z x
f .x/ D g.x/ k ek.xs/ g.s/ds:
b
Hence
f 0 .x/ D g 0 .x/ kg.x/ C k.g.x/ f .x//;

g0 .x/ D f 0 .x/ C kf .x/:
Hence Z x
g.x/ D f .x/ C k f .s/ds; since f .b/ D g.b/:
b
Hence
.I kL .k//1 D .I C kL.0//: t
u
Hence
O :/ kL.k; A.;
2.I C kL.0//T wO a .; :/ D A.; O ://.I C kL.0//h (5.41)
from which we see that all we need to evaluate is the functional:
O ://:
L.k; A.;
For which we take the functional L.k; :/ on both sides of (5.41). L.k; lhs/ D
L.k; rhs/ which yields
O :// D .L.k; .I CkL.0//2T w.;

L.k; A.; O :///=.1kL.k; .I CkL.0//h//: (5.42)
However we still need to verify that the denominator in (5.42) is not zero. Here we
have a remarkable result due to Sears [110].
Sears Formula
Lemma 5.6. (Sears)

For Re: k > 0:
Z 1
p 2b C
.1 kL.k; .I C kL.0//h// D k ek d:
0
Proof. First we calculate

kL.k; .I C kL.0//h/:
First, integrating by parts in
Z b Z x
k.bx/
L.k; kL.0/h/ D e k h.s/dsdx
b b
yields
Z b
D h.s/ds L.k; h/:
b
Hence
0 s 1
Z 1 Z
p 2b C bx
b
kL.k; .I Ck L.0/h// D k ek @1 1
dx A d;
0 b bCx xb
where the integral in parenthesis

p
D 1:
2b C
Hence we obtain that

Z 1
p 2b C
1 kL.k; .I C kL.0//h/ D k ek d: (5.43)
0
That it is analytic in k in the entire plane except for k 0 where it has a logarithmic
singularitytypifying the analyticity properties functions of k we shall use. This
can be seen from rewriting (5.43) in the formby a change of variable:
Z 1
p 2bk C t
et dt: (5.44)
0 t
Or expressing it in terms of Bessel K functions:

Z 1
k p 2b C
k e d D bkK0 .bk/ C K1 .bk/ebk : (5.45)
0
But now the interesting question is: what is

, Z
1
k t p 2b C t
1 k e dt : (5.46)
0 t
Is this a Laplace transform? Is (5.46) expressible as

Z 1
c1 .t/ek t dt; c1 . : / 0:
0
This is answered in the affirmative by Sears [110]as we have seenwho actually

gives an approximation to the function c1 . : /, which is referred to as the Sears
function. This is necessary to obtain the time domain structure response. But it is
easier of course to show just that (5.44) is bounded away from zero.
In the form (5.44) we see that as k goes to zero in the right-half-plane, the
function has the limit equal to 1, and goes to infinity as Re k goes to infinity. In
fact we have
Z 1 Z 1
p 2bk C t p 2bk
et dt 1 D et C 1 1 dt 0 for k 0
0 t 0 t
p
and Re 1 C z 1 0 for Re z 0. t
u
Solution of Possio Equation in Laplace Domain: M D 0
Hence finally we have the solution:

2 3
O :/ D .I C kL.0// 6 .L.k; .I C kL.0//2T wO a .; :/// 7

A.; 42T wO a .; :/ C h R1 q 5:
e t 2bkCt
dt
0 t
Or

O :/ D .I C kL.0// 2T wO a .; :/ C h .L.k; .I C kL.0//2T wO a .; :/// :
A.;
bk.K0 .bk/ C K1 .bk//ebk
(5.47)
To show that
O b/ D 0;
A.;
we make a change of variable
D t 2 .b x/
in s r
Z 1
1 bx k 2b C 1
h.x/ D e d
bCx 0 xb
yielding
r Z 1
1 1 p 2dt
ek t
2 .bx/
D .1/ 2b C t 2 .b x/ ;
bCx 0 1 C t2
and in this form it follows readily that h.x/ ! 1 as x ! b. Now from (5.34),
evaluating the limit as x ! b, we see that the left side goes to zero, while the right
side
O :/ kL.; A.;
D .I kL.k//A.; O ://h
Z b Z b
O
0 D A.; b/ k e k.bs/ O
A.; s/ds C k O s/ds:
ek.bs/ A.;
b b
O b/ D 0.
Or A.; t
u
Remarks: It is of interest in the sequel to determine the range of values of k beyond
Re k > 0 for which the solution continues to be valid.
Here we use the solution in the form (5.47) involving the Bessel K
functions. They are defined and analytic and nonzero except for k < 0I and we
have already covered k D 0: Hence the solution is valid for all k omitting the
negative real axis where we have a line of singularity. What about the kernel
in the equation? Here the problem is that 1=.k C i !/ is no longer the Laplace
transform of a function in the positive time domain for Re k negative. This ends the
incompressible case where we have presented a closed-form solution to the Possio
equation in the Laplace domain. We consider the timedomain solution later. Next
we go on to consider M > 0 where we no longer have the luxury of a closed-form
solution of the Possio equation (not as yet, of course).
5.4 Linear Possio Equation: Compressible Flow: M > 0
Now we go on to consider nonzero M: Unfortunately there is no longer a closed-

form solution available (except for M D 1). Nor does the technique of solution for
M D 0 carry over.
In fact we show that M D 0 is typical and may be labeled singular. We call
the flow compressible if the far field Mach number is bigger than 0. There is a
further qualification: transonic is somewhat vague, for M > 0:8 (this depends
on the Mach number of the flow in the interior; see [49]). This further distinction
plays no role in our theory and we do not invoke it. Our first objective is to develop
an abstract version of the equation.
5.4 Linear Possio Equation: Compressible Flow: M > 0 123
Abstract Version of Possio Equation M > 0
We use the language of semigroup theory of operators [16, 28]. We assume

throughout that
wa .t; :/ is in C1 b; b for every t
and Z 1
e t jjwa .t; :/jjdt < 1 > a > 0:
0
For most if not all of our purposes we may take the abscissa a to be zero. The C1
condition is indeed true for the downwash.
Let .!/ denote the Lp Lq Mikhlin multiplier
1 j!j p
.!/ D NO i !/ ;
1 M 2 1 C B.k;
2 i!
where
Z
NO i !/ D k 1 1 1
1
B.k; p a.M; s/ds :
1 M 2 k C i! 2 ks C i !
Let us convert this multiplier to the corresponding operator on Lp .b; b/,
1p
1 M 2 H.I C B.k//
2
so that the Possio equation (5.23) in abstract form is:
1p O
wO a .; :/ D 1 M 2 PH.I C B.k//PA.; :/; (5.48)
2
where H is the Hilbert transform
Z 1
1
Hf D gI g.x/ D f .s/ds; 1 < x < 1
1 .x s/
and B.k/ can be expressed:

Z
kR.k/A 1
B.k/A D p kR.ks/Aa.M; s/ds; A Lp .b; b/;
1 M2 2
where R.k/ is the bounded linear operator on Lp .R1 / into itself corresponding to
the multiplier
1
;
k C i!
and is the resolvent of the generator D, defined for all k off the imaginary axis.
Df D f 0
generating the negative shift group S. : / on Lp .R1 /:
S.t/f D f .: t/ t 0
R.k/ is short for R.k; D/, the resolvent of D defined for off the imaginary axis:
Z 1
R.; D/f D et S.t/f dt; Re > 0;
0
Z 1
R.; D/f D et S.t/f dt; Re > 0:
0
Again P denotes the projection operator on Lp .R1 / into Lp b; b 1 < p < 2. The
Hilbert transform can be expressed [10] as
Z 1
1 .S.t/f S.t/f /
Hf D dt;
0 t
where
HR.; D/ D R.; D/H
and we note that Z 2b
PR.; D/Pf D et PS.t/Pf dt
0
is an entire function of . In particular:

Lemma 5.7. As Re k goes to infinity, B.k/ converges strongly over Lp .R1 / to
M
I p H;
1 M2
which we denote as B.1/.
Proof. B.1/ is the operator corresponding to the multiplier:
M j!j
1 p :
1 M2 i!
The multiplier corresponding to B.k/ is

s
NO i !/ D i ! kM 2 .k C 2i !/
B.k; 1C 1;
k C i! ! 2 .1 M 2 /
which, as Re k ! 1; converges pointwise to the multiplier
M j!j
1 p :
1 M 2 i!
This is enough but we establish this directly using elementary semigroup theory:
Z
kR.k; D/f 1
B.k/f D p kR.ks; D/f a.M; s/ds:
1 M2 2
We note that
kR.k; D/f ! 0 as Re k ! 1:
Hence the first term goes to zero. But jjkR.k; D/jj does not. kR.k; D/ converges
only strongly. However, we do have that
jkj
jjksR.ks; D/jj ; Re k > 0; 2 < s < 1 :
Re k
And the integral can be expressed:
Z 1 Z 1
kR.ks; D/f a.M; s/ds a.M; 0/ kR.ks; D/f ds
2 2
Z 1
C kR.ks; D/f .a .M; s/ a.M; 0//ds
2
where
M
a.M; 0/ D p
1 M2
and
Z 1 Z 1Z 1
kR.ks; D/f ds D ke kst ds S.t/f dt
2 0 0
Z 1Z 2
ke kst ds S.t/f dt
0 0
Z 1Z 1
D ekst ds k.S.t/f S.t/f /dt
0 0
Z 1Z 2
ke kst ds S.t/f dt
0 1
Z 1
.S.t/f S.t/f /
D .1 ek1 t / dt
0 t
Z 1
.ek1 t ek2 t /
S.t/f dt;
0 t
which converges strongly to

Z 1
.S.t/f S.t/f /
dt D Hf:
0 t
Now Z 1
a.M; s/ a.M; 0/ kR.k; D/f
ksR.ks; D/f ds p
2 s 1 M2
converges boundedly in any sector of opening less than =2 to
Z
1
a.M; s/ a.M; 0/ f
f ds p :
2 s 1 M2
To evaluate this, let us note that

Z 1 Z 1
k k
a.M; s/ds D .a.M; s/ a.M; 0//ds
2 ks C i ! 2 ks C i !
Z 1
k
C a.M; 0/ ds;
2 ks C i !
where the second term goes to the multiplier corresponding to the Hilbert transform:
M j!j
p :
1 M i!
2
But
Z 1
N
O k 1 k
B.k; i !/ D p .a.M; s/ a.M; 0//ds
1M 2 k C i ! 2 ks C i!
Z 1 Z 1
M k k
Cp ds .a .M; s/ a.M; 0//ds
1 M 2
2 ks C i ! 2 ks C i!
Z 1
NO i !/ C
D B.k;
k 1
p
M k
p ds :
1 M 2 k C i! 1 M 2 2 ks C i w
Hence as Re k goes to infinity in the sector, we have

Z 1
1 1
.a .M; s/ a.M; 0// ds p D 1;
2 s 1 M2
which establishes the result. But we emphasize that the convergence is only
strong. t
u
Using this result we can express the Possio equation in abstract form as an
equation in Lp b; b:
p
2wO a .; :/ D O
1 M 2 PH.I C B.k//A.; :/
p
D O
1 M 2 PH.I C B.k/ B.1//A.; :/
p
M O
C 1 M 2 PH I p H A.; :/:
1 M2
Or
p
2 O 1 M2 O
wO a .; :/ D A.; :/ C PH.B.k/ B.1//A.; :/; 0 < M < 1;
M M
(5.49)
which is defined for all off the imaginary axis where R.k; D/ is defined. It is
important to note that in (5.49), we are restricted to 0 < M < 1. The cases M D 0
and M D 1 are treated separately below. Note that (5.49) does not impose the Kutta
condition.
The Kutta Condition
It is not evident from (5.49) that the solution will satisfy the Kutta condition: For
this we go to an equivalent but different version of the LDP (Laplace domain Possio)
by invoking the Tricomi operator.
Theorem 5.8. The Possio equation can be expressed in the form:
2T wO a .; :/ O
p D ..I C PB .k/P/ C T PH.I P/B.k/P/A.; :/; (5.50)
1 M2
where T is the Tricomi operator.

Proof. For nonzero M we multiply both sides of (5.48) by M to obtain:
p
2w O
O a .; :/ D M A.; :/ C O
1 M 2 PH.B.k/ B.1//A.; :/;
where
p p
1 M 2 PH.B.k/ B.1// D 1 M 2 PH.B.k/ C I / MI:
Hence
O a .; :/
2w O O :/;
p D PHA.; :/ C PHB.k/A.;
1 M2
where the right side can be expressed:
O
D PHPA.; O
:/ C PHPB.k/PA.; O
:/ C PH.I P/B.k/A.; :/: (5.51)
t
u
Lemma 5.9. Suppose f is in Lp b; b; 1 < p < 2 and
T f D 0:
Then
f D 0:
Proof. Let
T f D g;
s r
Z
1 bx b
b C s f .s/
g.x/ D ds;
bCx b bs sx
which is given to be
D 0; jxj < b:
Hence Z r
b
b C s f .s/
ds D 0; jxj < b:
b bs sx
Hence (see [11]):
r
bCs const
f .s/ D p ;
bs b2 s2
const
f .s/ D ; jsj < b;
bCs
which leads to a contradiction, because this function is not in Lp .b; b/;

1 < p < 2. t
u
Hence
O a .; :/
2w O O :/ D 0
T p PHA.; :/ C PHB.k/A.;
1 M2
is equivalent to
O a .; :/
2w O :/ C PHB.k/A.;

O :/ D 0;
p PHA.;
1 M2
proving the theorem. t

u
Lemma 5.10. PB.k/P is a Volterra Operator on Lp .b; b/ into itself, and its
range is contained in C1 b; b. It is defined for all k and is an entire function.
Proof.
Z 1
kPR.k/PA
PB.k/PA D p kPR.ks/PA a.M; s/ds
1 M2 1
Z 1
kPR .ks/PA a.M; s/ds;
2
where Z 1
PR.k/Pf D gI gD ek t PS.t/ P f dt:
0
And because
PS.t/P D 0 for t > b;
we have: Z x
g.x/ D ek.xs/ f .s/ ds; jxj < b:
b
Let us use the notation:
L.k/ D PR.k/P
L.k/ is then an entire function of k. And
Z
kL.k/ 1
PB.k/P D p kL.ks/ a .M; s/ds
1 M2 2
is a Volterra operator on Lp .b; b/ into itself, and L.k/ is defined for every k in
the plane. In fact we have for
A D PB.k/Pf
Rx
k ek.x/ f ./d
b
A.x/ D p
1 M2
Z x Z 1
k f ./ eks.x/ a.M; s/ ds d; jxj < b:
b 2
Furthermore, we see that A . : / is in C1 b; b; so that PB.k/P maps Lp b; b

into C1 b; b. t
u
Getting back to (5.51), we see that by the Tricomi theorem
h i
O
T PHPA.; O
:/ C PHPB.k/PA.; O
:/ D .I C PB.k/P/A.; :/
and is in Lp b; b. Next note that the third term on the right in (5.51),
O
PH.I P/B.k/A.; :/;
is in Lp b; b, 1 < p < 2. Hence by the Tricomi theorem [11],
O
T .PH.I P/B.k/A.; ://
is in Lp b; b for 1 0 by [5, 8] for A in Lp b; b; 1 <
p < 2,
Z 1
k
W .k/A D p h .k/L .k; A/ C k h .ks/L .ks; A/a.M; s/ds
1 M2 0
Z 2
Ck .hC .ks/ C j.ks//LC .ks; A/a .M; s/ds; (5.52)
0
where the functionals are

Z b
LC .k; A/ D ek.bC/ A./d;
b
Z b
L .k; A/ D ek.b/ A./d
b
and the functions are

s r
Z 1
1 bx 1
hC .k; x/ D ek d;
bCx 0 2b C b C C x
s r
Z 1
1 bx 2b C 1
h .k; x/ D ek d;
bCx 0 bC x
j.k; x/ D ek.bCx/; jxj < b:
Each function on the right in (5.52) is in Lp b; b; 1 < p < 2. Furthermore,
hC .k; b/ D 0;
h .k; b/ D 1:
Proof.
W .k/A D T PH.I P/B.k/PA. : /

Z
kR.k/A. : / 1
D T PH.I P/ p kR.ks/A. : /a.M; s/ds :
1 M2 2
Computing the first term we have

s s
Z
k bx b
bC 1
T PH.I P/kR.k/A. : / D
bCx b b x
Z 1 Z Z s
1 1
b ek.s/ .PA/./d ds d:
1 b s 1
We therefore have
s s
Z
k bx b
bC 1
k T PH.I P/R.k/A. : / D
bCx b b x
"Z Z # Z
b 1 s
1 1
C ek.s/ .PA/./ d ds d
1 b s b
s s
Z
k bx b
bC 1
D
bCx b b x
Z 1 Z b
1 1
ek.s/ .A/./d ds d:
b s b
Changing variables using s D C b, and changing the order of integration we

obtain
s
Z s
k b x b b C 1
T PH.I P/kR.k/A. : / D
b C x b b x
Z 1 Z b
1 ek
ek.b/ A./d d d
0 . C b/ b
Z Z
k 1 b k.bs/
D e A.s/ds ek
0 b
Z s
1 b bC 1 1
d d:
b b x . C b/
Evaluating the innermost integral
Z s
1 b
bC 1 1
I D d;
b b x . C b/
we get that
Z s
1 b
bC 1 1 1
I D d:
b b x . C b/ x . C b/
We next use the fact that

Z s
1 b bC 1
D 1 for jxj b and therefore;
b b z z2 b 2

1 2b C
I D 1 p 1
x . C b/ 2 C 2b
p
1 2b C
D p :
. C b/ x
Therefore, returning to the expression for T PH.I P/kR.k/A. : / we have

s p
Z 1
k bx ek 2b C
T PH.I P/kR.k/A./ D p
bCx 0 . C b/ x
Z b
ek.b/ A./d d
b
D kh .k; x/L .k; A/:
We now utilize this expression to compute

Z 1
T PHk .I P/R.ks/A. : /a.M; s/ds
0
Z 1
D kh .ks; x/L .ks; A/a.M; s/ds :
0
We finally compute the term

Z 2
T PHk .I P/R.ks/A. : /a.M; s/ds
0
starting with
Z 2
T PHk R.ks/A. : /a.M; s/ds
0
s s "Z #
Z b Z Z 1
k bx b
bC 1 1 b
1
D C C
bCx b b x 1 b b z
Z 2 Z 1
e ks.z/
.PA/./d a.M; s/ds dz d;
0 z
where we used the fact that R.k/ is the operator corresponding to the multiplier
1
ki !
which when Re k < 0 is
Z 1
R.k/f D ek.z/ f ./d
z
and hence we have

Z 2
T PHk R.ks/A. : /a.M; s/ds
0
s s "Z
Z b Z b#
k bx b
bC 1 1 1
D C
bCx b b x 1 b z
"Z Z #
2 b
eks.z/ A./d a.M; s/ds dz d: (5.53)
0 z
Now considering the first integral I1, we have

s s
Z Z b
k bx b
bC 1 1 1
I1 D
bCx b b x 1 z
"Z Z #
2 b
e ks.z/
A./d a.M; s/ds dz d
0 b
s "Z s #
Z Z 1
k 2
bx b
bC 1 1 1
D d
0 bCx 0 b b x C Cb
"Z #
b
ks. CbC/
e A./d a.M; s/ d ds:
b
Evaluating
Z s
1 b
bC 1 1
d
b b x C . C b/
Z s
1 b bC 1 1 1
D d
b b . C b/ C x x C . C b/

1
D 1 p 1
. C b/ C x 2 C 2b
p
1
D p ;
x C . C b/ C 2b
where we have used the fact that

Z s
1 b bC 1
D 1; for jxj 1:
b b z z2 b 2
Thus, I1 becomes
s p #
Z Z 1 Z
k 2 bx 1
I1 D p d
0 bCx 0 0 x C . C b/ C 2b
"Z #
b
eks. CbC/ A./d a.M; s/ d ds
b
Z 2
D k hC .ks; x/LC .ks; A/a.M; s/ds:
0
We now compute the second integral I 2 in (5.53) to get

s
Z s Z
k b x b b C 1 1 b 1
I2 D
b C x b b x b z
"Z Z #
2 b
eks.z/ A./d a.M; s/ds dz d
0 z
Z "Z #
2 b
D k T PHP e ks.:/
A./d a.M; s/ds
0 :
Z 2 Z b
D k eks.x/ A./d a.M; s/ds:
0 x
It now remains to subtract the term

Z 2
I3 D k T PHPR.ks/A. : /a.M; s/ds
0
Z 2
Dk L.ks/PA. : /a.M; s/ds
0
Z 2 Z x
Dk eks.x/ A./d a.M; s/ds:
0 b
Therefore, we get
Z 2
I1 C I 2 I 3 D k hC .ks; x/LC .ks; A/a.M; s/ds
0
Z 2 Z b
k eks.x/ A./d a.M; s/ds
0 b
Z 2
D k hC .ks; x/LC .ks; A/a.M; s/ds
0
Z 2 Z b
k eks.xCb/ eks.bC/ A./da.M; s/ds
0 b
Z 2
D k .hC .ks; x/ C eks.xCb/ /LC .ks; A/a.M; s/ds
0
as desired. Next we show that h .ks; :/ is in Lp b; b. We have

s r
Z 1
1 bx 2b C 1
h .ks; x/ D eks d; s > 0; (5.54)
bCx 0 bC x
where the integral:

Z r
1 2b C
1
eks d ; s>0
0 bC x
is Z r
1
1 2b C Re ks
e d
bx 0
and s
bx 1 1
Dp ; jxj 0;
bCx 0 2b C b C C x
s r
Z 1
1 bx 1 Re ks
jhC .ks; x/j e d; s > 0; (5.55)
bCx 0 2b C
and hence hC .ks; :/ is in Lp b; b; 1 0, Re k > 0, and
hC .ks; b/ D 0: (5.56)
Hence
Z 1 R b jA./jd r 1

h .ks; x/L .ks; A/a.M; s/ds b
0 b2 x2
Z 1r Z 1
1
a.M; s/eRe k s ds d; (5.57)
0 2b C 0
verifying that the second term in (5.52) is in Lp b; b; 1 < p < 2. Similarly
Z 2

.hC .ks; x/ C j.ks; x//LC .ks; A/a.M; s/ds
0
Rb s
Z r Z 2
b jA./jd bx 1 1
a.M; s/eRe k s ds d
bCx 0 2b C 0
Z b Z 2
C jA./jd eRe k sx a.M; s/ds;
b 0
and is thus in Lp b; b; 1 < p < 2. Finally
hC .k; b/ D 0 follows from (5.56)
and
h .k; b/ D 0 follows from (5.57). t

u
Theorem 5.12. Suppose the (Possio) equation given by (5.50) for any has a
unique solution. Then it is given by
O :/ D ..I C PB.k/P/ C W .k//1 2T wO a .; :/

A.; p
1 M2
O :/ satisfies the Kutta condition:
and A.;
O x/ ! 0
A.; as x ! b:
Lemma 5.13. The operator W .k/ is compact on Lp b; b into itself

1 < p < 2.
Proof. Follows immediately from (5.51) because it is defined in terms of the
continuous linear functionals LC .k; A/ and L .k; A/0 . t
u
It follows from the Lemma 5.13 that
I C PB.k/P C T PH.I P/B.k/P D I C C.k/;
where C.k/ is compact on Lp b; b into itself, 1 < p < 2. Hence either there is a
nonzero A such that
.I C C.k//A D 0
or
.I C C.k//
has a bounded inverse. Hence if the solution is unique, there is no nonzero A such
that
.I C C.k//A D 0
and the theorem follows.
Theorem 5.14. Suppose (5.50) has a unique solution. Then the solution will satisfy
the Kutta condition.
Proof. Let
O :/:
g D ..I C PB.k/P/ C T PH.I P /B.k/P /A.;
Then
Rx
k O /d
ek.x/ A.;
O x/ C
g.x/ D A.; b
p
1 M2
Z x Z 1
k O /
A.; eks.x/ a.M; s/dds; jxj < b
b 2
Z 1
k
Cp h .k; x/L .k; A/ C k h .ks; x/L .ks; A/a.M; s/ds
1 M2 0
Z 2
Ck .hC .ks; x/ C j.ks; x//LC .ks; A/a.M; s/ds:
0
Hence, by Lemma 5.6

Rb
k O /d
ek .b /A.;
O b/ C
g.b/ D A.; b
p
1 M2
Z b Z 1
k O /
A.; eks.b/ a.M; s/dds; jxj < b
b 2
Z 1
k
Cp C L .k; A/ C k L .ks; A/a.M; s/ds
1 M2 0
Z 2
Ck j.ks; b/LC .k; A/a.M; s/ds
0
O b/:
D A.;
But the right side by Theorem 5.16 D 0. t

u
Remarks: What is new is the introduction of the operator W .k/ which is defined
for all k, except for the negative real axis where it has a logarithmic singularity. See
below on the latter.
Existence Theorem
In lieu of a blanket existence theorem for every M and every k, we offer two results,
each with some limitation.
Existence Theorem for k in Bounded Vertical Strip
Theorem 5.15. For values of k in a bounded vertical strip, that is,
jRe: kj < kb < 1;
we can find M0 such that the LDP has a unique solution for all M < M0 ;
M0 > 0.
Proof. We begin with the equation in the form:
2T wO a .; :/ O
p D .I C PB.k/P C W .k//A.; :/:
1 M2
To make explicit the dependence on M , we decompose
PB.k/P as D PB0 .k/P C PBM .k/P;

where
PBP0 .k/P D kL.k/;

Z 1
1
PB M .k/P D k L.k/ p 1 k L.ks/a.M; s/ds;
1 M2 2
W .k/ D W0 .k/ C WM .k/;
where
W0 .k/A D kh .k/L .k; A/;

1
WM .k/A D .kh .k/L .k; A// p 1
1 M2
Z 1
Ck h .ks/L .ks; A/a.M; s//ds
0
Z 2
Ck .hC .ks/ C j.ks//LC .ks; A/a.M; s/ds;
0
so that
(Note P ! P B ! B here and below.)
2T wO a .; :/ O :/:
p D .I C PB0 .k/P C W0 .k/ C PB M .k/P C WM .k//A.;
1 M2
For M D 0; we have already proved that
.I C PB0 .k/P C w0 .k//
has a bounded inverse. Hence
O :/
.I C .I C PB0 .k/P C W0 .k///1 .PB M .k/P C WM .k//A.;
2T wO a .; :/
D .I C PB0 .k/P C W0 .k//1 p ;
1 M2
where the right side is the solution for M D 0.

Next we show that
.I C PB 0 .k/P C W0 .k//1 .PBM .k/P C WM .k//
can be made less than 1 in operator norm for all k, Re k > a for 0 < M < M0 for
some M0 .
We calculate
.PB M .k/P C WM .k//A

Z 1
1
D kL.k/A p 1 kL.ks/Aa.M; s/ds
1 M2 2

1
C.kh .k/L .k; A// p 1
1 M2
Z 1
Ck h .ks/L .ks; A/ a.M; s/ds
0
Z 2
Ck .hC .ks/ C j.ks//LC .ks; A/a.M; s/ds
0

1
D p 1 .kL.k/A C kh .k/L .k; A//
1 M2
Z 1 Z 1
kL.ks/Aa.M; s/ds C k h .ks/L .ks; A/a.M; s/ds
2 0
Z 2
Ck .hC .ks/ C j.ks//LC .ks; A/ a.M; s/ds;
0
where Z 1
kL.ks/Aa.M; s/ds
2
can be expressed
Z Z 1
1
.a.M; s/ a.M; 0//
D skL.ks/A ds kL.ks/Aa.M; 0/ds:
2 s 2
On a bounded strip continuous functions are bounded.

1
p 1 ! 0 as M !0
1 M2
and
p
1 .1 s/.2 C s/
a.M; s/ D ; 2 < s < 1
1s
q
2M 2 M2
1 s C 1M 2 s C 1M 2
2
D ;
1s
r
1 M2
a.M; 0/ D :
1 M2
Hence
2q 3
2M 2 M2
a.M; s/ a.M; 0/ 1 6 s 2 C 1M 2s C 1M 2 M 7
D 4 p 5
s s 1s 1 M2
"p #
1 s 2 .1 M 2 / 2M 2s C M 2 M.1 s/
D p
1 M2 .1 s/s
"p #
1 M 2 .1 s 2 / s 2 M.1 s/
D p
1 M2 .1 s/s
q
s2
M 1 1 M.1s/
D p :
1M 2 s
p
Hence for jsj < M 2 C 4M M; this is in absolute magnitude
p
1 s
< 1 M 2 :
2 .1 s/
And hence it follows that

Z M
1CM
ja.M; s/jds ! 0 with M;
M
1M
Z M
1CM
ja.M; 0/jds ! 0 with M;
M
1M
Z
a.M; s/ a.M; 0/
M
1CM
ds ! 0 with M:
M s
1M
Hence we can find M0 such that
jj.I C PB0 .k/P C W0 .k//1 .PB M .k/P C WM .k//jj < 1
for k in the strip.

Hence
.I C .I C PB0 .k/P C W0 .k//1 .PB M .k/P C WM .k///1 A

1
X 1
D .1/n ..I C PB0 .k/P C W0 .k//1 .PB M .k/P C WM .k///n A;
nD0
for any element A in Lp b; b, and finally
O :/ D .I C .I C PB0 .k/P C w0 .k//1 .PBM .k/P C WM .k///1

A.;
2T wO a .; :/
.I C PB0 .k/P C W0 .k//1 p
1 M2
2T wO a .; :/
D .I C PB 0 .k/P C W0 .k//1 .I PB M .k/P C WM .k// p
1 M2
up to the first two terms. t

u
Existence Theorem for Small k
Theorem 5.16. The Possio equation has a unique solution for small enough k,
given by the Neumann expansion:
X1
O :/ D 2T wO a .; :/
A.; .1/n .PB.k/P C W .k//n p :
nD0 1 M2
In particular for k D 0, we have:
O :/ D 2T wO a .0; :/
A.0; p
1 M2
and satisfies the Kutta condition if
wO a .0; :/ is in C1 b; b:
Proof. We dont need to invoke the fact that W .k/ is compact. We prove that
PB.k/P C W .k/ goes to zero in operator norm with k.
Now jj.I C PB.k/P/1 -I jj goes to zero as k goes to zero, because jjPB.k/Pjj
goes to zero as k goes to zero. Hence it is enough to show that jjW .k/jj is small for
small k.
Now
s
Z r
1 bx 1 1
hC .k; x/ D ek d
bCx 0 2b C b C C x
s r
Z 1
1 bx 1
D e d;
bCx 0 2bk C C k.b C x/
which is then defined for every k except k 0.

For k D i !, this yields

s r
Z 1
1 bx 1
hC .i !; x/ D e d;
bCx 0 2bi ! C C i !.b C x/
s r
Z 1
1 1 bx 1
jhC .i !; x/j p e d:
2bj!j bCx 0
And for k > 0,

s r
Z 1
1 bx 1
e d
bCx 0 2bk C
s r
Z 1
1 bx 1
e d
bCx 0 2bk
r s
Z 1
1 1 bx 1
p e d:
2bk bCx 0
Hence for k D jkjei; we have

p !
jkj
jjkhC .k/jj D 0 p
j cos j
p

D0 jkj if D :
2
We have a similar estimate for jjkh .k/jj.

Hence the theorem holds as k goes to zero in such a way that the real part is bounded
away from zero, or equal to zero. We note that the larger the M , the smaller the k
we will need. The Neumann expansion follows as soon as k is small enough so that
the operator norm
jj.I CPB.k/P/1 W .k/jj < 1: t

u
Existence Theorem for Large k
The Neumann expansion does not converge for large Re k.

Here we go back to the Possio equation in the form (5.50):

p
2 O :/ C 1 M2 O :/
wO a .; :/ D A.; PH.B.k/ B.1//A.;
M M
whereas Re k goes to infinity. Here PH.B.k/B.1// goes to zero strongly only, not
O :/ converges
in operator norm! Hence, for example, if wO a .; : / D f . : / and if A.;
then the limit will be .2=M /f . : /:
Note this cannot hold for M D 0: The limit blows up as we show in that case.
We show that it does hold for M D 1: Note also that in (5.50) we cannot assert that
PH.B.k/ B.1//
is compact. Indeed we cannot assert that

p
1 M2
IC PH.B.k/ B.1//
M
has a bounded inverse even for large Re k.

Theorem 5.17. Suppose the Possio equation has a unique solution.
Then p
1 M2
IC PH.B.k/ B.1//P
M
has a bounded inverse on Lp .b; b/.
Proof.
O :/ D ..I C PB.k/P/ C T PH.I P/B.k/P/1 2T wO a .; :/

A.; p :
1 M2
Hence by (5.50)
p !
1 M2
IC PH.B.k/ B.1// .I C PB.k/P
M
2T wO a .; :/ 2
C T PH.I P/B.k/P/1 p D wO a .; :/:
1 M2 M
Here wO a .; :/ is any element of C1 b; b, therefore we have (for nonzero M):
p !
1 M2
I PH.B.k/ B.1// .I C PB.k/P
M
M
CT PH.I P/B.k/P/1 p T D I;
1 M2
where I is the identity operator on C1 b; b. Hence

" p #1
1 M2
IC PH.B.k/ B.1// f
M
M
D .I C PB.k/P C T PH.I P/B.k/P/1 p Tf
1 M2
for f in C1 b; b. t

u
The convergence for large Re k being only in the strong sense, we cannot assert
that either side converges. However, assuming both sides converge strongly, we can
check whether:
M
f D .I C PB.k/P C T PH.I P/B.k/P/1 p T f:
1 M2
Now for f in C1 b; b

MH
.I C PB.1/P/f D I P I C p Pf
1 M2
M
D p PHPf;
1 M2

M
T PH.I P/B.1/Pf D T PH.I P/ P p HP f
1 M2
M
Dp T PH.I P/HPf
1 M2
M
Dp .T C T PHPH/f:
1 M2
Because
T PHPHf D PHf; for f in C1 b; b;
we have
M
.I C PB.k/P C T PH.I P/B.k/P/1 p Tf
1 M2
T 1 T f D f:
Hence at least under the assumption that the Possio equation has a solution we do
have that p !1
1 M2
IC PH.B.k/ B.1// f Df
M
for f in C1 b; b as Re k goes to infinity.
This is as far as we go in terms of the question of existence of solution of the

LDP.
We turn next to the time domain solutions.
Time Domain Solution: Incompressible Flow
So far we have followed the traditional (and Possios own original formulation in
the frequency domain where the main concern is to calculate the aeroelastic modes.
However, as we delve more into the problem of stabilityour main concernwe
need the time-domain solution.
For incompressible flow we simply set M D 0 in the Possio equation. Let us see
what the corresponding linear potential equation is.
Going back to (5.3) we set M D 0 therein. This yields
2
@2 @2 @2 @ @2 @2
C 2U 1 cos C sin a1
2
C C D 0:
@t 2 @t@x @t@z @x 2 @z2 @y 2
Next we note that M D 0 for any U is equivalent to a1 ! 1:

2
Hence dividing both sides by a1 and taking the limit, we obtain
@2 @2 @2
2
C 2 C 2 D 0;
@x @z @y
which is then our field equation for M D 0. This is of course the

celebrated Laplace equation and time does not appear in it but does go into the
boundary conditions:
Far Field Velocity D i U 1 :
Flow Tangency
@1 .t; x; y; 0/
D wa .t; x/
@z
P y/ C .x a/.t;
D .1/h.t; P y/ C U1 cos .t; y/;
x; y 2 : (5.58)
KuttaJoukowsky
With the KussnerDoublet function
1 .t; x; y/
A1 .t; x; y/ D (5.59)
U1
D 0 off the wing ! 0 as x ! b as defined before:
These are then the boundary dynamics.

We note that we can go over to the finite plane model, with the nonzero angle of
attack making no difference except for replacing U1 by U1 cos , and small at
that.
The Mikhlin multiplier (5.14) becomes
q
1 1
!12 C !22 : (5.60)
2 C i !1 U1 cos
At the present time we have no solution technique for the corresponding Possio
equation.
Note that we get the beam or high aspect ratio solution if we set !2 D 0 which
we may consider as a first approximation and try a power series expansion
q
1 1 1 j!1 j 1 !22
!12 C !22 D 1C :. :
2 C i !1 U1 cos 2 C i !1 U1 cos 2 !12
See [55] for more on this.

Hence we specialize to the following:
Time Domain Solution Incompressible Flow: Typical Section

Theory/Zero Angle of Attack
We can rewrite the potential field equation (5.43) now as simply that the divergence
of the flow is zero:
r:q D 0;
where q is the velocity vector and this is the defining property of incompressible
flow.
The potential flow is now 2D:
.t; x; z/ D 0; x; z R2 ; except z D 0I jxj < b;

which is then the boundary; we call attention to the fact that it is a singular boundary
which makes all the difference and characterizes aeroelasticity.
We start with the Laplace Fourier transforms we have calculated. Thus
1 1 O i !/ejzj! ;
O1 .; i !; z/ D A.;
2 k C i!
O i !/ D .I C kL.0//.2T w
A.; O a .; :/
3
L.k; .I C kL.0//2T wO a .; :// O 7
C R1 q h5 ;
k t 2bCt
0 e t
dt
s r
Z
O 1 b x 1 k 2b C 1
h.x/ D e d;
bCx 0 xb
jxj < bk D=U1 : (5.61)
To get the time domain response we need to find the inverse Laplace/Fourier
transforms.
Possio Equation Time Domain Solution: Incompressible Flow
Theorem 5.18. Let u.t; :/ D 2T wa .t; :/, where wa .t; :/ is given by (5.54) and is in
C1 .b; b/, as before.
We assume that the initial structure state is zero so that
wa .0; :/ D 0:
Assume the Sears inversion formula:

Z 1
1
R1 q D e.t =U1 / c1 .t/dt
k 0 ek 2bC
d
0
Z 1
D et U1 c1 .tU1 /dt:
0
Then for jxj < b Z x

1
A.t; x/ D .t; x/ C P .t; s/ds;
U1 b
where
Z Z !
t b
.t; x/ D u.t; x/ C hc .t ; x/ .Pu.; s/ds d;
0 b
where
R1 q p
hO ek 1 bCx
bx 2bC 1
xb d
hO c D
0
R1 q D R1 q (5.62)
k 0 ek 2bC
d k 0 ek 2bC d
and the inverse transform:

Z t
hc .t; x/ D U1 h.U1 ; x/U1 c1 .U1 .t //d;
0
where s r
1 bx 2b C 1
h.; x/ D ; jxj < b:
bCx xb
Proof.
2 3
Z
O i !/ D .I C kL.0// 6
b
hO 7
A.; 4uO .; :/ C uO .; x/dx R q 5;
b 1 k t 2bCt
0 e t dt
which follows from (5.61) because we have:

Z b Z x
L.k; .I C kL.0//Ou.; :// D ek.bx/ uO .; x/ C k uO .; /d dx;
b b
where
Z b Z x Z b Z b
k.bx/
e k uO .; /ddx D uO .; /d ek.bx/ uO .; x/dx:
b b b b
Hence
Z b
L.k; .I C kL.0//Ou.; :// D uO .; /d:
b
Hence Z !
b
O i !/ D .I C kL.0// uO .; :/ C k
A.; uO .; x/dx hO c :
b
Let Z b
hO c
O .; :/ D uO .; :/ C k uO .; x/dx :
b k
Then Z x
1
kL.0/ O .; :/ D O .; /d:
U1 b
Hence taking inverse transforms we have:
Z b Z t
.t; :/ D u.t; :/ C u.t ; x/dxhc .; :/d;
b 0
Rx
which is (5.62). and kL.0/ O .; :/ is the transform of .1=U1 / b P .t; x/dx; which
proves the theorem. t
u
Next using this result, we can derive the solution to the potential equation obtained
a different way in [6].
Time Domain Potential
Theorem 5.19. The time domain potential:

Z b Z t
z 1
1 .t; x; z/ A.t ; y/ ddy: (5.63)
b 0 ..x y U1 /2 C z2 /
Proof. Use
Z 1
.k C i !/1 D U1 et i !t U1 dt;
0
Z 1
1 2z
ej!jz D ei !x dx; z > 0;
2 1 x2 C z2
Z 1 Z 1
ej!jz t i wx 1 z
D e e dxdt
.k C i !/ 0 1 .x U1 t/2 C z2
yielding
Z 1
O 1 .; i !; z/ D eix!
1
Z 1 Z t Z b
z 1
et dt A.t ; x y/ddydx:
0 0 b ..x y U1 /2 C z2 /
Hence:
Z b Z t
z 1
1 .t; x; z/ D A.t ; y/ ddy: t
u
b 0 ..x y U1 /2 C z2 /
Remarks: The formula (5.63) actually holds for every M. So do:
1 .t; x; z/ D 1 .t; x; z/;
@1 .t; x; z/ @1 .t; x; z/

D ;
@x @x
@1 .t; x; z/ @1 .t; x; z/
D :
@z @z
Also the Fourier transform of
@1 .t; x; z/
D i !:
@x
Our next case where we are able to obtain a closed-form solution to the Possio
equation is for M D 1, which is referred to as the sonic case.
The Sonic Case: Laplace Domain Typical Section;

For M D 1; the Possio multiplier equation becomes:

s
1 k 2 C 2ki ! O
O

.; i !/ D A.; i !/; (5.64)
2 k C i!
where the multiplier can be expressed in the form:

s
1 k 2 C 2ki ! 1 p k 1
.1; k/ D D . k/ 2 p : (5.65)
2 k C i! 2 k C i! k C 2i !
Then let as before L./ denote the operator to the multiplier 1=Ci !
corresponding
p
and L2 .k/ correspond to the multiplier 1=. k C 2i !/
Z x
f .s/
L2 .k/f D gI g.x/ D e.k=2/.xs/ p ds; jxj < b: (5.66)
b 2.x s/
This is also a Volterra operator on Lp .b; b/ into itself.

Thus we have
p
P.k; 1/PA.; O :/ D k.2I k.k//L2 .k/A.;
O :/
D w.;
O :/: (5.67)
The Tricomi operator plays no role at all! .2 I kL.k// has a bounded inverse
but not of course L2 .k/. Thus we have
O :/ D p1 .2I kL.k//1 w.;

L2 .k/A.; O :/; (5.68)
k
where
1 k k
.2I kL.k//1 D IC L :
2 2 2
But
L2 .k/f D g
is essentially the Abel integral equation to find f given g. But we have to restrict g.
Thus following Tricomi [11] we require that g have an Lp .b; b/ derivative also.
This defines a Sobolev space; we denote it Wp1 . We can verify that the right side of
(5.68) is indeed contained in this space in as much as
O
w.; :/ is in C1 .b; b/:
Hence we have the following.

Theorem 5.20. The Possio equation (5.67) has the unique solution
O :/ D p2 w
A.;
2
O a .; b/ C p L2 .k/.wO a 0.; :/ C k wO a .; ://: (5.69)
k k
Lemma 5.21. For g in Wp1 the integral equation L2 .k/f D g has a unique
solution in Lp .b; b/ given explicitly by:
f .x/ D f1 .x/ C f2 .x/ C f3 .x/; jxj < b; (5.70)
where
Z x
k g.s/
f1 .x/ D p e..k.xs//=2/ p ds;
2 b xs
Z
1 x
g 0 .s/
f2 .x/ D p e..k.xs//=2/ p ds;
b xs
1 ek..xCb/=2 /
f3 .x/ D p g.b/ p :
xCb
We note that
p 0 k
f1 C f2 D 2L2 .k/ g C g :
2
Proof. First we show that zero is not in the point spectrum of L2 .k/: For if
L2 .k/f D 0 for f in Lp .b; b/ by (5.61) we must have
Z x
h.s/
p ds D 0; jxj < b; where h.x/ D ek.x=2/ f .x/:
b xs
Next we follow the ingenious argument by Tricomi [11] to show that h. : / must
be zero.
Z x Z y Z x Z x
1 h.s/ 1
p p dsdy D h./ p dyd;
b x y b y s b .x y/.y /
where the inner integral is D from which the result follows.

Next (5.61) is also proved by Tricomi [11] and we can use the notation:
f D L2 .k/1 g
proving the lemma.

Next we go back to (5.68):
!
1 kL. k2 /
O :/ D L2 .k/
A.; 1
p IC O a .; :/:
w
2 k 2
To simplify this further, we note that if

!
kL. k2 /
gD IC f; then
2
f .b/ D g.b/
and
k
g0 C g D f 0 C kf
2
from which (5.69) follows. t
u
It is interesting to look at some special cases first.
Special Cases
1. Case 1: Solution for
wO a .; x/ D f1 .x/ D 1 jxj < b
is
p Z
2 ek..bCx/=2/ x
1
AO1 .; :/ D p p C2 k e.k=2/.xs/ p ds; jxj < b:
k 2.b C x/ b 2.x s/
Note that the solution does not converge as k ! 0, and goes to the value 2, as
Rek ! 1.2=M more generally).
This is consistent with our general theory for nonzero M .
2. Case 2:
wO a .; x/ D f2 .x/ D x; jxj < b;
left as an exercise.
Time Domain Solution M D 1
Next let us look at the time domain solution.

To get the field equation we have only to make M D 1 in the general form (5.3):
@2 @2 2
2 @
2
C 2a1 a1 D 0: (5.71)
@t @t@x @z2
The far field is now
1 D xa 1 because U1 D a1 :
The boundary dynamics are the same as for every M , except that now
U1 D a1 .
Hence to get the time domain response we have only to take inverse Laplace
transforms in (5.70) which here unlike the case M D 0, where we needed the Sears
theorem, is more straightforward.
Here we merely state it, referring to [19] for more details as necessary.
A.t; :/ D g1 .t; :/ C g2 .t; :/ C g3 .t; :/;

Z t
g1 .t; x/ D 2 wa .t ; b/G.; x/d;
0
5.5 Linear Time Domain Airfoil Dynamics 155

1 .b C x/ .1=2/ .b C x/
G.t; x/ D p t ; t> ;
2.b C x/ 2 2
Z tZ x
1 .x / .1=2/ 0
g2 .x/ D 2 p wa .t ; /dd;
0 b 2.x / 2
Z Z
d t x 1 .x / .1=2/
g3 .t; x/ D 2 p wa .t ; /dd
dt 0 b 2.x / 2
C 2wa .t; x/:
5.5 Linear Time Domain Airfoil Dynamics
The Convolution/Evolution Equation
Having calculated explicitly the pressure jump across the wing, for M D 0 and
M D 1 we now go on to consider the time domain structure response to the
aerodynamic lift and moment.
We begin with incompressible flow: M D 0:
It turns out that we can save a lot of tedious analysis by first taking the Laplace
transforms of the structure dynamics equations and waiting until the end before
inverting to the time domain. Here we follow [21] and consider Re > 0, unless
otherwise specified.
First note that now for us
O s/ C .x ab/.;
wO a .; x/ D h.; O O
s/ C U1 .; s/ jxj < b: (5.72)
To find the corresponding lift and moment it is convenient (and customary) to

normalize the half-chord b to be equal to 1. Thus we redefine k as
b
kD
U1
and define
O
w.; x/ D wO a .; bx/; jxj < 1;
which can be expressed as
O s/ C .k.x a/ C 1/.;
D U1 k h.; O s/; jxj < 1: (5.73)
And the normalized Possio equation, cf. (5.23), becomes

Z 1
O
w.; x/ D PO .; x /A.;
O /d; jxj < 1;
1
where continuing the normalization,
PO .; x/ D bP .; bx/; jxj < 1;

O x/ D A.; bx/;
A.; jxj < 1:
So we may proceed with using our formulae for the solution just taking b D 1;
with k now being taken as b.=U1 /. In particular let AOi .; x/ be the solution
O
corresponding to w.; x/ D fi .x/; i D 1; 2 where
f1 .x/ D 1I f2 .x/ D x; jxj < 1
and Z 1
wij D fi .x/AOj .; x/dx; i D 1; 2I j D 1; 2: (5.74)
1
To calculate the lift and moment we take:
O s/ C .x ab/.;
wa .; x/ D h.; O O
s/ C U1 .; s/ jxj < b (5.75)
O s/ C .k.x a/ C 1/.;
D U1 k h.; O s/; jxj < 1; (5.76)
so that the transform of the lift

Z 1
O
L.; s/ D bU 1 O x/dx
A.;
1

k O
O s/ C .kw12 C .1 ka/w11 /.;
D bU1
2
w11 h.; s/ (5.77)
b
and the transform of the moment

Z 1
MO .; s/ D b 2 U1 O x/dx
.x a/A.;
1

k O s/
D b 2 U1
2
.w21 aw11 /h.;
b

O
C .kw22 C .1 ak/w21 akw12 a.1 ak/w11 /.; s/ : (5.78)
Next let us calculate the wij for M D 0I D 0.

Thus for AO1 .; :/; we go back to (5.61) and note that where now
u.; :/ D 2T wO a .; :/
r Z s
1x 2 1
1C 1
u.; x/ D d
1Cx 1 1 x
r
1x
D2 ; jxj < 1
1Cx
and hence
Z 1
u.; x/dx D 2:
1
Hence
AO1 .; :/ D .I C kL.0//.u.; :/ C 2 hO c .; ://
and in turn
Z 1 Z 1
w11 D 2 C 2 hO c .; x/dx C k .1 x/.u.; x/ C 2 hO c .; x//dx;
1 1
where
Z 1
1
hO c .; x/dx D q 1
1 R1
k 0 ek 2C

d
1 1
D 1;
k .K0 .k/ C K1 .k//ek
where K0 .:/ : K1 .:/ are Bessel K functions of order zero and one.
Z Z r
1 1
1x
k .1 x/u.; x/dx D 2k .1 x/ dx
1 1 1Cx
D 3k:
And
Z 1
1
k .1 x/hOc .; x/dx D ek K1 .k/ ..K0 .k/ C K1 .k//ek / 1:
1 k
Using
Z 1
1 1
.1 x/3=2 p dx
1 1 C x.x 1 /
r r !
2C
D 1 1
2C
and
Z r ! , Z r !
1
k
p 2C 1
k t 2Ct
e .2 C / d e dt
0 0 t
R1 p
ek t . .2 C / /d 1
D 0
R1 q ;
k t 2Ct k
0 e t
dt
where
Z 1 p
ek . .2b C / /d
0
Z r !
1
k 1
D e .2b C / 1 d
0
Z 1 r !
d k 1
D e .2b C / 1 d; bD1
dk 0

d 1
D .ek .K0 .k/ C K1 .k//
dk k

D ek .K0 .k/ C K1 .k// C .ek /

K1 .k/ 1
K1 .k/ K0 .k/ C 2
k k
K1 .k/ 1
D ek 2:
k k
Hence

1 1
w11 D 2 C 2 1 C 3k
k .K0 .k/ C K1 .k//ek

1
C 2 e K1
k
..K0 .k/ C K1 .k//ek /1 2k
k
D 2T .k/ C k;
where
K1 .k/
T .k/ D
K0 .k/ C K1 .k/
is called the Theodorsen function [6].
We may note also that
r
1x
AO1 .; x/ D 2
1Cx
r Z r !
1
1x 2 C k d
C 2 e ek .K0 .k/ C K1 .k//
1Cx 0 x1
Z r Z r
x
1s x
1s
C 2k ds C 2k ds
1 1Cs 1 1Cs
Z 1r
2 C k d
e .ek .K0 .k/ C K1 .k///:
0 s1
This is another example of an explicit solution to the Possio equation.

Note that in this form we can show
O x/ D 0:
Ltx ! 1 A.;
Next Z 1
w21 D x AO1 .; x/dx D T .k/
1
omitting the details of the calculation, which are similar to the previous case.
We need next
AO2 .; x/:
Here
p
u.; x/ D 2 1 x 2 ;
Z 1
u.; x/dx D :
1
Hence
p
AO2 .; x/ D 2 1 x 2
r Z r !
1
1x 2 C k d
C e ek .K0 .k/ C K1 .k//
1Cx 0 x1
Z x p
C 2k 1 s 2 ds
1
"Z r Z r
#
1
x
1s 2 C k d
C k ds e e .K0 .k/ C K1 .k// :
k
1 1Cs 0 s1
Hence exploiting the previous calculations of integrals we obtain:
w12 D T .k/;

k
w22 D 1 C T .k/ ;
2 4
w11 D k C 2T .k/;
w21 D T .k/: (5.79)
These hold for any b; with

b
kD :
U
Next let us calculate the lift:

O k O s/
L.; s/ D bU 1 2
.k C 2T .k//h.;
b

O
C .kT .k/ C .1 ak/.k C 2T .k///.; s/

O s/ C .1 ak/k O .; s/
D bU 21 k 2 h.;
b

O s/ C b 1 a O .; s/U1 .;
C T .k/.2bU 1 / h.; O s/
2
O s/ C 2 b 3 a.;
D 2 b 2 h.; O O
s/ b 2 U1 .; s/

O 1 O O
T .k/2bU 1 h.; s/ C b a .; s/ C U1 .; s/ :
2
(5.80)
Note the appearance of terms containing 2 , and the term containing the
Theodorsen function. We note first that
Z 1
dt
K0 .k/ D ek t p ;
1 t2 1
K1 .k/ D K00 .k/
Z 1
tdt
D ek t p :
1 t2 1
Hence
Z 1
r
k t t C1
K0 .k/ C K1 .k/ D e dt;
1 t 1
Z 1
r
k t t C2
e .K0 .k/ C K1 .k// D
k
e dt;
0 t
Z 1
p
ek K0 .k/ D ek t .1=.t.t C 2///dt;
0
Z 1
p
ek K1 .k/ D ek t ..t C 1/=. .t.t C 2////dt
0
and
T .k/
D ek K1 .k/=ke k .K0 .k/ C K1 .k//
k
Z 1
D ek t L.t/dt;
0
where
Z t
C1
L.t/ D c1 .t / p d t > 0;
0 . C 2/
known as the Wagner function [6], where
c1 .t/ is nonnegative and hence so is L.:/

and Z 1
c1 .t/dt D 1
0
because
1=.k.K0 .k/ C K1 .k//ek / ! 1 as k ! 0:
Here it is known also (KussnerSears [6]) that
c1 .t/ D O.t .1=2/ / for small t
and we see so is
t C1
p :
t.t C 2/
Hence L.t/ does not go to zero as t goes to zero!
To find the limit let us note that T .k/ has the following expansion.
At zero
T .k/ D1 C .EulerGamma log2 C logk/k

C .EulerGamma2 2EulerGamma log2
C log22 C 2EulerGamma logk 2log2logk
C logk2 /k 2 C Ok3
showing the logarithmic essential singularity along the negative real line.
Also
kT 0 .k/ ! 0 as k ! 0:
At infinity:
3
1 1 1 1
T .k/ D C 2
CO :
2 8k 16k k
Note that
kT 0 .k/ ! 0 at infinity:
The inverse Laplace transform of T .k/ contains a delta function at the origin and
hence
T .k/ 1
k goes to D L.0C/ as Re k ! 1:
k 2
Now
1 1
T .k/ D .K1 .k/ K0 .k//=.K1 .k/ C K0 .k//
2 2
1
D k.ek K1 .k/ ek K0 .k//=.ke k .K1 .k/ C K0 .k///
2
and
Z 1
r
t k t
k.e K1 .k/ e K0 .k// D
k k
ke dt
0 t C2
Z 1 r
k t d t
D e dt
0 dt t C 2
Z 1
p
D ek t 1=. .t.t C 2/3 // dt:
0
Hence Z 1
1
T .k/ D ek t `.t/dt;
2 0
where Z t
p
`.t/ D c1 .t /1=. .. C 2/3 //d;
0
where Z 1
1
`.t/ 0I `.t/dt D :
0 2
Hence we have that
Z t
1
L.t/ D C `.t/dt;
2 0
L.1/ D1 D T .0C/;
Z 1 Z 1
1
ek t `.t/dt Dk ek t L.t/dt : (5.81)
0 0 2
Hence
as jkj ! 0; Re k > 0;
Z 1
k ek t L.t/dt ! 0: (5.82)
0
Next the moment:

k O s/
MO .; s/ D U1
2 2
b .T .k/ a.k C 2T .k// h.;
b

k
C k 1 C T .k/ .1 ak/T .k/ akT .k/
2 4

O
a.1 ak/.k C 2T .k// .; s/

O 1 O
D b a h.; s/ b a C
3 2 4 2
2 .; s/
8

C.U1 b 3 / O
a .; s/ C T .k/U1 2 2
b
2

k k O O
2a h.; s/ C k a.1 ak/2 .; s/ ;
b b 2
where the last term containing the Theodorsen function can be expressed as

kO O
2 2
T .k/U1 b h.; s/ C k .; s/
b 2
O O kO
2a .; s/ C a2 k2 .; s/ 2a h.; s//
b

1 O s/Cb.12a/.;
O O
D T .k/U1 b 2 Ca 2h.; s/C2U1 .; s/ :
2
It is convenient now to define a new function
O
w.; O s/ C b.1 2a/.;
s/ D 2h.; O O
s/ C 2U1 .; s/
with the inverse Laplace transform
P s/ C b.1 2a/.t;
w.t; s/ D 2 h.t; P s/ C 2U1 .t; s/; t > 0:
Hence using this function:
O
L.; s/ D 2 b 2 h.; O
O s/ C 2 b 3 a.; O
s/ b 2 U1 .; s/
T .k/bU 1 w.;
O s/

O s/ b 4 a2 C 1 2 O .; s/ C .U1 b 3 /
MO .; s/ D b 3 2 a h.;
8

O 1
a .; s/ C T .k/U1 b
2
C a w.;
O s/
2 2

O 1
D b a h.; s/ b a C
3 2 4 2
2 O .; s/
8
b 2 U1 O
O s/ C b 2 U 2 .;
O
w.; s/ C b 2 U1 h.; 1 s/
2

1
CT .k/U1 b
2
C a w.;
O s/:
2
Using

1 O
.U1 b 3 / a .; s/
2
b 2 U1 O s/ C b.1 2a/.;
O
D .2h.; s/
2
O
C2U1 .; 2 O
s// C b 2 U1 .; s/
b 2 U1 b 2 U1 O
O s/ C b 2 U 2 .;
D O
w.; s/ C 2h.; 1 s/:
2 2
We recall now our Hilbert space formulation of the structure dynamics in Chap. 2
with the Hilbert space H as therein and M and A defined as therein along with the
notation:

h.t/
x.t/ D ;
.t/

L.t/
R C Ax.t/ D
M x.t/ :
M.t/
We take Laplace transforms on both sides, with x.0/ D 0 and using the notation
O
x./ for the Laplace transform we have

O
L.; :/
2
O
M x./ C Ax./
O D O :
M .; :/
Substituting for
O
L.; :/
O
M .; :/
and collecting terms containing 2 and we have
2 Mx./
O C Dx./
O C Kx./
O C Ax./
O D FO ./x.;
O :/; (5.83)
where
!
m C b 2 s b 3 a
MD ;
S b 3 a I C b 4 .a2 C 18 /
which is again positive definite because

1 1 2
b 2 b 4 a2 C .b 3 a/2 D b 3 :
8 8
Next
!
0 b 2 U1
DD ;
b 2 U1 0
!
0 0
KD ;
0 b 2 U1
2
0 T .k/
1
k bU1
FO ./ D @ T .k/
b 2 U1
A:
k
U1 b 2 1
2
C a 2
We have thus obtained the structure dynamics under aerodynamic loads in the
Laplace transform domain.
The next step is to take the inverse Laplace transforms. Here the only slight
complication is that the inverse transform of T ./ contains a delta function. Thus
in taking the inverse transform of
T .k/wO a .; s/;
we note that we are already assuming that wa .t; :/ is differentiable in t and hence
we may write:

T .k/
O
T .k/w.; s/ D k wO a .; s/
k

b T .k/
D .wO a .; s//
U1 k
and hence the inverse transform can be expressed

Z t
b
L.t /wP a .; s/d;
U1 0
where L./ is given by:

Z t
p
L.t/ D c1 .t /. C 1/=. .. C 2///d:
0
Or as Z t
b 1
wa .t; s/ C `.t /w.; s/d :
U1 2 0
The first form is preferred traditionally [6] and so we use it for the time domain
version.
Hence with !
x.t/
Y .t/ D
P
x.t/
with range in H H and defining
8 9

0 >
>
< =
2U1
BD
2 >
>
: ;
b .1 2a/

B1
D ;
B2
we have
wa .t; :/ D B Y .t/ D B1 x.t/ C B2 x.t/:
P
We can now rewrite the structure equations in the time domain as
Z t
Mx.t/
R C Dx.t/
P C Ax.t/ C Kx.t/ D F .t /B YP ./d;
0
where !
bU1 L.U1 t/
F .t/ D :
b 2 a C 2l U1 .L U1 t/ 12
As usual we convert this to a first-order equation in time:
Z t
YP .t/ D AS Y .t/ C T Y .t/ C L.t /YP ./d
0
recognized as a convolution/evolution equation, where

0 I
As D
M1 A 0

0 0
T D
M1 K M1 D
(not to be confused with the Tricomi operator) the subscript S denoting structure.
The operator T reflects the so-called non-circulatory terms in the literature; see [6].
0 1
0
@L.t/Y D 0 A:
M1 F .t/B Y
We now modify the energy space H using M in place of M . Thus modifying (5.3)
we have:
The energy space H consists of elements of the form:

x
Y D 1 ;
x2
p
x1 2 D. A/; x2 2 H
endowed with the energy inner product:
p p
Y; ZE D Ax1 ; Az1 C Mx2 ; z2 ; (5.84)
where
z D z1
z2 ;
p
z1 2 D. A/; z2 2 H:
Note that if we set U1 to be zero, we get back to the pure en vacuo structure
equations of Chap. 2 without the aerodynamic forcing terms.
Hereinafter we simply continue to use ; and drop the subscript E, because the
context should make clear which inner product is being used. Also we use:

x h1 h2
Y D 1 I x1 D I x2 D :
x2 1 2
As is the generator of a C0 semigroup, therefore so is
A D As C T
as is well known [11, 16]. Let us denote the semigroup by S.t/; t 0. Unlike the
semigroup generated by As , this semigroup is not necessarily a contraction. In fact
we have:
ReAY; Y D ReT Y; Y ;
because
ReAS Y; Y D 0I ReDx2 ; x2 D 0
and
ReT Y; Y D ReKx1 ; x2 D CU1
2 2
b Re1 ; 2 ;
which shows that the semigroup S./ may not be a contraction.
But the growth bound is of course finite because the generator is a perturbation of
As by a bounded operator; and the precise growth bound is not needed here anyway.
What is important, however, is that the resolvent of A, denoted R.; A/ is
compact because the resolvent of As is compact.
What can we say about the convolution part?
The properties of L./ are readily obtained from those of the numerical function
L./. We have seen that
1
L.0/ D I L.1/ D 1
2
and the derivative of L.t/ is given by `.t/ which is continuous and is integrable
0; 1: L./ is nonnegative and
Z 1
ek t L.t/dt < 1 Re k > 0:
0
Hence we define Z 1
O
L./ D et L.t/dt Re > 0:
0
Again thus defined, its properties depend upon .T .//= which is not defined along
the negative real axis, including zero, and can be extended analytically to the whole
plane, except the negative real axis and D 0. On the other hand L./ O which
involves only T ./ is now defined at D 0 by continuity. Hence L./ O is defined
on the whole finite plane excepting < 0:
The calculation of the aeroelastic modes is the central part of Flutter analysis,
which is really part of the spectral theory of the stability operator.
Spectral Theory: Aeroelastic Modes
M D 0I D 0:
But before we prove existence and uniqueness of the solution of the
convolution/evolution equation we need to first study spectral theory, which also
plays a crucial role in stability theory.
Thus we go back to the Laplace transform theory.
We have: for every except the negative real axis:
2 Mx./
O C Dx./
O C Kx./ C Ax./
O D FO ./x.;
O :/;
which by taking Laplace transforms of

Z t
YP .t/ D AS Y .t/ C T Y .t/ C L.t /YP ./d
0
takes the form:
O
.I A L.// YO ./ D .I L.//Y
O .0/;
where YO ./ is the Laplace transform:

Z 1
YO ./ D et Y .t/dt
0
and
O
L./ O
D M1 F./B
;
which is defined in the whole plane except for Re < 0. Hence, following
semigroup theory techniques, we consider the generalized resolvent equation:
O
.I A L.//Y DZ
in H:
Exploiting the compactness of R.; AS / we have the next theorem.
Theorem 5.22. Given any in the resolvent set of As (not an en vacuo structure
mode in other words), and omitting the negative half-line, either
O
.I A L.//Y D0 Y 0
O
for some nonzero Y or .I A L.// has a bounded inverse, which we call the
generalized resolvent.
Proof.
O
I A L./ O //:
D I As .T C L.Y
Hence
O
R.; AS /.I A L.//Y O
D .I R.; AS /. C L.///Y;
where
O
R.; As /.T C L.//
is compact because R.; As / is.
Hence either
O
.I R.; AS /.T C L.///Y D0
for some nonzero Y in which case
O
.I A L.//Y D0
or
O
I R.; AS /.T C L.//
and equivalently
O
.I A L.//
has a bounded inverse. t
u
We call the bounded operator
O
.I A L.// 1
the generalized resolvent, denoted R./.

Note, however, that R./ does not satisfy the resolvent equation [10, 41].
Aeroelastic Modes
The zeros of I A L./ O are called the aeroelastic modes. They are not
identified as eigenvalues at the expense of changing the space, as we show later.
The zeros of the adjoint:
O
.I A L.//
Y D0
are called the adjoint modes.

The main objective in flutter analysis is to calculate these modes and the
corresponding mode shapes Y .:/.
Calculating Aeroelastic Modes: M D 0
We begin with M D 0: Here we recall the analysis in Chap. 2, including now the
aerodynamic loading. Thus solving
O
.I A L.//Y D0
leads us back to where we began:

O
L.; :/
2
O
M x./ C Ax./
O D O ;
M .; :/
which unravels to
1 h 2 O i
hO .; s/ D O
0000
.mh.; s/ C S .; O
s// L.; s/ ;
EI
1 h 2 i
O .; s/ D O s// MO .; s/ ; 0 < s < `
00
.I C S h.;
GJ
with the end conditions:
O 0/ D 0I
h.; hO .; `/ D 0I
000
hO .; `/ D 0;
00
O
.; 0/ D 0I O
.; `/ D 0:
Note immediately that if is a mode, so is the conjugate.

This is recognized as a two-point boundary value problem for ordinary differen-
tial equations for each .
Two-Point Boundary Value Problem
To solve this let us define the 6 1 vector

0 1
O s/
h.;
B hO 0 .; s/ C
B C
B O 00 C
B h .; s/ C
Y .; s/ D B O 000 C:
Bh .; s/ C
B C
O
@ .; s/ A
O 0 .; s/
Then the equation becomes:
d
y.; s/ D A./Y .; s/;
ds
where now (cf. (2.13)) 0 1
0 10 0 0 0
B0 0C
B 01 0 0 C
B C
B0 00 1 0 0C
A./ D B C;
Bw1 00 0 w2 0C
B C
@0 00 0 0 1A
w3 00 0 w4 0
where now:
1
w1 D .m2 C U1 bw11 /;
EI
1 h 2 i
w2 D S C bU 21 ..1 ak/w11 C kw12 / ;
EI
1 h 2 i
w3 D S C bU 21 .w21 aw11 / ;
GJ
1 h 2 i
w4 D I C b 2 U1
2
.w21 C kw22 a.1 ak/w11 ak.w21 C w12 //
GJ
and is defined for every except the negative half-line. The wij are given in (5.79)
for M D 0.
The solution is given by
y.; s/ D esA./ y.; 0/;
where to satisfy the end conditions (CF) we must have:
y.; 0/ D Qz./;
and
P y.; `/ D P e À./ Qz./; z./ 0: (5.85)
Hence we must have
d.0; ; U1 / D det P e À./ Q D 0;
where as before in Chap. 2:

0 1
001000
B C
P D @0 0 0 1 0 0A;
00000 1
0 1
0 00
B0 0 0C
B C
B C
B1 0 0C
QDB
B0
C
B 1 0C
C
B C
@0 0 0A
001
D P :
The main result here is the following

Theorem 5.23. The function
d.0; ; U /
for each U positive is an analytic function of except for a branch cut along the
negative real axis, logarithmic singularity along Re 0, with at most a countable
number of isolated zeros in a half-plane that do not have an accumulation in the
finite part of the plane.
The aeroelastic modes are the roots of
d.0; ; U1 / D 0:
Proof. The analyticity properties are inherited from those of the matrix function
A./ and in turn by those of T .k/; and ultimately the analyticity properties of the
Bessel K functions. Hence the stated analyticity properties follow. t
u
Let us explore the aeroelastic modes.
Let

x1 h1 h2
Y D ; x1 D ; x2 D I Y:Y D 1:
x2 1 2
Then
O
.I A L.//Y D0
or, equivalently:
x1 D x2 ;
Mx2 D Ax1 Kx1 Dx1
0 1
2
T .k/U1
C@ b 2 U1 A .2U1 1 C2h2 Cb.12a/2 /;
T .k/b.aC 12 /U1
2

2
where
h2 D h1 I 2 D 1 :
We begin with the special case D 0; of interest because it is an unstable mode
even though it is not a structure mode.
Theorem 5.24. Zero is an aeroelastic mode for a sequence Un of values of U1
given by p p
Un D .2n C 1/. GJ /=.2b` ..1 C 2a///:
Proof. We have:
x2 D 0;
3

2U1 1
0D Ax1 :
2b .a C 12 / U1 31
Or
!
EI h1 0000 .s/ 2
2U1 1 .s/
D ;
GJ 1 00 .s/ 2b.a C 12 / U1
2
1 .s/
where we see that

1
GJ 1 00 .s/ D 2b a C 2
U1 1 .s/ 0<s<`
2
with the CF end conditions:
1 .0/ D 0I 10 .`/ D 0
is an eigenvalue problem and has solution only for a sequence of values of

U1 given by
p
p
Un D .2 n C 1/ GJ .2b` ..1 C 2a/// ;
s
1 .s/ D const sin 0 < s < ` for all n: t
u
2`
The minimal value is called the divergence speed. See Sect. 5.5; of course the speeds
are also determined from solving
d.0; 0; U / D 0;
where d.0; 0; U / is an entire function of U of finite order and has a countable

number of zeros.
Theorem 5.25. The aeroelastic modes are confined to a finite vertical strip in the
complex plane; the real part is bounded.
Proof. Let be an aeroelastic mode and Y the mode shape, where we normalize so
that
Y; Y D 1:
Hence
O
Y; Y D D AY; Y C L./Y; Y
O
D 2Rex1 ; Ax 1 jj2 Dx1 ; x1 jj2 Kx1 ; x1 C L./Y; Y ;
where:

x1
Y D ;
x1
Y; Y D x1 ; Ax 1 C jj2 Mx1 ; x1 D 1
Dx1 ; x1 D 0I Kx1 ; x1 D b 2 U1
2
1 ; 1
Re.1 C 2x1 ; Ax 1 C b 2 U1
2
1 ; 1 /
O
D ReL./Y; Y ;
where the right-hand side goes to zero as Re ! 1. Hence Re is bounded. The
aeroelastic modes are confined to a vertical strip in the complex plane. Furthermore
there can only be a finite number of zeros with positive real part [8]. Indeed if we
have a sequence of the form: k D C i !k and !k ! 1 we would contradict the
asymptotic properties of the roots. t
u
Mode Shapes
By the mode shape corresponding to an aeroelastic mode k , we mean the

component xk;1 of Yk . : /:

xk; 1
Yk D ;
xk; 2

hk .s/
xk;1 D ; 0 < s < `:
k .s/
Let us calculate Yk .
O k/ D 0
k Yk AYk k L.
and hence we have:
xk;2 D k xk;1 ;
2k Mxk;1 D Axk;1 Kxk;1 k Dxk;1
0 1
k b 2
B T U k C
Uk
CB
@ k b C
1 k b Uk A
2
T b a C Uk
2
Uk 2 2
.2U1 k C 2 hk C b.1 2a/k k /:
Here we follow [56].

Recall that
Det:P e À.k / Q D 0: (5.86)
Hence there is a nonzero zk such that
P e À.k / Qzk D Q eÀ.k / Qzk D 0: (5.87)
Note that the dimension of the nullspace of the matrix is at most two. Define
zk .s/ D esA.k / Qzk 0<s<`
D colhk .s/; h0k .s/; h00k .s/; h000 0

k .s/; k .s/; k .s/;
zk .0/ D Qzk :
Thus defined zk .s/ is an analytic function of s.

Next define
Yk .s/ D colhk .s/; k .s/; k hk .s/; k k .s/; 0 < s < `:
Note that we can express Yk . : / as
Yk .s/ D T46 .k /esA.k / Qzk 0 < s < `;
P eÀ.k / Qzk D 0;
where T46 .k / is the 4 6 matrix given by:

0 1
10 00 00
B0 0 00 1 0C
T46 ./ D B
@ 0
C:
00 0 0A
00 00 0
And we have thus determined the mode shapes

xk;1
Yk D :
xk;2
Generalized Resolvent
We now derive a closed-form expression for the generalized resolvent.

The generalized resolvent equation

O
I A L./ Y D Yg ;
where

x1;g hg . : /
Yg D I x2;g D ;
x2;g g . : /

xl
Y D (5.88)
x2
reduces to the nonhomogeneous equations
x1 x2 D x1;g ;
1 h 2 O i h .s/
hO 0000 .; s/ D O O g
mh.; s/ C S .; s/ L.; s/ C ; (5.89)
EI EI
1 h 2 i
O s/ MO .; s/ C g .s/ ;
O 00 .; s/ D I C S h.;
GJ GJ
0 < s < `; (5.90)
d
y.; s/ D A./y.; s/ C yg .s/; (5.91)
ds

hg .s/ g .s/
yg .s/ D col 0; 0; 0; ; 0; ;
EI GJ
y.; 0/ D Qz./;
Z s
y.; s/ D e sA./
Qz./ C e.s /A./ yg ./d; (5.92)
0
Z
`
P eÀ./ Qz./ C v./ D 0I v./ D e.` /A./ yg ./d;
0
or,
D./z./ D P v./I D./ D P e À./ Q:
So, omitting the aeroelastic modes this has the solution:
z./ D D./1 P v./

and in turn:
Z s
1
y.; s/ D e sA./
QD./ P v./ C e.s /A./ yg ./d;
0
0<s<` (5.93)
and finally the generalized resolvent

x1 ./
R./Yg D ;
x2 ./
x1 ./ D T26 y.; :/I x2 ./ D x1 ./ x1;g ; (5.94)
where

h .; s/
T26 y.; s/ D 0 < s < `:
.; s/
Modal Expansion of Solution
Let us now get back to the solution to the initial value problem:

O
I .A/ L./ O
YO ./ D I L./ Y .0/;
which has the solution

YO ./ D R./ I L./
O Y .0/; k :
Because D./ has only simple zeros at k , it follows that R./ has only isolated
simple poles at k , confined to a finite strip.
Hence the inverse Laplace transform
Z L
1
W .t/ D lim et . Ci !/ R. C i !/d!; t > 0:
L!1 2 L
Following [33, 44] we deform the contour of the inverse Laplace transform to
obtain [8]
1 Z 1 !
X
rt
W .t/Y D Pk Y e C
k t
e J .r/Y dr
kD1 0
(where we define the integral term presently),

I
1
Pk Y D R./Y d; D k C rei ; r small enough
2 i

O k / Pk Y D 0:
k I A k L.
Or Pk is an idempotent mapping into the null space of

k I A k L.O k/ :
The null space is of dimension at most 2, but numerical calculations indicate it is

only l and we may assume this for simplicity.
Next note that the generalized resolvent is defined for every except fk g and
the negative half-line. We also exclude k D 0.
Then with r > 0 we have to consider the integral along the bounding lines:
D r C i ! and D r i !;
where we note the jump:
R.r; i !/ R.r; Ci !/ D J .r; !/
can be calculated in many ways, our main interest being in the limit as r goes to
zero.
We now define
1
J .r/Y D lim R.r i !/ R.r C i !/Y; r > 0;
2 i !!0
where we need to first show that the limits exist. Now
R.r; i !/ D ..r C i !/I A .r C i !/L.r C i !//1 Y;
where
0 1
0
L.r C i !/Y D @ 0 A;
Ml FO .r C i !/B Y
where
0 T .k/ 1
bU
B k C
FO .r C i !/ D @ T .k/ 1 b 2 U A ;
2
b U Ca
k 2 2
k D .r C i !/:
Note the appearance of the scale factor b=U for the first time.
We need next to calculate the jump. Now
z 2m
1
X
K0 .z/ D I0 .z/logz C I0 .z/log2 C 2 :
mD0
.m/2
So in the definition we need to go to the Riemann sheet where the value of the
logarithm jumps by 2 i every time we cross the negative half-line. Using for r >
0; ! > 0,
p !
logr C i ! D log r 2 C ! 2 C i. /; Tan D ;
r
p
logr i ! D log r 2 C ! 2 i. /;
logr C i ! logr C i ! D 2 i 2i;
which leads to the jump in the Bessel K functions:
K0 .r C i !/ D I0 .r C i !/logr C i ! C I0 .r C i !/log2

1
,
X 2k
C .r C i !/ 2k
2 .k/2 .k C 1/;
kD0
lim ! ! 0 .K0 .r C i !/ K0 .r i !// D I0 .r/2 i:
We can define:
K0C .r/ D limit ! ! 0 K0 .r C i !/ D K0 .r/ i I0 .r/;
K0 .r/ D limit ! ! 0 K0 .r i !/ D K0 .r/ C i I0 .r/:
Similarly:
K1C .r/ D limit ! ! 0 K1 .r C i !/ D K1 .r/ C i I1 .r/;
K1 .r/ D limit ! ! 0 K1 .r i !/ D K1 .r/ i I1 .r/:
Let
JT .r/ D lim ! ! 0.T .r i !/ T .r C i !//

K1 .r/ C i I1 .r/

D
K1 .r/ C i I1 .r/ C K0 .r/ C i I0 .r/
K1 .r/ i I1 .r/

:
K1 .r/ i I1 .r/ C K0 .r/ i I0 .r/
And
K1 .r/ i I1 .r/

lim ! ! 0 T .r C i !/ D D TC .r/;
K1 .r/ i I1 .r/ C K0 .r/ i I0 .r/
K1 .r/ i I1 .r/

lim ! ! 0 T .r i !/ D C JT .r/:
K1 .r/ i I1 .r/ C K0 .r/ i I0 .r/
Then
1
J .r/Y D lim R.r i !/ R .r C i !/Y; r >0
2 i !!0
1
D lim ..r i !/I A .r i !/L.r i !//1 Y
2 i !!0
..r C i !/I A .r C i !/L.r C i !//1 Y
1
D .rI A C rL .r//1 .r.L .r/
2 i

LC .r///.rI A C rLC .r//1 Y;
where
L .r/ D lim ! ! 0 L.r i !/;

LC .r/ D lim ! ! 0 L.r C i !/:
We are most interested in the limit as r ! 0, because of the familiar relation for
Laplace transforms: following (5.20)
Z 1 Z 1
rt
lim t ! 1 t e J .r/Y dr D eR J .R=t/Y dR
0 0
! J .0C/Y as t ! 1
and
J .0C/Y D 0:
Also we have: Z 1
1
X Pk Y J .r/
R./Y D C Y dt:
k 0 Cr
kD1
The idempotants Pk are not orthogonal but Pk Pj D 0 for k j and defining

Z 1
QY D J .r/Y dr;
0
we have that
lim R./Y D W .0/Y D Y
Re !1
and hence
1
X
Y D Pk Y C QY;
kD1
where
QP k D 0 D Pk QI Q2 D Q:
The mode shape vectors do not span the space H, nor are they orthogonal.
Evanescent States
What is novel here is the existence of evanescent states that decay faster than any
modal response, a phenomenon that becomes apparent only in continuum theory
and lost in the discretized models of CFD.
First the Definition
The evanescent states are the states in the range space of the operator Q. We note
that the response to evanescent states is given by
Z 1
W .t/QY D ert J .r/QY dr
0
and are thus Laplace transforms in t, and thus are stable for all values of U . In
[33] the structure is discretized and the aerodynamics is also approximated (Pade
approximation) and the authors call this part the nonrational component. Here we
see the exact solution and in particular it is stable for all values of the speed U .
Next we derive the time domain solution of the convolution/evolution equation
using state space theory.
5.6 State Space Theory
We are ready to consider now the time domain solution of the convolution/evolution
equation for given initial conditions. The solution given in [8, 22] involves a fair
degree of operator theory.
Here we use a different and more self-contained technique: state space represen-
tation which is of interest on its own for us; but requires some knowledge of the
theory of semigroups of operators.
State Space Representation: M D 0
Here we follow [16]. Our time domain equation is the initial value problem in H:
Z t
YP .t/ D AS Y .t/ C T Y .t/ C L.t /YP ./d (5.95)
0
with initial condition Y (0).

We show that this can be expressed in the state space form:
P
Z.t/ D AZ.t/;
Y .t/ D PZ.t/ (5.96)
in a Banach space and A is the infinitesimal generator of a C0 semigroup, and P is

a projection operator.
We make strong use of the fact that L.:/ is in C 0; 1 (the space of
continuous functions with finite limit at 1) and is absolutely continuous therein. In
particular:
L. : / 2 ;
where is the Banach space C 0; 1 of continuous 4 by 4 matrix-valued functions.
It is also absolutely continuous with derivatives therein.
Remarks: In view of these properties we could integrate by parts and get:

Z t Z t
L.t /YP ./d D P /Y ./d L.t/Y .0/
L.t
0 0
but this makes it awkward, bringing in the initial value in the state equation which
we want to avoid!
We begin with the convolution part which is defined by the function L. : /.
Consider the system inputoutput relation where u . : / is the input and v. : / is
the output
5.6 State Space Theory 185
Z t
v.t/ D L.t /u./d t > 0; (5.97)
0
RT
where the input u. : / takes its values in a separable Hilbert space and 0 jju.t/jjdt <
1 over every finite T . For representation theory, it is convenient to embed this class
of inputs in a larger class where u. : / is defined on .1; 1/ but vanishes on a
negative half-line:
u.t/ D 0 fort < L for some L; 0 < L < 1:
In that case we can rewrite the relation as

Z t
v.t/ D L.t /u./d:
1
Then of course we can extend this definition to any t, not necessarily positive. But
we are only interested in the t > 0 with which we started. The implication is that we
embed this in a controllable system [16] but this is irrelevant in our current context.
Theorem 5.26. Let X denote the Banach space C 0; 1I H . The inputoutput
relation Z t
v.t/ D L.t /u./d t >0
1
can be represented as
v.t/ D C x.t/;
P
x.t/ D Ax.t/ C Lu.t/ (5.98)
for t > 0; where x.t/ is in X for each t > 0; and A is the infinitesimal generator of
a C0 semigroup over X , and L is a linear bounded operator on H into X , and C is
a linear bounded operator on X into H .
Proof. Define L by
Lu D xI
x.t/ D L .t/u; t 0
t
u
and in as much as
jj x.t/ jj jj L .t/ jj jjujj;
it follows that x is in X .
Next define C on X by
C x D x.0/
and C is linear bounded because
jj x.0/ jj jjxjj:
Let S. : / denote the shift semigroup over X defined by:
S.t/x D yI y.s/ D x.s C t/; s 0:
It is a C0 semigroup. Let A denote the infinitesimal generator which is then

defined by
D.A/ D x. : / in X such that x 0 . : / is in X ;
Ax D yI y.t/ D x 0 .t/ t 0:
Next let denote the closed linear subspace generated by elements in X of the
form:
S.t/Lu u in H and t 0:
Then is an invariant subspace for the semigroup: S.t/ is contained in for
every t: We may take our state space to be and let Sc . : / denote the semigroup
restricted to , and let Ac denote the generator, being then a restriction of A.
Then we have the next lemma.
Lemma 5.27. The nonhomogeneous Cauchy problem:
P
x.t/ D Ac x.t/ C Lu.t/ t >0
with x.0/ in the domain of Ac , has a unique solution in such that:
x.t/ " D.Ac /

jjx.t/ x.0/jj ! 0 as t ! 0C
given by Z t
x.t/ D Sc .t/x.0/ C Sc .t /Lu./d:
0
Proof. The proof is standard; see, for example, [16]. t
u
Next let us calculate v. : /. We have
Z t
C x.t/ D CS c .t/x.0/ C CS c .t /Lu./d
0
Z t
D x.0; t/ C L.t /u./d
0
D v.t/
by taking x.0/ to be zero, where we have used the notation
x.t; s/ s 0
to denote x.t/.
Remarks: As is well known (see [16]) the method of variation of parameters
solution holds in the generalized or weak sense even if x.0/ is not in the domain
of the generator, with the continuity at the origin, but of course the solution need
not be in the domain of the generator.
Next let us calculate the resolvent of Ac :
Resolvent of Ac
Let R.; Ac / denote the resolvent of Ac . Then for Re > 0 we have
Z 1
R.; Ac /x D et Sc .t/xdt
0
and
.I Ac /x D LY; Y in H
has the unique solution for Re > 0 given by
x D R.; Ac /LY
and Z 1
Cx D C O
et Sc .t/LY dt D L./Y:
0
Lemma 5.28. Given any Y in H; we can find x./ in such that
.I Ac /x./ D LY
and
O
C x D L./Y
O
for in the region of analyticity of L./:
Proof. For Re > 0
Z 1
x./ D et Sc .t/LY dt
0
is an element of .
And the corresponding function

Z 1
x.; s/ D et L.t C s/Y dt s0
0
Z 1
D e s
e L./Y d
s
Z s
O
D es L./Y es e L./Y d (5.99)
0
0 s 1;
x.; 1/ D L.1/Y; (5.100)
O
x.; 0/ D C x.; : / D L./Y: (5.101)
The main point here is that thus defined for Re > 0 x.; s/ can be continued
O
analytically as an analytic function of in the region of analyticity of L./Y
which includes the whole plane except 0. Hence x.; : / can be continued
analytically satisfying
.I Ac /x.; : / D LY (5.102)
as required. t
u
Let us use x. to denote the state space, the product space xH which is then a
Banach space and denote the elements therein as

x
ZD x"; Y "H;
Y
And the projection operator P on X. into H by
PZ D Y:
Then we can recast the convolution/evolution equation (5.72) as
YP .t/ D AY .t/ C C x.t/;

XP .t/ D Ac x.t/ C LYP .t/: (5.103)
And further with

x.t/
Z.t/ D
Y .t/
and defining the linear bounded operator B on X. into itself by

LY
BZ D ;
0
we have:

P Ac 0 P
Z.t/ D Z.t/ C B Z.t/:
C A
Or

P Ac 0
.I B/Z.t/ D Z.t/:
C A
Because B is nilpotent we have
.I B/ D .I C B/1
and hence finally we can express the equation as

P Ac 0
Z.t/ D .I C B/ Z.t/
C A
D A.Z.t/; (5.104)
where

Ac 0 Ac C LC LA
A. D .I C B/ D :
C A C A
It is immediate that thus defined A. is closed linear with dense domain in X.
.Domain of Ac / .Domain of A./:
Thus we have the new state space form
P
Z.t/ D A.Z.t/; (5.105)
where Z. : / is the state, the state space is X.; and A. is the system
stability operator. To prove existence and uniqueness of the convolution/evolution
equation (5.95) we only need to show that A. generates a C0 semigroup.
We begin with the spectrum of A.. In turn, we begin with the point spectrum
(eigenvalues).
The point spectrum of A. consists of such that
x D .Ac C LC /x C LAY;
Y D C x C AY:
Suppose
Y D 0:
Then
Cx D 0
and hence
x D Ac x:
But then
x.s/ D es C x D 0:
Hence Y cannot be zero.
Suppose
x D 0:
Then we must have
LAY D 0
and the second equation yields
Y D0
and hence neither x nor Y can be zero.
Next, the first equation can be rewritten as:
x Ac x D LY
and can have a solution even if is not in the resolvent set of Ac , and the second as
Y AY D C x;
where
O
C x D L./Y;
so that
O
Y AY D C x D L./Y:
Or
O
Y AY L./Y D0
and thus is an aeroelastic mode.
We can say more.
Theorem 5.29. The point spectrum of the system stability operator A. is the set of
all aeroelastic modes.
Proof. Let be an aeroelastic mode, defined by
O
.I A L.//Y D 0; Y 0:
Or
O
.I A/Y D L./Y:
O
Recall that L./ is defined and analytic in the whole plane except the
negative real axis.
Let us next consider the equation in ,
x Ac x D LY I
this does not require that be in the resolvent set of Ac . Then
O
C x D L./Y:
By Lemma 5.28, this equation has a solution in . And
O
.I A/Y D L./Y:
Hence let

x
ZD :
Y
Then we verify that is in the point spectrum of A.:
x D Ac x C LY
D Ac x C L.AY C C x/;
Y D AY C C x: t
u
Let us next consider the resolvent.
Resolvent of A.
Theorem 5.30. Let denote the region in the complex plane
\
D fresolvent set of Ag
\
fComplement of the set of aeroelastic modesg fRe > 0g:
Then
is contained in the resolvent set of A.:
Moreover, for in
O
.I R.; A/L.// has a bounded inverse.
Define 1
O
R./ D I R.; A/L./ R.; A/:
Then the resolvent of A. denoted R.; A./ is given by

x
ZD ;
Y

R.; AC / .L LY C x/
R.; A./Z D ;

where
O
D R./.Y C CR.; Ac /x L./Y /:
Proof. Note first that includes a right-half-plane. For in , we note that
R.; A/ is compact, and hence so is
O
R.; A/L./:
Hence either

O
I R.; A/L./ Y D 0; Y 0;
or

O
I R.; A/L./
has a bounded inverse. But
O
.I R.; A/L.//Y D0
would imply that

O
.I A L.//Y D 0;
or is in the point spectrum of A.: Hence
O
.I R.; A/L.//
has a bounded inverse for in . Hence so does

O
I A L./
and let us denote it R./. Then

1
O
R ./ D I R.; A/L./ R.; A/
and is analytic in . To find the resolvent of A. we need to solve:
.I A./Z D Zg
for given

xg
Zg D :
Yg
Let

x
ZD :
Y
Then
x Ac x L.AY C C x/ D xg ;
Y AY C x D Yg :
Because Re > 0; is in the resolvent set of Ac . Hence
x D R.; Ac /.LY LYg C xg /;
O
CR.; Ac /L D L./:
Hence
O
C x D CR.; Ac /xg C L./Y O
L./Y g;
O
Y AY L./Y O
D Yg C CR.; Ac /xg L./Y g:
Hence

O
Y D R./ Yg C CR.; Ac /xg L./Y g ;
x D R.; Ac /.LY LYg C xg /;
which then defines R.; A./Z for in . t

u
Remarks: The resolvent above not being compact implies that the resolvent set is
properly contained in the complement of the point spectrum.
Note also that

0 O
PR.; A./ D R./.I L.//Y:
Y
Finally let us consider the semigroup generated by A..
Semigroup Generated by A
.
Rather than estimating the resolvent, we simply construct the semigroup explicitly.
First let us note that
O
jjR.; A/L./jj !0 as Re ! 1
and hence
O
jjR.; A/L./jj <1
in a right-half-plane contained in . Hence we have the Neumann expansion
1 1
X k
O
I R.; A/L./ DI C O
R.; A/L./ ;
kD1
where Z 1
O
R.; A/L./Y D et J.t/Y dt
0
and Z t
J.t/Y D S.t /./;
0
O
where . : / is the inverse Laplace transform of L./:
Hence it follows that
1
O
I R.; A/L./ D I C rO ./;
Z 1
rO ./ D et r.t/dt;
0
where
1
O
rO ./ D I R.; A/L./ I
1
O
D I R.; A/L./ O
I I R.; A/L./
O
D R./L./
and hence Z 1
R./Y D et W .t/Y dt
0
and
Z t
W .t/Y D S.t/Y C S.t /r./Y d;
0
R./Y D R.; A/Y C R.; A/Or ./Y:
The function W . : / was obtained by a contour integration technique in [60, 74].

Here we have obtained it in a more constructive way that also displays explicitly the
nature of the function. In particular
W .0/ D I
and W . : / is absolutely continuous on the domain of A. In fact
WP .t/Y D AW .t/Y C r.t/Y:
Armed with this function we can now proceed to construct the semigroup, by simply
taking the inverse Laplace transform of the resolvent. Thus, given

x
ZD ;
Y
where x is in the domain of Ac and Y in the domain of A, define for t 0:

Z t
Y .t/ D W .t/Y C W .t /.CS c ./x L./Y /d
0
and
0 Z t 1
@Sc .t/x C Sc .t /LYP ./d A
Z.t/ D Sz .t/Z D 0 ;
Y .t/
where we note that the derivative is well defined for x in the domain of Ac and Y in
the domain of A.
Thus defined, the Laplace transform of the derivative
Z 1
et SPz .t/Zdt D R.; A./Z Z D A.R.; A./Z;
0
which by our construction

Z 1 Z 1
t
D A. e Sz .t/Zdt D et A.Sz .t/Zdt ;
0 0
or
P
S.t/Z D A.S.t/Z;
which is enough to prove that A. generates the C0 semigroup Sz . : /:
And hence we have proved the state space representation
P
Z.t/ D A.Z.t/;
Y .t/ D PZ.t/: t
u
Time Domain Airfoil Dynamics: The Sonic Case
Next we consider the case M D 1 which differs radically from the case M D 0:
Here we follow [71]. We begin by evaluating the coefficients wij defined as
before.
k p
e
w11 D 4 p C erf k ;
k
p
erf k ek
w12 D p ;
k2 k3
p !
1 p k k
w21 D 2 erf k C 2e 1 ;
k
" p r #
2 3 k k 1 2 3
w22 D 2 C 3 erf k C 2e :
3 k k k2 k3
Note that in contrast with the case M D 0, there are no polynomials in k. In fact
wij .k/ does not converge to a finite limit as k goes to zero.
The role of the Theodorsen function is now played by the erf function
Z z
2
et dt
2
erfz D p
0
p
and erf z has a logarithmic singularity along the negative real axis and hence so
does wij . In fact:
p 1
erf s 2 X
p D p .1/k s k =.2k C 1/ entire function;
s kD0
Z 1
p
erf s D est ..t/ f .t//dt;
0
where
f .t/ D 0 0<t <1

1 . p
D 1 .t .t 1// ; t > 1:

The Laplace domain equations are
O s/ C 2 S .;
2 m h.; O O s/000
s/ C EI h.;
O
D L.; s/

k O s/ C .kw12 C .1 ka/w11 /O .; s/
D bU 21 w11 h.;
b
k p
k e O s/
D bU 21 4 p C erf k h.;
b k
p ! k ! #
erf k ek e p
C p C .1 ka/4 p C erf k O
.; s/ ;
k k k
O
2 I .; O s/ GJ .;
s/ C S 2 h.; O s/00 D MO .; s/

k O s/
D b 2 U1
2
.w21 aw11 /h.;
b
i
C.kw22 C .1 ak/w21 akw12 a.1 ak/w11 /O .; s/
" p !
k 1 p k
D b 2 2
U1 erf k C 2ek 1
b k2
p p
ek O 2 3
a4 p C erf k h.; s/ C k 2 C 3 erf k
k 3 k
r p !
k 1 2 3 1 p k
C 2ek C .1 ak/ 2 erf k C 2ek 1
k k2 k3 k
p k ! #
erf k ek e p
a p 4a.1 ak/ p C erf k O
.; s/ :
k k3 k
Let us next isolate the non-circulatory terms by taking the nonzero limits as Re
goes to infinity.
O
M 2 x.; s/ C Ax.;
O s/ D F ./x.;
O s/;
where

F11 ./ F12 ./
F ./ D ;
F21 ./ F22 ./
where
p
ek
F11 ./ D 4U1
2
k p C erf k ;
k
p ! k !
erf k ek e p
F12 ./ D bU 12
p C 4.1 ka/ p C erf k ;
k k k
" p ! k !#
k 1 p k e p
F21 ./ D bU 12
erf k C 2ek 1 4a p C erf k ;
b k2 k
p r
2 3 k k 1 2 3
F22 ./ D b U1 k
2 2
2 C 3 erf k C 2e
3 k k k2 k3
p ! p
1 p k erf k ek
k
C .1 ak/ 2 erf k C 2e 1 a p
k k k3
k p
e
4a.1 ak/ p C erf k :
k
Hence
F ./
lim D 0;
!1 2
0 1
F ./ 4U1
2
4bU 2
1a
lim D@ 4 A;
!1 k 0 b 2 U1
2
C 4a2
3
0 1
4U1
2
4bU 2
1a 0 4bU1 2
lim .F ./ k/ @ 4 A D :
!1 0 b 2 U1
2
C 4a2 0 4ab 2 U1 2
3
Having evaluated the non-circulatory terms we can formulate the equations in the
Laplace domain as
2 Mx./
O C Dx./
O C Kx./ C Ax./
O D FO ./x.;
O :/;
where
0 1
4bU1 4b2 U1 a
DD@ 4 A;
0 b 2 U1 2
C 4a2
3

0 4bU1 2
KD ;
0 4ab 2 U1 2
0 1
4U1 2
4bU 2
1a 0 4bU1 2
O
F ./ D F ./ k @ 4 A
0 b 2 U1
2
C 4a2 0 4ab 2 U1 2
3
Z 1
D et L1 .t/dt Re > a :
0
Aeroelastic Modes
As before
D./ D P e À./t Q;
d.1; ; U / D det D./:
The aeroelastic modes are the zeros fk g:
d.1; k ; Uk / D 0:
They are again confined to a finite strip. The corresponding mode shape vector is
given by

xl;k
Yk D ;
x2;k
x2;k D k x1;k ;
D.k /zk D 0; zk 0:
Assume D.k / is of rank 2
yk .s/ D esA.k / Qzk ;

x1;k .s/ D T26 yk .s/ 0 < s < `:
We can now go on, as in the case M D 0, to define the generalized resolvent R./
and the modal representation.
The big difference now, unlike M D 0, is that
YO ./ D R./Y .0/:
We can proceed as in the general case M > 0; treated below.

The unique feature here is that zero is not a mode; indeed D./ is not defined at
zero. There is no continuity at zero.
The General Case: 0 < M < 1I D 0
We now get back to the general case 0 < M < 1 having treated separately the cases
M D 0 and M D 1 where we exploited the luxury of explicit solution of the Possio
equation. We continue with the notation as in Sect. 5.3 and the normalization there,
so we may take b D 1; with the normalized frequency k D b=U .
The Possio equation for 0 < M < 1 in the abstract form is given by:
p
2 O 1 M2 O
wO a .; : / D A.; :/ C PH.B.k/ B.1//A.; :/ (5.106)
M M
and we simply assume that it has a unique solution, and the solution is then given by
p !1
O 1 M2 2
A.; :/ D I C PH.B.k/ B.1// wO a .; : /: (5.107)
M M
(We should recall here that we do have in Sect. 5.4 a constructive solution for
small , which happens to be the region of major interest in practice.)
We define the coefficients wij ./ as in Sect. 5.4 (5.3), which are now recognized
as functions of , and this is all we need from the solution to the Possio equation for
all M.
Note that for every M:
wij ./ D wij ./;
implying that the aeroelastic modes come in conjugate pairs. The main
difference for M > 0 from M D 0 is that from (5.72) we have that
lim Re ! 1
Z 1
2
wij D d D 0; i j;
M 1
4
w11 D ;
M
Z 1
2 4
w22 D 2 d D :
M 1 3M
The implication is that in the breakdown wi i ./ D polynomial in C

non-polynomial term the polynomial term is a constant (of zero degree), unlike
the case for M D 0, where it is of degree 1. The polynomial terms are called
the non-circulatory terms in [6]. To see the implication of this let us consider the
Laplace transform version of the structure dynamics:
O s/ C 2 S .;
2 m h.; O O s/0000 D L.;
s/ C EI h.; O s/

k O
O s/ C .kw12 C .1 ka/w11 /.;
D bU 21 w11 h.; s/ :
b
Or, collecting the non-circulatory terms into the structure state terms:

O s/ C 2 S .;
O 4 O s/ ab .;
O
2
m h.; s/ C bU1 .h.; s//
M
4 O O s/0000 D bU 2 k wM 11 h.;

O s/
C bU 21 .; s/ C EI h.; 1
M b
O
C .kw12 C .1 ka/wM 11 /.; s/;
where
4
wM 11 ./ D w11 ./ and ! 0 as Re ! 1:
M
Similarly
O
2 I .; O s/ GJ O .; s/00
s/ C S 2 h.;

O k O s/
D M .; s/ D b U12 2
.w21 aw11 /h.;
b
i
O
C.kw22 C .1 ak/w21 akw12 a.1 ak/w11 /.; s/ :
Hence collecting the non-circulatory terms into the structure side:

O 2O 4ab O 4ab O
I .; s/CS h.; s/ C bU1
2
h.; s/ C bU 1 h.; s/
M M

4 4 2 O 4 O O
C bU 1 a2 b 2 C b .; s/ bU 21 ab .; s/GJ .; s/00
M 3M M

k O
D b U1
2 2
.w21 awM 11 /h.; s/ C .k wM 22 C .1 ak/w21 akw12
b
i
a.1 ak/wM 11 / O .; s/ ;
where
4
wM 22 ./ D w22 ./
3M
and goes to zero as Re ! 1.
The main thing is now to note that the structure inertia matrix is not changed
unlike in the case where M D 0. But damping and stiffness terms are added which
are reflected in the abstract version as bounded operators. Thus collecting terms
containing 2 and we can now write the structure with the aerodynamic loading
Laplace domain equation:
2 Mx./
O C Dx./
O C Kx./
O C Ax./
O D F ./x.;
O : /; (5.108)
where

m s
MD ;
s I
0 1
4 4
B bU1 ab bU1 C
B M M C
DDB C ;
@ 4 4 4 A
ab bU1 2 2
bU 1 a b C b 2
M M 3M
0 1
4 2
0 bU1
B M C
KD@ A; (5.109)
4
0 bU1 ab
2
M
where MDK are constant 2 2 matrices and FO ./ is the Laplace transform:
0 1
k
bU 2
M
w h C .kw C .1 ka/ M
w /
B 1 C
11 12 11
B
b C
h B C
FO ./ DB k C (5.110)
Bb 2 U1 2
.w21 awM 11 /h C
@ b A
C .k wM 22 C .1 ak/w21 akw12 a.1 ak/wM 11 /
and A is the structure differential operator as defined in Chap. 2.
Next we take inverse Laplace transforms
Z t
Mx.t/
R C Dx.t/ P C Kx.t/ C Ax.t/ D F ./x.t /d; (5.111)
0
where F . : / is the inverse Laplace transform:

Z 1
FO ./ D et F .t/dt Re > 0: (5.112)
0
Next with

x.t/
Y .t/ D ;
P
x.t/
we obtain the convolution/evolution equation in H:

Z t
YP .t/ D AY .t/ C L.t /Y ./d;
0
where

0 0
L.t/ D 1 4 4 matrix function ; (5.113)
M F .t/ 0

0 I
AD l 1 (5.114)
M .A C K/ M D
with the energy space H as defined in Chap. 2.

Here as before, A generates a C0 semigroup with compact resolvent; and the
analysis is similar to that for M D 1: M D 0 is a special case, whereas M D 1 is
typical of M > 0.
Taking Laplace transforms in
Z t
YP .t/ D AY .t/ C L.t /Y ./d;
0
we have:
O
.I A L.// YO ./ D Y .0/:
Thus we define the aeroelastic modes as the zeros:

O
I A L./ Y D0
with the generalized resolvent defined again by

1
R./ D I ALO ./ :
Note that the change from the M D 0 case is that (as in the case M D 1/
O
L./ O
is replaced by L./ O
(the two L./ are not the same functions, of course!)
which makes the problem less complex in that now

O
I A L./ YO ./ D Y:
Or,
YO ./ D R./Y:
The aeroelastic mode fk g are of course obtained as the zeros of the
determinant
d.M; ; U / D det D.M; ; U /;

D.M; ; U / D P e A./ Q
with A./ just as defined in Sect. 5.4 except that the wij are no longer expressible
explicitly.
We assume that AOi .; M / is the solution of the Possio equation (5.2) special-
ized to
p !1
1 M2 2
AOi .; : / D I C PH.B.k/ B.1// fi ;
M M
where
fi .s/ D 1; b < s < b

D s; b < s < b
and note that the solution is defined for all omitting the modes k and the negative
half-line. Moreover the solution is analytic in this open set and hence so are the wij
and so is A./.
Note that the roots again come in conjugate pairs, if complex. There can be real
roots for some speeds; zero is a root for some speeds.
For the existence of the roots for U positive, we need to draw on the
analyticity properties of D.M; ; U /. These are determined by those of fwij ./g
which in turn are dependent on those of the solution to the Possio equation.
However, we no longer have the luxury of an explicit solution. And thus we have to
go back to (5.49), the abstract form of LDP for values other than Rek > 0, assumed
there. The Hilbert transform operator H is what causes trouble. The right-hand side
of (5.49) is
b
PHPB.k/PA C T PH.I P/B.k/PA; k D ;
U
where the first term is an entire function of and the second term
WA D T PB.I P/B.k/PA;
we now show is analytic except for the logarithmic singularity along the negative
half-line, of the same kind as we have seen for M D 0. Thus we have

kR.k/A. : /
T PH.I P/B.k/PA D T PH.I P/ p
1 M2
Z 1
kR.ks/A. : /a.M; s/ds ;
2
where the second term is kind of an average of the first term and inherits the
analyticity properties of the first term. Denoting the first term by g. : / we have:
s p
Z 1 Z b
k bx ek 2b C
g.x/ D p d ek.b/ A./d;
bCx 0 . C b x/ b
jxj 0; take b D 1;
0 . C s/
dF
D sF .z/ .K0 .z/ C K1 .z// Re z > 0
dz
and because .K0 .z/ C K1 .z// is analytic except for z 0; F .z/ can be extended
analytically to the left-half-plane, but will have an essential singularitybranch
cutalong the negative axis exactly as is the case for M D 0.
Thus the operator PH.B.k/B.1// in the Possio equation can be extended to be
analytic in the whole plane except for the branch cut along the negative axis, and this
property then extends to the solution, and hence to wij , and finally to D.M; ; U /
and d.M; ; U /.
Modes and Mode Shapes
Hence in particular the zeros of d.M; ; U /the aeroelastic modesare countable.

Further more they are limited to a finite strip (we call it the spectral strip) because
the contribution of the circulatory terms:
O
L./ !0 as Re ! 1 (5.115)
and hence we may imitate the arguments for M D 0 (cf. Sect. 5.3) where really only
this argument is needed.
Again define xk by:
D.k /xk D P e A.k /` Q xk D 0; xk 0; (5.116)
D./ being a 3 3 matrix the null space of the matrix is at most 2, and we assume
it to be 1, as is the case for U D 0. Similarly we assume that the multiplicity is also
1, so that
D 0 .k /xk 0: (5.117)
Lemma 5.31. Let
D.k /? zk D 0 zk 0:
Then
zk ; xk 0: (5.118)
Proof. The 3-space has the orthogonal decomposition:
null space of D.k /? C range space of D.k /:

Hence if
zk ; xk D 0;
we must have that xk is in the range space of D.k /. Or,
xk D D.k /h for h 0:
Hence
D.k /2 h D 0:
Taking the derivative with respect to , we have
D 0 .k /xk D 0;
which contradicts (5.11). t

u
The corresponding Yk satisfies
O k //Yk D 0:
.k I A L.
Or, with

xl k
Yk D ;
x2 k
we have
k x1;k D x2;k ;
2k Mx1;k C Dx1;k C Kx1;k C Ax1;k FO .k /x1;k D 0: (5.119)
Hence
x2;k D k x1;k : (5.120)
Define the 2 6 matrix:
100000
T26 D : (5.121)
000100
Then
x1;k .s/ D T26 eA.k /s Qx k ; 0 < s < `k
and xk is defined by (5.10).
We may normalize xk by:
" Z ! #
`
A? .k /s
Mx1k ; xlk D 1I Q? e T26 ? MT26 eA.k /s Qds xk ; xk D 1: (5.122)
0
This determines Yk completely.

T26 eA.k /s Qxk
Yk .s/ D ; 0 < s < `:
k T26 eA.k /s Qxk
In fact we can express it as a linear bounded transformation on C 3 into H:

fl
Lk x D I f1 .s/ D T26 eA.k /s Qx; 0 < s < `;
k fl
Yk D Lk xk ;
and
Yk ; Yk D Lk ? Lk xk ; xk D j k j2 :
The energy space norm of Yk is not equal
to 1.
O k / is the dimension of the
The dimension of the null space of k I A L.
null space of P e A.k /` Q and is 1, with multiplicity 1.
We note that in as much as the conjugates
N
D./ N
D D./I N
A./ N
D A./;
the conjugate YNk is the eigenvector corresponding to the conjugate root N k .
Yk YNk
R./Yk D I R./YNk D
k N k
corresponding to
xk
D./1 xk D :
k
The case where zero is an aeroelastic mode is treated separately below, as it is
involved in defining the divergence speed.
We note that zero is an aeroelastic mode only for a sequence of speeds. The
functions (for M 1) are defined as the limit from above (Im nonzero) or from
the right (Re > 0).
And
d.0; 0C; 0C/ D 1:
Generalized Resolvent
The generalized resolvent is defined for excepting the aeroelastic modes and the
negative half-line:
1
O
R./ D I A L./
just as in the case M D 0 in Sect. 5.5.

O 1 xl;g
R./Yg D .I A L.// Yg W Yg D :
x2;g
Let

hg
D Mx2;g C .D C K/x1;g :
g
Then (as in Sect. 5.5)

hg .s/ g .s/
yg .s/ D col 0; 0; 0; ; 0; ;
EI GJ
Z s
y.; s/ D esA./ Qz./ C e.s /A./ yg ./d;
0
Z `
P .eÀ./ Qz./ C vg .// D 0I vg ./ D e.` /A./ yg ./d:
0
Or,
D./z./ D P vg ./I D./ D P e À./ Q:
Hence
z./ D D./1 P vg ./;
which is defined where wij are defined excepting the modes fk g and
Z `
y.; s/ D esA./ QD./1 P e.` /A./ yg ./d
0
Z s
C e.s /A./ yg ./d; (5.123)
0
and finally:

xl ./
R./Yg D ;
x2 ./

h.; : /
x1 ./ D T26 y.; : / D ;
.; : /
x2 ./ D x1 ./ x1;g :
Greens Function Representation
We proceed first to deduce the Greens function for the generalized resolvent. For
this purpose it is convenient to use the notation:

h. : / . : /
C6 ./Y D col 0; 0; 0; ; 0; ;
EI GJ
where

h. : / xl
D Mx2 C .D C K/x1 I Y D :
. : / x2
Then R./Y is the function:

Z `
x1 .; s/
D R.; s; /Y ./d; 0 < s < `; (5.124)
x2 .; s/ 0
where the kernel is deduced from

Z `
x1 .; s/ D T26 esA./ QD./1 P e.` /A./ C6 ./Y .s/ds
0
Z s
C e.s /A./ C6 ./Y ./dx2 .; s/ D x1 .; s/ x1 .s/:
0
The main thing to note is that the kernel is square integrable:

Z ` Z `
jjR.; s; /jj2 dsd < 1:
0 0
In particular this implies that the eigenvalues are square integrable:

X 1
< 1;
j k j2
where the summation is over all roots, the index k being chosen such that the
imaginary part
j !kC1 j j !k j:
Our convention for the modes is:
k D k C i !k ; !k > 0
and the conjugate is of course always a mode as well, if complex. And in particular
X 1
< 1 nonzero k : (5.125)
jk j2
Laplace Inversion Formula
Equation (5.9) shows that the generalized resolvent has simple poles at the
aeroelastic modes confined to a finite strip. We can use this to determine the inverse
Laplace transform by the inversion formula
Z L
1
W .t/Y0 D lim e. Ci !/t R. C i !/Y0 d!; (5.126)
L!1 2 i L
where is to the right of the strip containing the point spectrum,

x1 .t/
W .t/Y0 D ;
x2 .t/

x1;0
Y0 D ;
x2;0
W .0/Y0 D Y0 in D.A/;
and
d
x2 .t/ D x1 .t/;
dt

h0 ./
D Mx2;0 C .D C K/x1;0 ;
0 ./

h0 .; s/ 0 .; s/
y0 .; s/ D col 0; 0; 0; ; 0; ;
EI GJ
Z L
1
x1 .t; s/ D lim et T26
L!1 2 i L
Z `
esA./ QD./1 P e.`/A./ y0 .; /d
0
Z s
C e .s/A./
y0 .; /d id!jD Ci ! 0 < s < `;
0
where we have seen that D./1 has simple poles at D k with residue
I
1
Dk1 D D./1 d D k C rei :
2 i
(The convergence of the infinite series is assumed here for the moment; the proof is
given below.)
Hence we deform the contour as before following [33,44] where the functions are
meromorphic but with a branch cut along the negative axis.
Thus we have
1 Z ` !
X
1
x1 .t; s/ D T26 e sA.k /
QD k P e .` /A.k /
y0 ./d ek t C x1;R .t; s/;
kD1 0
x2 .t; s/ D xP 1 .t; s/; Y0 in D.A/; (5.127)
where the second term x1;R .t; : / is the evanescent term we have already treated
in Sect. 5.5.
The evanescent term can be expressed:
Z 1
U
x1;R .t; : / D eRU=bt RJ .R/Yg dR; (5.128)
b 0
where RJ . : / is the jump across the line of singularity: 1 < 0 and for all
M 0 including M D 0.
Note the slight difference in notation from Sect. 5.5. Thus we define
1
RJ .r/Y D lim R.r C i !/ R.r i !/Y; ! 0;
2 i !!0
where the subscript J is for jump and

x
Y D l
x2
and R./Y is the function defined by

" Z
sA./ `
.R./Y /.s/ D T26 e QD./1 P e.` /A./ y./d
0
Z s
C e.s /A./ y./d ; 0 < s < `;
0

h.; s/ .; s/
y.s/ D col 0; 0; 0; ; 0; ;
EI GJ

h./
D Mx2 C .D C K/x1
./
defined and analytic for k , and omitting 0.

" Z
sA./ `
.R./Y /.s/ D T26 e QD./1 P e.` /A./ y./d
0
Z s
C e.s /A./ y./d ; 0<s<`
0
is defined and analytic for k ; and omitting < 0 which is a logarithmic

branch cut.
Moreover the jump R.r C i !/ R.r i !/Y; ! 0 converges strongly as
we approach the line of singularity as ! goes to zero for each r > 0. And the limit
RJ .r/Y goes to zero as r goes to zero.
Hence finally:

x1 .t/ x1;0
W .t/Y0 D I Y0 D ;
x2 .t/ x2;0
1 Z !
X `
x1 .t; s/ D T26 e sA.k /
QD 1
k P e .` /A.k /
yg ./d ek t
kD1 0
Z 1
C ert .RJ .r/Y0 /.s/dr;
0
x2 .t; s/ D xP 1 .t; s/; t > 0; 0 < s < `: (5.129)
Again: let us take the Laplace transform
Z 1
xO 1 .; : / D et x1 .t; : /dt
0
Z !
X `
1
D T26 e sA.k /
QD 1
k P e .` /A.k /
yg ./d
0 k
Z 1
1 .RJ .r/Yg /.s/
dr; 0 < s < `;
0 Cr
where the sum is over all roots here and below.

Z 1
et x2 .t; : /dt D xO 2 ./ D xO 1 ./ x1;g :
0
Hence we have the representation R./Y is the function

0P R ` .` /A. / 1
T26 esA.k / QD 1
1
k k P 0 e
k y ./d
g
B C
B R C
B C 1 01 .RJ .r/Y /.s/
dr C
B Cr C
DB R C;
BP 1 ` C
B T26 e sA.k /
QD k P 0 e .` /A.k /
yg ./d k C
@ A
R
1 1 .RJ .r/Y /.s/
C 0 Cr
dr x1;g .s/
0 < s < `: (5.130)
Let us define the linear bounded operator on H into H by the contour integral
Z
Pk Y D R./dI .k / D f D k C rei ; 0 2g
.k /
Z 2
1
D R k C rei Y d for 0 < r;
0 2
small enough to include no other poles. Pk Y is the function

0 R ` .` /A. / 1
T26 esA.k / QD 1
k P 0 e
k y ./d
g
@ R ` .` /A. / A ; 0 < s < `: (5.131)
k T26 esA.k / QD 1
k P 0 e k y ./d
g
And more generally:

Z
f ./R./d D f .k /Pk :
.k /

xl
For Y D ;
x2
0 Z 1 1
l RJ .r/Y
X Pk Y dr
B Cr C
R./Y D CB Z
0 C:
k @ 1 1
RJ .r/Y A
C dr x1 .s/
0 Cr
Now

O
X I A L./ PkY
O
Y D I A L./ R./Y
k
0 Z 1 1
1 RJ .r/Y
B dr
Cr C
O
C I A L./ B Z
0 C:
@ 1 1
RJ .r/Y A
dr x1 .s/
0 Cr
H
Then the contour integral .1=2 i /Yd; D k C rei yields

O k / Pk Y:
0 D k I A L.

Hence Pk Y is the slant (not orthogonal because Pk Pk ) projection on the null
space of k I A L.O k / which is of dimension 1 and is spanned by Yk . Hence

the dimension of the range of Pk is 1. Or
Pk Y; Yk
Pk Y D k Yk D Yk ;
Yk ; Yk
Pk Yk D Yk ;
Z 1 0 1
l RJ .r/
X Pk Y B Y dr C
R./Y D CB Z 01 C r C:
k @ l rRJ .r/Y dr
A
k
0
Note that this runs over all roots: the conjugates are treated as separate roots.
Next let Q denote the linear bounded operator
0 R1 1
1
0 RJ .r/Y dr
QY D @ R A: (5.132)
1 1
0 rRJ .r/Y dr
The range of Q are again the evanescent states as in the incompressible case, stable
for all U . From (6.9)
Pk Y
R./Pk Y D :
k
Now
lim R./Y D Y:
Re!1
Using which in particular we have the representation:

X
Y D Pk Y C QY (5.133)
consistent with (5.11).

Lemma 5.32.
Pk Pj D Pj Pk D 0; j k
D Pk j D k;
Q2 D Q;
Pk Q D 0 D QP k :
The Pk as well as Q are idempotents, but not projections.

Proof. Here we essentially follow the theory of pseudoresolvents in [10, p. 208
et seq].
Thus we note that for k j ,
Z Z
1
Pk Pj x D R./R./dd;
.2 i /2 .k / .j /
where h i
O
R./ R./ D . /I L./ O
L./ R./R./
and hence
!
R./ R./ O
L./ O
L./
R./R./ D C R./R./ (5.134)

and
Z Z
1 R./
dd D 0; k j D Pk ; k D j;
.2 i /2 .k / .j /
Z Z
1 R./
dd D 0; k j D Pk ; k D j;
.2 i /2 .k / .j /
Z Z !
1 O
L./ O
L./
R./R./dd
.2 i /2 .k / .j /
O k / L.
L. O j/
D Pk Pj k j:
j k
Or
O k / L.
L. O j/
Pk Pj D Pk Pj k j:
j k
Hence
Pk Pj D 0 k j:
Also by a similar integration procedure:
Pk Qx D 0 D QP k :
t
u
Next, using (5.22)
0 1
X
Pk x D Pk @ Pj x C Qx A D Pk Pk x;
j
0 1
X
Qx D Q @ Pj x C Qx A D QQx:
j
Hence the Pk and Q are idempotents but not necessarily orthogonal

projections because not self-adjoint.
To proceed further we need to go on to the operator adjoints.
Adjoint Theory
Let us consider now the properties of the adjoint operator: (Note that the adjoint
now is with respect to the energy inner product.) We calculate
O
.I A L.// ?
ZD0
by
O
Z; .I A L.//Y D0 for every Y:
Or
Az1 ; x1 x2
h i
C z2 ; 2 Mx1 C Dx1 C Kx1 C Ax1 C Dx1 C Kx1 FO ./x1 D 0
with
h
x1 D I

the second term

h i
O
D h2 ; 2 mh C 2 S C EI h0000 L.; :/
h i
C 2 ; 2 I C S 2 h GJ 00 MO .; :/ ;
where

O k
L.; s/ D bU1 2
w11 h.s/ C .kw12 C .1 ka/w11 /.s/ MO .; s/
b

k
D b 2 U1 2
.w21 aw11 /h.s/
b
C .kw22 C .1 ak/w21 akw12 a.1 ak/w11 / .s/ :
As in the case of Yk we can calculate

z k
zk D 1 ;
z2 k
Q
z2k .s/ D T26 eA.k /s Qzk 0 < s < `; (5.135)
Q k /zk D 0;
D. zk 0; (5.136)
Q Q
D./ D P e A./` ;
0 1
0 100 0 0
B0 010 0C
B 0 C
B C
Q B0 001 0 0C
A./ DB C;
Bw1 0 0 0 w3 0C
B C
@0 000 0 1A
w2 0 0 0 w4 0

N 2k Mz2k C Az2k D K? C FO .k / N k D? z2k ; (5.137)

N k z1k D I A1 K? C A1 F.
O k /? z2k : (5.138)
Q
This time z1k is determined from z2k . The zeros of det D./ are the conjugates of
those of det D./.
Recall that we normalized Yk so that
jjYk jj2 D x1k ; Ax 1k C jk j2
and asymptotically in k,
x1k ; Ax1k
!1
jk j2
and
jjYk jj2
! 2:
jk j2
We would want a similar normalization for Zk . We now normalize so that
Mz2k ; z2k D jk j2 :
Then " Z ! #
`
Q k /? s Q k /s
? A.
Q e .T26 / MT26 e A.
Qds zk ; zk D jk j2 ;
0
which is then how zk has to be normalized.

Then we calculate
1 h 1 ? 1 O
i
O k /? z2k :
z1k ; Az1k D I C A K A F . k / ?
z2k ; A C K ?
F.
jk j2
Now

N 2k Mz2k C Az2k D K? C F.
O k / N k D z2k
using which we have:

,
1 ?
z1k ; Az1k D 2k z2k ; Mz2k C D z2k Mz2k ; z2k
N k

1 ? 1 O 1 ?
k A K A F .k / z2k ; Mz2k C
2 ?
D z2k
N k
,
Mz2k ; z2k :
Hence asymptotically in k
z1k ; Az1k 2k

j ! 1; noting that ! 1:
jk j2 jk j2
Hence it follows in particular that

.
Zk ; Zk jk j2 ! 2:
Second Normalization
We renormalize Zk without changing norms. We have seen that Yk ; Zk

0, therefore we define
ZQ k D k Zk ; where jk j D 1; and

,q

Q
Yk ; Zk ZQ k ; ZQ k Yk ; Yk D 1: (5.139)
We simply keep using Zk again in this renormalized version. Again we can write
zk D LQ k zk ;
where LQ k is a linear bounded transformation on C 3 into H given by

Q g1 Q
Lk z D I g2 .s/ D T26 eA.k /s Qz; 0 < s < `;
g2

N k g1 D g2 A1 K ? A1 FO .k /? g2 ;
LQ k MLQ k g2 ; g2 D jk j2 :
Note that
Pk? Z; Pj Y D 0 j k:
Hence
Yk ; Zj D 0 for kj
and similarly:
QY; Zj D 0 for everyj:
The sequences fYk g; fZk g are biorthogonal. Note the set of eigenvalues is the same.
Then we have the representations:
X
Y D k Yk C QY (5.140)
X
D k Zk C Q Y: (5.141)
Problem: Given any two sequences: Can we always biorthogonalize them?

We note that the fPk Y g do not span the space. In particular
Pk QY D 0 D QP k Y:
So then the question is: Given Pk Y D k Yk , how do we determine k ?

We use the representation:
Y; Zj D Pj Y; Zj
D j Yj ; Zj :
If
Yj ; Zj D 0 for any j; then for every Y;

1
X
Yj ; Y D Yj ; k zk C Yj ; Q Y D 0;
kD1
which is a contradiction. Hence
Y; Zk
k D :
Yk ; Zk
Next
p
jYk ; Zk j jjY jj Zk ; Zk :
Next
p p
Yk ; Zk D .1/ Yk ; Yk Zk ; Zk :
Hence
jjY jj
j k jjjYk jj p ;
Yk ; Yk
which asymptotically
jjY jj
:
jk j
Because
1
X 1
< 1;
jk j2
kD1
it follows that
1
X
jk j2 jjYk jj2 < 1: (5.142)
kD1
And similarly that

1
X
jk j2 jjZk jj2 < 1: (5.143)
kD1
This is called the Riesz property of the biorthogonal sequence proved in a more
abstract way in [22].
Note that
X
Y; Y D k Nk Yk ; Zk : (5.144)
This is enough to prove the convergence of all the infinite sums considered above.
Next we can calculate the Zk in a different way. Thus
Ak Pk Y D k Pk Y:
Hence for any Z
Ak Pk Y; Z D k Pk Y; Z D Y; N k Pk Z:
Hence
Pk? Z D.Ak /
and

N k I A?k Pk Z D 0:
We define:
Pk? Zk D Zk :
To calculate Pk? we use

z x h
Pk Y; Z; Z D l I Y D l I C62 x D col 0; 0; 0; ; 0;
z2 x2 EI GJ
Z `
D T26 esA.k / .QD 1
k P /C6 Y .s/; .Az1 /.s/ ds
0
Z `
C k T26 esA.k / .QD 1
k P /C6 Y .s/; Mz2 .s/ ds
0
Z ` h i
D x1 .s/; .C6 /? .QD 1 sA.k /?
k P/ e .T26 /? .Az1 /.s/ ds
0
Z ` h i
C x2 .s/; N k .C6 /? .QD 1 sA.k /?
k P/ e .T26 /? Mz2 .s/ ds
0
Z ` h i
D Ax1 .s/; A1 .C6 / .QD 1
k P / ? sA.k /?
e .T 26 / ?
.Az1 /.s/ ds
0
Z ` h i
C Mx2 .s/; N k M1 .C6 /? .QD 1
k P / sA.k /?
e .T 26 / ?
Mz2 .s/ ds:
0
Hence
!
A1 .C6 / .QD l .T26 / .Azl /.s/
? sA.k / ?
k P/ e
Pk Z D ;
N k Ml .C6 / .QD l ? sA.k /?
k P/ e .T26 /? Mz2 .s/

QD 1
k P D P .Dk1 /? Q? ; : : : ; a 6 6 matrix;
I
1
.Dk1 /? D D ./1 d D N k C r ei :
2 i
Finally let
0 Z 1 1
1
B ert RJ .r/Y dr C

Q.t/Y D B
@1 Z 1
0 C
A (5.145)
rt
.r/e RJ .r/Y dr
0
so that in particular Q D Q.0/:

Furthermore, QY; Zk D 0 for every k. Hence the range of Qthe space of
evanescent statesis orthogonal to the modal space spanned by the sequence fZk g.
And similarly
Q? Z; Yk D 0:
Then finally the solution to the aeroelastic equation can be expressed:
X
Y .t/ D ek t Pk Y .0/ C Q.t/Y .0/; (5.146)
where
Y; Zk
Pk Y D Yk :
Yk ; Zk
Here the most intriguing part is the term that cannot be expressed in terms of modes!
Now we can return to the properties of the nonrational component: the evanescent
states:
Q.t/Y; t 0:
Lemma 5.33. For each U:
S.t/Y D W .t/.Y QY /; t 0
X
D ek t Pk Y
k
D W .t/Y Q.t/Y
defines a C0 semigroup, actually a group, over the subspace spanned by the

modes fYk g.
S.t/QY D QS.t/Y D 0
Q.t/Y is stable for all U:
Proof. Is immediate from the representation (5.29). t

u
Greens Function: Time Domain
From (5.35) or directly from (5.16)we have the Greens function representation for
the time domain solution:
W .t/Y .0/ D Y .t/
is the function: Z `
Y .t; y/ D W .t; y; s/Y .0; s/ds; (5.147)
0
where Z 1
et W .t; y; s/dt D R.; y; s/: (5.148)
0
Next we extend the state space theory for M > 0.
State Space Theory: M 0
In view of the fact that the convolution/evolution is different from that for M D 0:
Z t
YP .t/ D AY .t/ C L.t /Y ./d
0
(no derivative in the integral) we need to examine the changes needed in the state
space representation (5.2)
P
Z.t/ D A.Z.t/ Y .t/ D PZ.t/:
Thus we begin with the analogue of Theorem 5.22. The main question is the nature
of the function L. : /. We begin with:
2
AOi .0; :/ D p T fi ;
1 M2
r
2 1x
AO1 .0; x/ D p ;
1 M2 1 C x
2 p
AO2 .0; x/ D p 1 x2 :
1 M2
Hence
2
w11 .0/ D p ;
1 M2
2 4
wM 11 .0/ D p ;
1 M2 M
w22 .0/ D 0;
4
wM 22 .0/ D :
3M
Which is enough to show that L. : / here enjoys the same sort of properties as in
P M: Hence Theorem 5.22 holds in this case as well and so does the
the case for zero
definition of and that in turn of X.; of C and Ac , and Z denoting the elements
therein, as there. But now with

x.t/
Z.t/ D ;
Y .t/
we have:
P
x.t/ D Ac x.t/ C LY .t/;
YP .t/ D AY .t/ C C x.t/
and the representation is thus much simpler:

P Ac L
Z.t/ D A.Z.t/I A. D ;
C A
PZ.t/ D Y .t/: (5.149)
Let us examine the changes needed with this definition. The domain of A. remains
the same: i.e., Domain of Ac D Domain of A.
Spectrum of A.
The spectrum of A. consists of such that
x D Ac x C LY;
Y D C x C AY:
Or
x Ac x D LY:
Theorem 5.34. The point spectrum of A. consists of the aeroelastic modes.

Proof. For Re > 0; is in the resolvent set of Ac so that given any Y in H, we
have:
x./ D R.; Ac /LY

Z 1
D et Sc .t/LY dt:
0
Hence the function

Z 1
x.; s/ D et L.s C t/Y dt; 0s1
0
and hence Z 1
C x D x.; 0/ D O
et L.t/Y dt D L./Y:
0
Hence we have the representation:
Z s
O
x.; s/ D e L./Y
s
e.s / L./Y d; (5.150)
0
where
L.1/Y
x.; 1/ D :

O
But this defines an analytic function of in the region of analyticity of L./Y =
which is the whole plane omitting the branch cut along < 0: In other words we
have proved that given Y , we can find x./ in such that
x./ Ac x./ D LY:

Next we are given

x Ac x D LY
even if is not in the resolvent set of Ac , and thus not necessarily unique. Hence
O
Y AY D L./Y
if is in the spectrum of A.. But this means that is an aeroelastic mode.

Next suppose is an aeroelastic mode. Then
O
Y AY D L./Y:
But given Y we have shown that we can find x./ such that
x./ Ac x./ D LY:
Hence is in the spectrum of A.. t

u
Resolvent of A.
To find the resolvent we need to solve:

R xg
Z Az D ;
Yg
x Ac x LY D xg ;
Y AY C x D Yg : (5.151)
We show the following analogous to Theorem 5.24.

Theorem 5.35. Let denote the region in the complex plane: D
fresolvent set of Ag \ fcomplement of the point spectrum of A.g \ fRe > 0g.
Then is contained in the resolvent set of A.. Moreover, it has the representation

R.; Ac /.x C L/
. /Z D
R.; A

;
D R./.CR.; Ac /x C Y /;
which is simpler than the case for M D 0.

Proof. For in we can solve (5.40) as:

x R.; Ac /.xg C LY /
D :
Y R.; A/.C x C Yg /
But
O
C x D CR.; Ac /xg C L./Y
O
D CR.; Ac /xg C L./Y;
O
Y AY CR.; Ac /xg L./Y D Yg ;
O
Y AY L./Y D Yg C CR.; Ac /xg :
But is not an aeroelastic mode and hence
Y D R./.Yg C CR.; Ac /xg /;
where R./ is the generalized resolvent and in turn
LY D LR./.CR.; Ac /xg C Yg /:
Hence

x R.; Ac /.xg C L/
D ;
Y
D R./.CR.; Ac /xg C Yg /
as required. t
u
Semigroup Generated by A.
We show next that A. generates a C0 semigroup. The main step here is again to
evaluate the inverse Laplace transform of the generalized resolvent. We note that
O
jjR.; A/L./jj !0 as Re ! 1:
Hence we have that

O
jjR.; A/L./jj <1
in a right-half-plane and furthermore, also the Neumann expansion therein
1
X k
O
I R.; A/L./ 1
D O
R.; A/L./ ;
kD0
where
Z 1
O
R.; A/L./Y D et J1 .t/Y dt;
0
Z t
J1 .t/ D S.t /L./d;
0
whose Laplace transform is defined for Re > the growth bound of the semi-
group S .
Define
Z t
Jn .t/Y D J1 .t /Jn1 ./d; n 2;
0
whose Laplace transform is:

n
O
R.; A/L./ :
Let
1
O
rO ./ D I R.; A/L./ I;
which is defined for Re > the growth bound of the semigroup S. : /.

Then
1
O
rO ./ D I R.; A/L./ O
R.; A/L./ O
D R./L./:
Let
1
X
r.t/Y D Jn .t/Y; t 0:
nD1
Then the Laplace transform

Z 1 1
X k 1
et r.t/dt D O
R.; A/L./ O
D I R.; A/L./ I
0 kD1
D rO ./:
Next let
Z t
W .t/Y D S.t/Y C S.t /r./Y d t > 0: (5.152)
0

Z 1
et W .t/Y dt D R.; A/ .I C rO .// Y
0
1
D R.; A/ I R.; A/L./O Y
D R./Y:
Define
Rt
Sc .t/x C 0 Sc .t /L./d x
Sz .t/Z D ; ZD ;
.t/ Y
Z t
.t/ D W .t/Y C W .t /CS c ./ xd:
0
The Laplace transform is the resolvent of A., proving that it is the semigroup
generated by A.. Thus we have the state space representation:
P
Z.t/ D A.Z.t/;
Y .t/ D PZ.t/: (5.153)
This establishes in particular the existence and uniqueness of solution to the

aeroelastic equations. This result is essential for control theory treated in Chap. 8.
We can now go on to the main objective of the theory: flutter analysis next.
5.7 Flutter Analysis
Divergence Speed
We begin with the divergence speed (4.42) we have already seen in connection with
the steady-state or static solution in Chap. 4. Here we examine it from the dynamic
side as corresponding to zero frequency.
Thus here we consider U such that d.M; 0; U / D 0. Setting D 0 is equivalent
to setting all the time derivative terms to be zero.
Theorem 5.36.
p
d.M; 0; U / D cosh ` w4 ;
5.7 Flutter Analysis 231
where
b 2 .1 C 2a/ 2
w4 D p U :
GJ 1 M 2
Proof. We calculate wij .0/: Using:
O :/ D p 2
A.0; T wO a .0; :/;
1 M2
r
O 2 1x
A.0; :/ D p jxj < 1:
1M 2 1Cx
Hence
2
w11 .0/ D p ;
1 M2

w12 .0/ D p ;
1 M2

w21 .0/ D p ;
1 M2
w22 .0/ D 0:
From which we calculate the entries in the matrix A(0):
w1 D 0;
w3 D 0;
2
w2 D bU 2 p ;
1 M2
b 2 .1 C 2a/ 2
w4 D p U ;
GJ 1 M 2
which in turn yields

s
b 2 .1 C 2a/ 2
d.M; 0; U / D cosh ` p U : t
u
GJ 1 M 2
p
Hence d.M; 0; U / D 0 yields cosh ` w4 D 0:
Or s !
b 2 .1 C 2a/ 2
cos ` p U D 0:
GJ 1 M 2
Hence p

2 1=4 1 GJ
U D .2n C 1/ 1 M p ;
2b` .1 C 2a/
n positive integer and hence the divergence speed:
p
1=4 1 GJ
Ud D 1 M 2 p ; (5.154)
2b` .1 C 2a/
which of course checks our previous result based on the time-invariant solution
(5.42).
Remarks: A question of interest here (see below) is what happens as we increase

speed at the divergence speed. We want to calculate the nearest aeroelastic mode as
we increase the speed. We show that the derivative with respect to U is positive at
this point, thus as we increase speed the system becomes unstable, with a positive
aeroelastic mode.
Thus we want to calculate as a function of Ud C U such that
d.M; ; Ud C U / D 0:
Because
p
d.M; 0; U / D cosh ` w4 ;
we see that s
@d.M; 0; Ud / b 2 .1 C 2a/
D ` p :
@U GJ 1 M 2
However we run into the difficulty that @d.M; 0; Ud /=@ is not defined at D 0
singularity of the function at D 0! We hence use the perturbation formula: for
small
d.M; ; Ud / D det P e A./ Q

Z !
`
D det P e A.0/
QCP e A.0/.`s/
.A./ A.0//e A.0/s
Qds ;
0
where we may use the approximation for small for the wij ./ given in [21],
following the expansion in Sect. 5.4.
Here, however, we follow a less computation-intense type of approximation.
Thus note that d.M; ; U / is a function of the variables, w1 ; : : : w4 and we may
denote it d.w1 ; : : : w4 / We note that
p
d.0; w2 ; 0; w4 / D cosh ` w4
and for small k:
@d.0; w2 .0/; 0; w4 .0// @d.0; w2 .0/; 0; w4 .0//

d.M; ; U / D w1 C w3
@w1 @w3
p
C cosh ` w4 : (5.155)
This is in fact a relation we use many times. And as a further approximation we may
simply use
p
d.M; ; U / D cosh ` w4
D cosh
" r #
1
` 2 I C b 2 U 2 .w21 C kw22 a.1 ak/w11 ak.w21 C w12 //
GJ
b
kD :
U
Hence we need to find such that for
U D U C Ud ; U > 0;
4 2
2 I C b 2 U 2 .w21 C kw22 a.1 ak/w11 ak.w21 C w12 // D GJ ;
`2
where we know that
4 2
b 2 Ud2 .w21 .0/ aw11 .0// D GJ :
`2
Subtracting the bottom from the top we have

2 I C b 2 U 2 w21 w21 .0/ C kw22 a2 kw11 a.w11 w11 .0//

C.w21 .0/ aw11 .0// b 2 U 2 Ud2 ak.w21 C w12 / D 0:
We can approximate the wij here by their limit values at D 0, and obtain the
quadratic equation in :
p
2 1 M 2 I b 2 Ua2 2b .1 C 2a/b 2 .U 2 Ud2 / D 0:
But for U > Ud we see that the term independent of is negative, and the roots are
1 2 2
p b Ua b
1 M I
2

p 2 2 2 p
b Ua b C 1 M 2 I .1 C 2a/b 2 U 2 Ud2 ;
and we note that there is a positive root which also depends on M and I . This
verifies our claim. The main point is that we wind up with a quadratic equation in
, where the term independent of is negative; this is what continues to hold even
with higher-order approximations.
The divergence speed for nonzero and the transonic dip has been covered in
Chap. 4. So far we have considered the CF case. Following similar lines it is easy to
see that for the FF or CC case the speeds are determined by
p
sinh` w4 D 0
or the divergence speed is now given by

p

2 1=4 1 GJ
Ud D 1 M p ;
4b` .1 C 2a/
one-half the value for the CF case.
Root Locus and Stability Curve
Our first objective is to make precise the notion of flutter speed. For this we begin
with a closer examination of the aeroelastic modes. The function d.M; ; U / fixed
M , is analytic in both and U (even though we are interested only in positive values
of U ), and for fixed U , with a branch cut along the negative axis, including zero,
in . For each U there is a countable number of roots of d.M; ; U / D 0 which
we denote by .U /. We consider only nonzero roots, and assume that there are no
multiple roots. If there are, we need to work with each branch in a similar way. We
have in fact seen that there are two near the divergence speed.
We can apply the implicit function theorem by which we see that .M; U / is an
analytic function of U , omitting isolated singularities where
@d.M; ; U /
D 0I d.M; ; U / D 0
@
and in particular we see that:
,
d @d.M; ; U / @d.M; ; U /
.M; U / D j D .M; U /:
d @U @
Also

@2 d.:/ @2 d.:/ d 2 @2 d.:/
C 2
d2 @U 2 @@U dU @2
.M; U / D j D .M; U /;
d2 @d.M; ; U /
@
where of course the denominator cannot be zero. At U D 0, the aeroelastic modes
are the structure modes, and we assume that the damping can be neglected, actually
that the coupling S can be neglected. As before, let us order these modes in terms of
increasing magnitudes, within each classbending or torsionso we can talk for
instance about the first few modes: k .0/ where k .0/ D i !k .0/. For nonzero U
we have difficulty in classifying them. So we use root locus. Thus we consider the
function k .U / starting with the value at zero, the structure mode i !k .0/. Hence we
can talk about a bending mode, or torsion mode. Of course the mode and the
mode shape will change as U increases. We call this function a root locus starting
at i !k .0/. We use the notation
k .U / D k .U / C i !k .U /
and refer to k .U / as the damping termstable mode if the real part is strictly
negative.
We note that if complex, the modes occur in complex conjugate pairs; so we may
only consider the ones with the positive imaginary part for most purposes.
By a stability curve we mean the function k .U /; U > 0 so that we have
instability if it is nonnegative, an unstable mode. Our first result is that the damping
decreases at U D 0, whatever the mode.
Theorem 5.37.
@k .0/ 2b

D at every bending mode (5.156)
@U mM

2b 3 a2 C 13
D at every torsion mode; (5.157)
MI
@!k .0/
D 0: (5.158)
@U
Remarks: Note that the slope does not depend on the mode number. This result does
not appear to be in the aeroelastic lore.
Proof. We again use the local perturbation formula

Z `
Pe `.ACA/
Q D Pe Q C P
À
e.`s/A AesA dsQ;
0
where A is the matrix A./ corresponding to U D 0, the structure-only case we

have already considered in Chap. 2
0 1
0 1 0 0 0 0
B 0 0 1 0 0 0C
B C
B 0C
B 0 0 0 1 0 C
A./ D B m2 C
B EI 0 0 0 0 0C
B C
@ 0 0 0 0 0 1A
2 I
0 0 0 0 GJ
0
and A is the increment to get back to A./. Now, recall from Chap. 2:
P e À Q D
0 sinw11=4 `Csinhw11=4 `
1
1
cosw11=4 ` C cosh w11=4 ` 0
1=4 1=4 1=4 2w1

2 1=4
B1 C
@ 2 w1 1=4
sin w1 ` C sinh w1 ` 1
cos w1 ` C cosh wl 1=4 ` 0 A
2
p
0 0 cosh w4`
and
d0 D determinant P eÀ Q
1 hp i
D 1 C cos w11=4 ` cosh w11=4 ` cosh w4`
2
and the perturbation term (see [21])
Z `
P e.`s/A A esA dsQ D
0
0 1
0
B C
B 0 C
B p C
B w3 2e w4` w11=4 pw42epw4` ;w11=4 pw4ew11=4 ` .pw1Cw4/Cew11=4` .pw1Cw4/2.pw1w4/sinw11=4` C
B C
B C
B 4w11=4 .w1w42 /
C
B C
B 0 C
B C
B C
B 0 C
B p p p p p p p C
B w3 2e w4` w12e w4` w1Cew1 ` . w1Cw4/Cew1 ` . w1Cw4/C2. w1w4/cosw11=4` C
1=4 1=4
B p C
B .4 w1.w1w42 // C
B C
B p p p p p p p C
B w2 2e w4` w11=4 w4C2e w4` w11=4 w4Cew11=4 ` . w1Cw4/ew11=4 ` . w1Cw4/C2. w1w4/sinw11=4` C
B C
B .4w11=4 .w1w42 // C
B C
B p p p C
B w2.. w1Cw4/cosw11=4` C. w1Cw4/coshw11=4` 2w4 cosh w4`/ C
@ 2.w1w42 / A
0
and we see that we can write
d.M; ; U / D d0 C w2 w3 det. ; /; (5.159)

where . : ; : / is a 3 by 3 matrix depending only on ; .
From (5.6) we can calculate that the derivatives we need:
@d @d0 @ @d0 @ @
D C C .w2 w3 Det:. ; //;
@U @ @U @ @U @U
@d @d0 @ @d0 @ @
D C C .w2 w3 Det:. ; //:
@ @ @ @ @ @
And hence at U D 0:
@w4
d.U /
D @U at a pitching (torsion) mode
dU @w4
@
@wr1
D @U at a bending (plunge) mode:
@w1
@
Using:

@wij @wij b
D ;
@U @k U2

@wij @wij b
D ;
@ @k U
we have:

@w1 1 @w11
D bw11 kb ;
@U EI @k

@w1 1 @w11
D b C bU w11 C 2m ;
2
@ EI @k

@w1 @w1
b w11 k
@U @k
D ;
@w1 w11 @w11
2m C C b 2
@ k @k
which we note depends only on k.

Similarly we can calculate that
@w4 3 @c.k/
2b 3 c.k/
k b
@U D @k ;
@w1 b 2 @c.k/
2I C
@ k @k
where
c.k/ D w21 C kw22 a.1 ak/w11 ak.w21 C w12 /
and again the ratio depends only on k. Note that U D 0 means that k D 1.
For M > 0 we see from (5.1) that
2
AOi .; : / ! fi as k ! 1
M
and hence we see that
4
w11 .1/ D ;
M
wij .1/ D 0; i j;
4
w22 .1/ D ;
3M
@wij .k/
k !0 as k ! 1:
@k
Hence
w2 ! 0 as U ! 0.S D 0/;
w3 ! 0 as U ! 0.S D 0/:
Hence
@w1

@U U D 0 D b 4 ;
@w1 2m M
@
which yields (5.3).
Next:
c.k/ 4 4
! C a2 as U ! 0;
k 3M M
@c.k/ 4 4
! C a2 as U ! 0:
@k 3M M
Hence (5.4) and (5.5) follow. t
u
Remarks: The appearance of .1=M / is interesting, showing in particular that these

are not continuous with respect to M at zero. In fact we have to deal with the case
M D 0 separately.
The stability curve for M D 0

Noting that
T .k/ ! 1 as k ! 1;
we have:

@.U / b
U D0 D at every bending mode
@U 2.m C b 2 /
2
b 3 a 12
D at every pitching mode: (5.160)
2 I C .a2 C 18 /
Definition of Flutter Speed
We can now make a precise definition of flutter speed.

We begin with the stability curve:
k .U / as a function of U with k .0/ D i !k; !k > 0:
Theorem 5.37 shows that the slope of this curve is strictly negative at U D 0. Hence
the speed U at which k .U / becomes zero again for the first time we call the flutter
speed for this mode denoted UF .k/. Thus UF .k/ is defined by:
k .UF .k// D 0I k .U / < 0 for U < UF .k/I

d
k .UF .k// > 0:
dU
The second condition means that the system becomes more unstable as we increase
the speed. Of course at this point, Im k .UF / is not necessarily D !k and may have
changed considerably.
Flutter Speed
By flutter speed we mean

inf
k UF .k/ D UF ; (5.161)
where the infimum is taken over all structure modes.
In the event that no structural mode flutters, we would have to define it to be

infinity.
The situation changes if we modify the definition (as we do) to be instead the
infimum over all aeroelastic modes (rather than merely structure modes) because
this would then include zero. First we prove that the flutter speed in either definition
is positive.
Corollary 5.38. The flutter speed is positive. The infimum in (5.8) is attained.
Proof. Suppose the infimum in (5.8) is zero. Then we can find a sequence of mode
frequencies !n and corresponding speeds UF n such that UF n converges to zero.
Suppose the infimum is attained for some n. Then the corresponding flutter speed
must be positive because the slope at zero speed has to be negative by Theorem
5.37. Suppose then it is not attained. Consider the stability curve for each n:
n .U / D Re n .U / for 0 < U < UF n :
We have for all n: by Theorem 5.37
d
n .0/ < < 0
dU
and by definition
d
n .UF n / 0:
dU
The function .d=dU /n.U /is continuous on finite intervals. Hence for some U
denoted U n the slope
d
n .Un / D 0; where Un < UF n
dU
and Un goes to zero by hypothesis.
Let us look at the corresponding mode sequence !n . Suppose the sequence is
bounded. Then we can find a convergent subsequence with finite limit !1 which
would then attain the infimum. Hence the sequence must be unbounded.
Then
n b
Kn D I d.M; n; Un / D 0
Un
and in as much as !n ! 1; we must have that
jn j ! 1:
The Mikhlin multiplier converges to a finite limit as jn j ! infinity, and

,
@ @d @d
D
@U @U @
and
@d @2 d.:/ @2 d.:/ @d

@2
D @U @@U @U 2 @ D .M; U /
2 2
@U @d
@
converge to finite values as jn j ! 1.

Now for each n we have, with primes denoting derivative with respect to U :
Z Un
0 D n0 .Un / D n0 .0/ C n00 .U / dU ;
0
where the integral ! 0 and the first term goes to a nonzero value, leading to a
contradiction.
Hence the infimum is attained and is positive. u
t
A computer program for generating stability curves in incompressible flow is
given in the Appendix of this chapter with numerical calculations for the Goland
model. We note that the first torsion mode flutters.
Theorem 5.39. Zero is an aeroelastic mode for 0 M < 1.
Proof. For M D 0:
We note that from Sect. 5.4
O
L./ D0 at D 0:
Hence to prove that zero is an aeroelastic mode we need to find Y such that
AY D 0; Y 0:
Or, we need to find a nonzero solution of (see Sect. 5.4)

x
Y D l x2 D 0;
x2
Ax 1 C Kx1 D 0:

h
Let x1 D ,

GJ 00 .Y / b 2 U 2 .Y / D 0;

EI h0000 .y/ D 0 plus end conditions: (5.162)
But this is precisely the linear steady-state aeroelastic equation treated in Chap. 3.
For a nonzero solution we have an eigenvalue problem, which has a solution for
only a sequence of far field speeds: under CF end conditions
p
1 GJ
U1 D Un D .2n C 1/ p ;
2b` .1 C 2a/
with .y/ determined by (5.9.) and h.y/ D 0 and also, of course d.0; 0; Un/ D 0.
For 0 < M < 1 we have
AY L.0/Y D 0
and from the evaluation of the wij .0/ we see that

x1 h
Y D I x1 D ;
0
4
EI h0000 .y/ C bU 2 .y/ D 0;
M
1
GJ 00 .y/ D b 2 U 2 p .y/; (5.163)
1 M2
plus end conditions. Thus for nonzero solution we have an eigenvalue problem with
solution (under CF end conditions):
p
1 GJ
U1 D Un D .2n C 1/.1 M / 2 1=4
p
2b` .1 C 2a/
and .y/; h.y/ are nonzero, and of course d.M; 0; Un / D 0: t

u
Remarks: Note that it is required that M < 1. For M D 1; the formula for Un
yields zero. Indeed we have seen that for M D 1 zero is not an aeroelastic mode.
So what is new is that we have evaluated the mode shape corresponding to the
aeroelastic mode equal to zero. And we see that Un is the sequence of flutter speeds
corresponding to the flutter mode zero, which however is not a structure mode.
Moreover if we consider this as corresponding to k D 0, the stability curve
starting at 0 .Un /, we see that the slope is nonnegative, by the Remark under
Theorem 5.37.
In particular this shows that there are zeros other than those given by the root loci
of the structure modes.
Which definition do we use? Note that we do not know whether UF .k/ is < Ud :
This depends on whether a is positive or negativewhether the cg is above
or below the elastic axis; see [6]but of course we have offered no mathematical
proof. We use k D 0 to indicate that the mode frequency is zero.
Note that for the mode zero we were essentially calculating the mode shape .x1
in our notation above).
Note that by our definition, if we include k D 0 in (5.9) we have
0 < UF Ud < 1: (5.164)
The appendix presents a computer program for calculating the flutter speed in
incompressible flow for the Goland model. Here the first torsion mode flut-
ters.
At the present time we have no idea, no formula for determining which mode
will flutter and which wont, without numerical calculation. Fortunately we are
interested in practice only in the first few modes, which makes it doable by
computation.
Now we return to proving the convergence of the many series representations for
the solution of the linear equation. The basic idea here is that as the mode number
increases in the limit they all are like the structure modes fk .0/g. Also we know that
physically the modes cannot grow indefinitely (there are no microwaves) but the
model shows that they are of diminishing importance as the mode number increases.
Dependence on Far Field Speed
The roots fk g depend on U1 which we shorten to simply U and use the notation
k .U / when we need to emphasize the dependence on U .
Thus
A.0/Yk D i !k Yk I A.0/ Zk D i !k Zk I Zk D Yk I Yk; Zk D 2!k 2 :
Normalization
x1 k
Mx1k; x1k D 1I Yk D :
i !k x1 k
Similarly we now use R.; U / to emphasize the dependence on U.
We can express
O
A C L./
O U /;
A.U / D A.0/ C T .U / C L.;
where T .U / depends on U but not on :

0 0
T .U / D
M1 K.U / Ml D .U /
and defines a compact linear bounded operator on H into H for each U.

O U/ D 0 0
L.; ;
M1 FO .; U / 0
which is also a bounded linear operator as T .U / but goes to zero in operator

norm as jj goes to infinity in the spectral strip. Our main result is the asymptotic
equivalence of the aeroelastic modes with the modes of the structure.
O
Note that it holds in particular if the convolution term L./ is absent.
Theorem 5.40. For each U
n .U /
! 1 as n ! 1:
n .0/
Proof. The Mikhlin multiplier:

p
1 M 2 2 C 2M i ! C .1 M 2 /! 2
2 C i!
behaves badly as ! 0, however, it goes to a finite limit as jj ! 1. This is what

we exploit.
As jj ! 1 in any strip that houses the roots, the multiplier goes to
M=2 and
2
.; M / ! ;
M
2
w11 ! ;
M
wij !0 i j;
w22 !
and similarly for the derivatives with respect to .

We use the same technique as in proving Lemma 5.41.
The (nonzero) roots fn .U /g are defined along the root locus and we have:
Z U
@n .u/
n .U / D n .0/ C du: (5.165)
0 @u
As n increases so does jn .U /j; and

Z
U
@n .u/
du goes to a finite limit as n ! 1:
@u
0
This follows from the calculations above. Dividing by n .0/ in (5.8), we obtain
RU @n .u/
n .U / du
D1C 0 @u
; (5.166)
n .0/ n .0/
where the second term goes to zero with n. Hence the ratio
n .U /
! 1 as n ! 1:
n .0/
t
u
In particular we see that
Re n .U / ! 0 as n ! 1 (5.167)
and that
n .U / D O.n/ (5.168)
because this is true of n .0/.
What we have proved is that the sequences fk .U /g as U varies are
asymptotically equivalent to fk .0/g; see [2] for similar property in control
problems.
Dependence on Mach Number
The next big question is the dependence of the flutter speed on M; UF .M /.

Actually we are interested in the flutter speed at a given altitude. The speed of
sound a1 depends on the altitude. Thus the system is stable if

U
U < UF :
a1
Dependence of Flutter Speed on Mach Number
The Federal Aviation Administration (FAA) in the United States mandates a 15%
margin. The question of whether the Flutter speed decreases or increases with M
has been of interest from the early days of aeroelasticity. The first attempt at this was
by the pioneer Garrick [38] where he determines an empirical formula drawing on
the Possio integral equation for nonzero M for M up to 0:6 or so. We may express
this as the Garrick formula:
UF .M / D .1 M 2 /1=4 UF .0/:
Note that according to this, the flutter speed decreases as M increases, and goes to
zero as M goes to 1. This is of course for zero angle of attack. It is remarkable that
there is a marked change for nonzero angle of attack as seen for D 0 in Chap. 4.
Because we are interested in the dependence on M , we now write .M; U / in
place of .U /I and similarly k .M; U /; UF .M; k/ for the kth mode.
It has been observed [19] that a mode which flutters at M D 0 does not flutter
at M D 1, and vice versa. This would indicate that for some value of M there are
multiple roots for the same U . This is true near the flutter speed for k D 0; as we
have already noted. We return to this question in Chap. 6.
Numerical calculations indicate that typically for current commercial aircraft
(heavy wings) the value of k for flutter is small, around 0.1 or so. In as much as
we know that
p
d.M; 0; U / D cosh` w4 ;
we may use the perturbation formula to obtain:
1
d.M; ; U / D .2.2880.4w32w33
5760w49
4w1 w22 w23 w4 C w94 C 2w2 w3 w44 .w1 w24 //
C 1440w2 w3 w4 .4w22 w23 C w64 2w2 w3 w4 .2w1 C w24 //`2
C 120w24 .w2 w3 C w1 w4 /.12w22 w23 C 2w64 C w2 w3 w4 .2w1 5w24 //`4
4w2 w3 w34 .w2 w3 w1 w4 /.24w2 w3 4w1 w4 5w34 /`6
p
C w44 .w2 w3 w1 w4 /2 .w2 w3 w34 /`8 /cosh w4 `
C w2 w3 .1440.12w22w23 12w1 w2 w3 w4 C 5w2 w3 w34 C 4w1 w44 8w64 /
C 2880w4 .3w22 w23 2w1 w44 C 3w2 w3 w4 .w1 C w24 //`2
C 60w24 .w2 w3 C w1 w4 /.5w2 w3 C 4.3w1 w4 C w34 //`4
C 8w34 .w2 w3 C w1 w4 /.21w2 w3 C 26w1 w4 /`6 C 6w44 .w2 w3 w1 w4 /2 `8
60.4w1 w44 .24 C w1 `4 / C w22 w23 .96 C w4 `2 .48 C 5w4 `2 //
p
3w2 w3 w4 .40w24 C w1 .32 C w4 `2 .16 C 3w4 `2 ////cosh2 w4 `
p
C w4 `..2880.4w22w23 C 2w1 w44 C 3w64 C 4w2 w3 w4 .w1 2w24 //
480w4 .17w22 w23 C 2w1 w24 .w1 C 3w24 / w2 w3 w4 .19w1 C 3w24 //`2
48w24 .w2 w3 C w1 w4 /.21w2 w3 C w1 w4 C 10w34 /`4
C 4w34 .w2 w3 C w1 w4 /.w2 w3 C 4wlw4 /`6 C w44 .w2 w3 w1 w4 /2 `8 /
p
sinh w4 ` C 24.w2 w3 w1 w4 /.w24 .120w4 C 20w1 `2 C w1 w4 `4 /
p
C w2 w3 .240 C w4 `2 .50 C w4 `2 /// sinh2 w4 `///: (5.169)
In addition we may use the approximation for wij following [21] using the Neumann
expansion in Sect. 5.4. Thus we have:

2P i k
w11 D p 2 p BesselK.0; k/
1 M2 1 M2
C BesselK.1; k//.BesselI.0; k/
Z 1
BesselI.1; k// k a.s/.BesselK.0; ks/ C BesselK.1; ks//
0
.BesselI.0; ks/ BesselI.1; ks//ds

Z 2
C a.s/.BesselK.0; ks/
0

BesselK.1; ks//.BesselI.0; ks/ C BesselI.1; ks//ds ;

Pi k
w12 Dp 1C p
1M 2 1 M2

1 2BesselI.1 ; k/
.BesselK.0; k/ C BesselK.1; k//
k k
Z 1
1 2BesselI.1; ks/
Ck a.s/ .BesselK.0; ks/
0 ks ks
Z 2
C BesselK.1; ks// ds C k a.s/
0

1 2BesselI.1; ks/
C .BesselK.1; ks/ BesselK.0; ks// ds ;
ks ks

2P i 1 k
w21 D p Cp
1M 2 2 1 M2

1 1 1 BesselK.1; k/
BesselI.0; k/ BesselI.1; k//
k k 2 k
Z 1
1 1 1
Ck a.s/
0 ks ks 2

BesselK.1; ks/
BesselI.0; ks/ BesselI.1; ks// ds
ks
Z 2
1 1 1 BesselK.1; ks/
Ck a.s/ C
0 ks ks 2 ks

.BesselI.0; ks/ C BesselI.1; ks// ds;

1 1 Pi 1
w22 D p p 2P i BesselI.1; k/BesselK.1; k/
1 M2 1 M2 k k
Z 1
1
C a.M; s/ .P i 2P i BesselI.1; ks/BesselK.1; ks//ds
0 ks 2
Z 2
1
C a.M; s/ .P i 2P i BesselI.1; ks/BesselK.1; ks//ds:
0 ks 2
Here we have a still further approximation:

1 p 1
d.M; ; U / D w1 `4 cosh w4 ` C w3 w2 .11520w64
12 5760w94
p
2.5760 w64 C 1440 w74`2 240w84 `4 /cosh w4 `

13=2 p p
C 8640 w4 ` sinh w4 `/ C cosh` w4 :
We may use this to calculate
@d
@UF .0; / @M
D ;
@M @d
@U
at D 0, so that UF .0; k/ D Ud; and
b 2 .1 C 2a/ 2
w4 D p U ;
GJ 1 M 2
and hence
@w4
@UF .0; / M Ud
D @M D
@M @w4 2.1 M 2 /
@U
as obtained in [21].
This is consistent with the Garrick formula above but of course it is limited to
small . It is also consistent with divergence as a flutter speed for zero frequency,
except that zero is not a structure mode as we have noted.
Finally we extend the theory to the nonlinear structure models we introduced in

Chap. 2.
Linearization: BeranStraganac Model
We are fortunate that for this model in spite of its extreme complexity, the static
solution is the zero solution for the structure and constant air flow. Hence the
linearization yields exactly the same equations as in the case of the linear Goland
model.
Linearization: DowellHodges Model
Here as we have noted the presence of the gravity terms complicates matters. Of
course they have to be considered only along with nonzero far field velocity, as we
have noted before.
The static solution which is nonzero, is given in Chap. 4, Sect. 4.5. Here we go
on to linearize the solution to the aeroelastic equations about this solution. This
requires substantial effort because the static structure solution is no longer zero. An
important question is the role played by the gravity terms.
Thus we seek the solution to the structure equations in the power series form in
the structure state variables. Using the parameter , we define
x.; t; s/ D x.0; s/ C x.t; s/; 0 < s < `;

0 1
h
x D @v A (5.170)

and determine the linear equation characterizing x.t; :/ by retaining only linear
terms in . Correspondingly .; t; x; z/ satisfies the field equation for every :
@2 .:; / @
C jjr.; :/jj2
@t 2 @t
2
1 U1 jjr.; :/jj2 @.; :/
D a1 2
1C 2 .; :/
a1 2 2 @t

jjr.; :/jj2
r.; : / r (5.171)
2
and
@ jrj2
.; t; x; z/ D C ;
@t 2
p.; : / D .; : :/
with the boundary conditions detailed in Chap. 3 plus the flow tangency condition
and the Kutta Joukowsky condition.
And
0 .x; z/ D the static potential;

Z b
L.; t; y/ D 1 .; x; y/dx;
b
Z b
M.; t; y/ D 1 .x ab/ .; x; y/dx
b
and x.; t; s/ satisfies the equations:
R t; y/ C EI 1 h0000 .; t; y/ C .EI 2 EI 1 /..; t; y/v.; t; y/00 /00

mh.;
D mg sin' C L.; t; y/; (5.172)

0000 00 00
mRv.; t; y/ C EI 2 v .; t; y/ C .EI 2 EI1 /..; t; y/h.; t; y/ /
D mg cos '; (5.173)

R
I .; 00
t; y/ GJ .; t; y/ C .EI 2 EI 1 /h.; t; y/ v.; t; y/00 00
D M.; t; y/ 0 < tI 0 < y < `: (5.174)
Taking the derivative with respect to at zero (or equivalently equating coefficients
of the first power of ), we see that x.t; :/ satisfies the linear equation:
R y/ C EI 1 h0000 .t; y/ C .EI 2 EI 1 /..t; y/vo .y/00

mh.t;
C o .y/v.t; y/00 /00 D L1 .t; y/;
mRv.t; y/ C EI 2 v0000 .t; y/ C .EI 2 EI 1 /..t; y/ho .y/00
C o .y/h.t; y/00 /00 D 0;
R y/ GJ 00 .t; y/
I .t;
C .EI 2 EI 1 /.h.t; y/00 vo .y/00 C ho .y/00v.t; y/00 / D M1 .t; y/;

where
Z b
L1 .t; y/ D p1 .t; x; y/dx;
b
Z b
M1 .t; y/ D .x ab/p1 .t; x; y/dx;
b
d
p1 D p.; :/ D 0 0 < tI 0<y<`
d
and
ho .y/ D h.0; y/;
vo .y/ D v.0; y/;

o .y/ D .0; y/: (5.175)
Note that these functions depend on the gravity terms.

Consider first the case where the air speed is zero, U1 D 0, but we retain the
gravity terms, so that we have that x.0; s/ is the solution of:
EI 1 h0000 .s/ C .EI 2 EI 1 /..s/v.s/00 /00 D mg cos';
EI 2 v0000 .s/ C .EI 2 EI 1 /..s/h.s/00 /00 D mg sin';
GJ 00 .s/ C .EI 2 EI 1 /v.s/00 h00 .s/ D 0:
Plus CF end conditions.

A power series solution is given Chap. 4, Sect. 4.5. From which we can see in
particular that the functions are real-valued.
The linear structure dynamics equations then become:
R y/ C EI 1 h0000 .t; y/ C .EI 2 EI 1 /..t; y/v0 .y/00 C 0 .y/v.t; y/00 /00 D 0;

mh.t;
mRv.t; y/ C EI 2 v00 .t; y/ C .EI 2 EI 1 /..t; y/h0 .y/00 C 0 .y/h.t; y/00 /00 D 0;
R y/ GJ 00 .t; y/ C .EI 2 EI 1 / .h.t; y/00 v0 .y/00 C h0 .y/00 v.t; y/00 / D 0:

I .t;
This can be given a Hilbert space formulation as in the case of the Goland model in
Chap. 2.
Let H D L2 0; `3 with elements

0 1
h
x D @vA :

Let A denote the operator with domain:
D.A/ D fxjh0000 ; v0000 ; 00 L2 0; `;
with CF or FF end conditions as in Chap. 2

0 1
EI1 h0000
Ax D @ EI2 v0000 A :
GJ 00
Define the operator B with domain: D.B/ D D.A/.

Then because the coefficient functions h0 .y/; v0 .y/, and 0 .y/ have continuous
fourth derivatives in 0; `, it follows that
0 1
.v000 C 0 v00 /00
Bx D .EI2 EI1 / @.h000 C 0 h00 /00 A
h00 v000 C h000 v00
is in H. Because x0 .: / is real-valued, it is readily verified that
B D B D B
and is closed on the domain of A. Hence
.A C B/ D .A C B/
and is closed on the domain of A. Hence the equations can be written in abstract
form:
Mx.t/
R C Ax.t/ C Bx.t/ D 0;
where
M D diagm; m; I :
We note that, as in Chap. 2, A is self-adjoint and nonnegative definite, and Ax D 0
implies x D 0.
Hence we may, as in Chap. 2, introduce the energy norm space
p
H D D. A/H
with inner product:

p p
Y; Z D Ax1 ; Az1 C Mx2 ; z2
and rewrite the structure equation as
YP .t/ D AY .t/ C BY .t/;
where

0 I
AD ;
Ml .A C B/ 0
where we note that D.A/ is contained in D.B/, and

xl
D.A/ D ; x2 H; x1 D.A/ :
x2
Let us calculate first the resolvent of A. Thus we consider the equation:

xl z
Y AY D ZI XD DI ZD l :
x2 z2
Or
2 Mx1 C Ax1 C Bx1 D M.z2 C z1 / x2 D x1 z1 : (5.176)
Lemma 5.41.
B.2 M C A/1
is bounded (but not necessarily compact) and the norm

1
jjB.2 M C A/1 jj D 0 for > 0:
2
Proof. Only the last part is new. Here we note B is closed and that the range of
.2 M C A/1 is contained in the domain of A which is contained in the domain of
B. And we can write
1 1
B 2 M C A D BA1 2 MA1 C I ;
2 1 2 2
2 X k
MA1 C I 1 x D x; k ;
C
2
k
kD1
which is enough to show via the uniform boundedness principle that

1 1
jj 2 MA1 C I jj D 0
2
and in as much as BA1 is bounded, this is enough to show that

1
1
jjB. MCA/ jj D 0
2
: t
u
2
Hence we can go back to (5.7).

Lemma 5.42. The resolvent R.; A/ is compact for all in the resolvent set that
includes a half-plane
X D R.; A/Z;
1 1
x1 D 2 M C A I C B 2 M C A M.z2 C z1 /;
x2 D x1 z1 (5.177)
for all large enough.

Proof.
The compactness follows from (5.8). t
u
We next proceed to calculate the modes following the procedure outlined in
Chap. 2, Sect. 2.4, for the Goland beam. Here we simplify the analysis by taking
advantage of the fact that mg.EI 2 EI1 / is small.
Let
Y D Colh; h0 ; h00 ; h000 ; v; v0 ; v00 ; v000 ; ; 0 :
Then (5.7) can be expressed:
Y 0 ./ D .A../ C B./Y ./;
where
0 1
0 100 0 0 00 0 0
B0 010 0 0 00 0 0C
B C
B0 001 0 0 00 0 0C
B C
Bw 0 0C
B 1 000 0 0 00 C
B C
B0 000 0 1 00 0 0C
A../ D B C;
B0 000 0 0 10 0 0C
B C
B0 000 0 0 01 0 0C
B C
B0 000 w2 0 00 0 0C
B C
@0 000 0 0 00 1 0A
0 000 0 0 00 0 w3
m
w1 D 2 ;
EI 1
m
w2 D 2 ;
EI2
I
w3 D2 ;
GJ
where B.Y .s/ D .EI 2 EI1 /Col0; 0; 0; .v0 00 C 0 v00 /00 ; 0; 0; 0;

00 00
.h0 C 0 h / ; 0; .h v0 C h0 00 v00 /, which we solve as
00 00 00
Z s
Y .s/ D eA. ./s Y .0/ C eA. ./.s / B. Y ./d;
0
which is a Volterra integral equation that we solve by retaining only the linear term
in B. to obtain:
Z s
Y .s/ D eA. ./s Y .0/ C eA. ./.s / B. eA. ./ Y .0/d;
0
so that
Z !
`
Y .`/ D eA. ./` C eA. ./.` / B eA. ./ d
. Y .0/:
0
The eigenvalues for the CF end conditions then are the zeros of
Z !
`
d./ D det P eA. ./` C eA. ./.` / B. eA. ./ d Q;
0
the determinant of a 5 5 matrix, where P D Q , where now P is 5 10, given by:

0 1
000 00
B0 0 0 0 0C
B C
B1 0 0 0 0C
B C
B0 1 0 0 0C
B C
B C
B0 0 1 0 0C
P DB C:
B0 0 0 1 0C
B C
B0 0 0 0 0C
B C
B0 0 0 0 0C
B C
@0 0 0 0 0A
000 01
The main point to note here is that d./ is an entire function of and has a countable
number of zeros with no limit point in the finite plane.
Let us see what we can say next about the perturbed eigenvalues.
First we see that
d0 ./ D det PeA. ./` Q;

D i !k D .1 C cosh k cos k /.1 C coshk cosk /cos!k` ;
where the mode frequencies are:

1=4
1 EI1
!k D k
` m
or
1=4
1 EI2
!k D k
` m
or
s
GJ
!k D .2k C 1/ :
2` I
Let the perturbation be denoted:
Z `
d1 ./ D Det P eA. ./.` / B. eA. ./ dQ:
0
We note that this is a function of 2 . And hence d1 .i !k / is real-valued, although the

derivative is imaginary. Hence a one Newton step shows that the root continues to be
imaginary. Hence it follows that the perturbed eigenvalues are also imaginary. Usu-
ally for perturbation by a controller, for example, we assume that the eigenfunctions
are approximately the same and only the eigenvalues change. If we do that here we
would find no change in the eigenvalues because Bk; k D 0.
Linearizing the Euler Full Potential Equation
Next let us linearize the potential equation about the static solution. Let 0 .:; :/
denote the static solution, and .t; :; :/ the linearized potential. Then .t; :/ satisfies
the linear equation with nonconstant coefficients:
2
@2 @r 1 U1 jro j2
C 2r0 D a1 1 C 2
2

@t 2 @t a1 2 2

@
. 1/0 r0 r C
@t

jr0 j2
r0 r.r0 r/ r r
2
with the boundary condition

@.t; x; 0/ P y/ C .x ab/.t;P y/ @0 .t; y/
D h.t;
@z @x

0 .y/.@.t; x; 0// @x; jxj < b:
This is of course quite a formidable taskas statedbecause the coefficients are

no longer constant. So we seek only an approximate solution, as in the case above
for the structure. Thus we break up the linearized equation with
00 .x; z/ D xU
and correspondingly
.t; x; z/ D 00 .t; x; z/ C 01 .t; x; z/
with the boundary conditions

@.t; x; 0/ P y/ C .x ab/.t;P y/ @0 .t; y/
D h.t;
@z @x

C 0 .y/.@ .t; x; 0//=@x : (5.178)
The boundary conditions dont involve the in-plane bending v.t; : /.

Now .t; : / satisfies the field equation:

@2 @2 2
2 @ @2
C 2U C a1
2
.1 M / C
@t 2 @x@t @x 2 @z2
2
@ U1 jr0 j2
D '.t; x; z/ D .r00 r0 /: r C . 1/
@t 2 2

@
. 1/0 r0 r C .r0 r00 /
@t

jr0 j2
r..r0 r00 / r/ r r:
2
So we express the solution as the sum of two: One (denote it F ) that satisfies
the nonhomogeneous field equation with zero boundary conditions and the other
(denote it B ) that satisfies the homogeneous field equation:

@2 @2 2
2 @ @2
C 2U C a1 .1 M / 2 C 2 D 0
2
@t 2 @x@t @x @z
with the boundary conditions given by (5.7).

The solution to the first is given by (as in [7])
Z t Z
F .t; x; z/ D d L.t ; x ; z /'.; ; /dd;
0 R2
where
1 1
L.t; x; z/ D p s ;
2
2a1 1 M2 2 1 x2 z2
t U x
C 2
c12 1 M2 c12 c2
where
c12 D a12
.1 M 2 /I c22 D a1
2
;
Z t
z A.t ; /d
B .t; x; z/ D d
0 ..x U/2 C z2 /
and
A .t; :/ is the solution of the Possio equation:
Z b
P y/ C .x ab/.t;
P .t ; x /A.; /dd D h.t; P y/
b
Z
@0 d t
.t; y/ C 0 .y/ A.t; x/ C A.t ; x U/d ;
@x dt 0
where P .., . / is the linear TDP kernel as given in Sect. 5.6.

This equation is a little bit more involved than the standard TDP, because of the
function 0 .y/ but we can still take advantage of our solution of the Possio equation
and rewrite the equation as
A C 0 .y/ .M /LA D .M /w; (5.179)
where
Z
d t
LA D A S.U/A.t ; : /d;
dt 0

w.t; x/ D h.t; P y/ C @0 .t; y/
P y/ C .x ab/.t; for fixed y:
@x
And hence
A D .I C 0 .y/ .M /L/1 .M /w;
which may be approximated by
A D .I 0 .y/ .M /L/ .M /w: (5.180)
Hence we have:
Z tZ
.t; x; z/ D B .t; x; z/ C dL.t ; x ; z /'.t; ; /dd;
0 R2
which is a Volterra equation in the time domain, and has the solution (up to the first
term):
Z t Z
.t; x; z/ D B .t; x; z/ C d L.t ; x ; z /:
0 R2
2
@ U1 jr0 j2
.r00 r0 / . rB C . 1/ B
@t 2 2
. 1/0 .r0 :rB / .r0 r00 / r..r0 r00 / rB /

r.jr0 j / rB dd:
2
What is unique about this example is that we have modes (eigenvalues) even
though the system is not linear which continue to be the modes for the linearized
approximation.
We stop here because further analysis would involve explicit use of the function
0 .., . /.
5.9 Appendix: Computer Program for Flutter Speed

in Incompressible Flow: Goland Model
Dynamic Response of a Wing in Unsteady Aerodynamics
This Program Calculates the Flutter Speed of a Wing in

Unsteady Aerodynamics Using Full 2D Continuum Model
(*Initialization & System Parameters*)

(*Initialization & Output Log File*)
Off [General::"spelll", General :: "spell"];
(*Initialize*)
ClearGlobal[ ] :=ClearAll "Global0 "I Clear Derivative];
Remove Global0 I /I
ClearGlobal[ ];
sciNumut[num ] := ToString [MantissaExponent
[Abs[num]] [[1]]] <>
"E"<> ToString" [MantissaExponent[Abs[num]] [[2]]];
writeNum name ; num; desc W "" WD name <>" = n t"<>
ToString[N[num]] <> "n t" <> desc <>"n n";
nbInfo = NotebookInformation [EvaluationNotebook[ ]];

sLogFileName =
"z. RunLog."<> StringReplaceList Extract ["FileName"/.
nbInfo, 2],
". nb" ! ""];
nbFileName = Extract["FileName"/. nbInfo, 2];
nbDir = DirectoryName "ToFileName["FileName"/. nbInfo]];
nbFilePath =
"FileName"/. nbInfo /. FrontEnd FileName dir ;
fname ; $ ToFileName [dir, fname];
sWinTtl =": Initializing, creating log file. . . ";

Setoptions [EvaluationNotebook[ ], WindowTitle ! Dynamic
[nbFileName <> sWinTtl]];
SetDirectory [nbDir];
sTimeStampForm = f"Year", "-", "Month", "-", "Day",
",", "Hour", ".", "Minute", ".", "Second",",",
"Millisecond"g,
sOutFile = DateString Join [fsLogFileName, ", "g,
sTimeStampForm]];
sLogFile = sOutFile <>".xls";
sPDFFile = sOutFile <>". pdf";
sStartTime = DateString Join [f"Start time: ntg,
sTimeStampForm, f"nng]];
5.9 Appendix: Computer Program for Flutter Speed in Incompressible Flow... 261
(*formatted time stamp, write in log file *)

dtStartTime = DateList [ ]; (*raw time stamp for run
time
calculation *)
fTimeUsedO = TimeUsed [ ]; (*total CPU processing time
used by
Mathematica Kernel *)
strm = OpenWrite [sLogFile];
(*Parameters for the Structure*)

m = 0.7; (* mass per unit length *)
a = -0.3; (* location of elastic axis *)
b = 3; (* half-chord length *)
1 = 20; (* span *)
S = 0.447;
I D 1:943I
GJ D 2:39 106
EI D 23:6 106 I
mt D 0I
st D 0I
Iyt D 0I
It D 0I
(*Parameters for Control*)

ghi D 0I gi D 0I steu D 010I numu D 100I
(*Parameters for the Air*)

= . 0022973;
b2
D I
m
2:39 10 6
(*Calculation of 1st Bending and 1st Torsion Modes of the

Structure *)
sWinTtl =": Calc lst bending & 1st torsion of
structure...";
s
xD I
q
mb
r D mbI2 I
Iyt 2
n1 D EI C gh
EI I n2 D mt=EI2I n3 D St =EI2 I
g
n4 D St =GJ2 I n5 D It =GJ2 C I
GJ
m
tw1 D 2 .1 C /I
EI
m
tw2 D b2 .x a/I
EI
m
tw3 D b2 .x a/I
GJ
m 2 2 2 1
tw4 D b r C Ca 2
I
0 GJ 1 8
0 100 0 0
B 0 0 1 0 0 0C
B C
B C
B 0 0 0 1 0 0C
ADB C =: U > 0I
Btw1 0 0 0 tw2 0C
B C
@ 0 0 0 0 0 1A
tw3 0 0 0 tw4 0
0 1
0 n1 1 0 0 0
p3 D @n2 0 0 1 n3 0A I
n4 0 0 0 n5 1
0 1
000
B0 0 0 C
B C
B C
B1 0 0 C
q3 D B CI
B0 1 0 C
B C
@0 0 0 A
001
2
r
2 EI
!b1 D .0:597/ 2 I
r ` m
GJ
!t1 D I
I 21 r
2 EI
!b2 D .1:49/2 2 I
r ` m
GJ 3
!t2 D I
I 21
g D Detp3:MatrixExpA 1:q3=: > !I=:gh > 0=:g > 0I

!2 D !=:FindRootg DD 0;f!; !b1; !b1 C 0:01gI
!3 D !=:FindRootg DD 0;f!; !b2; !b2 C 0:1gI
!1 D !=:FindRootg DD 0;f!; !t1; !t1 C 0:1gI
!4 D !=:FindRootg DD 0;f!; !t2; !t2 C 0:01gI
mode D Chopf!1; !2; !3; !4gI
mode=2= I
(*System Equations*)
sWinTtl D W Calc system equations : : : I
ClearAllA; p3; q3I
z D b=UI
w11 D Pi z C 2 Pi T;
w12 D Pi TI
w21 D Pi TI
w22 D Pi=2.1 C z=4 T/I
T D BesselK1; z=.BesselK0; z C BesselK1; z/I

w1 D 1=EI.2 m C Ubw11/I
w2 D 1=EI.2 S C bU2 ..1 az/w11 C zw12//I
w3 D 1=GJ.2 S C b2 U.w21 aw12//I
w4 D01=GJ.2 I C b 1 U .w21 C zw22 a.1 az/w11 az.w21 C w12///I
2 2
0 100 0 0
B 0 0 1 0 0 0C
B C
B C
B 0 0 0 1 0 0C
ADB C
Bw1 0 0 0 w2 0C
B C
@ 0 0 0 0 0 1A
w3 0 0 0 w4 0
0 1
0 n1 1 0 0 0
p3 D @n2 0 0 1 n3 0A I
n4 0 0 0 n5 1
0 1
000
B0 0 0 C
B C
B C
B1 0 0 C
q3 D B CI
B0 1 0 C
B C
@0 0 0 A
001
1
d0 D Detp3:MatrixExpA1:q3=:Cz > I
2
d1 D Detp3:MatrixExpA1:q3=:gh > ghi=:g > gi=:
BesselK1; z
Cz > I
BesselK0; z C BesselK1; z
(*Root Loci with U as the parameter*)

(*Root Loci*)
sWinTt1 =": Calc Root Loci w/ param U.";
ClearAll; gh; g; UI gr D fgI tp D fgI
st0 = "Root Loci with U as the parameter:"; (*Prepare

Plot Titles*)
sPlotTtl = fst0 <> "1st Torsion"; st0 <> "1st Bending";
st0 <> "2nd Bending"; st0 <> "3rd Bending"; st0 <> "2nd
T orsi on"I gI
10 D "; Listing for"I
11 D f10 <> "1st Torsion"; 10 <> "1st Bending"; 10 <>
"2nd Bending";
10 <> "2nd Torsion gI

Do U D 0I ht D fgI rp D fgI ip D fgI tmp D modekiI
WriteString [strm, "UntSigma=Re(L)ntf=Im(L) ntRe/Imnt
|dL|n
tmoden n"];
( column headers )

Do U D U C steuI D =:FindRootd1 DD 0, f;
tmp; tmp C 0:010 igI tmp D I Um D UI
sWinTt1 =": Calc Root Loci w/ param U = "<> ToString[U];
WriteString [strm, StringForm ["" n t"n t"n t"n t"n t"n
n",
NumberForm[Um, 10], NumberForm[Re[], 10], NumberForm
[Im[], 10],
NumberFormRe=Im; 10; sciNumOutAbsd1,

NumberForm[mode[[k]],
10]]];
Re
rp = Append rp; Um; I
Im
Im
ip = Append ip; Um; I
2
Re Im
ht = Append ht; ; ,
Im 2
fj,
1, numug
;
sTimeUsed = "CPU time used (min): nt"<> ToString
[(TimeUsed
[]-- fTimeUsed0)/60.0]<> "nnnn"; WriteString [strm ,
sTimeUsed];
tp = Append [tp, frp, ipg];
p1 = ListLinePlot [rp, AxesLabel ! f"U"; "Re"gI
p2 = ListLinePlot [ip, AxesLabel ! f"U"; "Im"gI
p3 = ListLinePlot [ht, AxesLabel ! f"Re"; "Im"gI
Print [GraphicsGrid ffp1; p2; p3gg, ImageSize ! Full,
Frame ! True,
FrameStyle ! Directive [Blue, Dotted], PlotLabel !
sPlotTt1
k;
fk; 1; 4g
I
sEndTime = DateString[Join[f"End time: nt"g,
sTimeStampForm, f"nn"gI
sTimeUsed =
"CPU time used (min): nt" <> ToString
[(TimeUsed[]-fTimeUsed0)/60.0] <> "nn"IWriteString[strm,
sStartTime, sEndTime, sTimeUsed];
(*Print time stamps for performance review*)
WriteString[strm, writeNum["m", m, "mass per unit
length"] <>
writeNum["a", a, "location of elastic axis"]<>
writeNum["b", b, "half-chord length"] <> writeNum["1",
1,
"wing span"]<>
writeNum["S", S]<> writeNum ["I alpha", I]<> writeNum

["GJ", GJ]<>
writeNum["EI", sciNumOut[EI]] <> writeNum["mt", mt] <>
writeNum
"S t"; S t<>
writeNum "I yt"; Iyt <> writeNum "I t"; It <> writeNum
["ghi", ghi]<>
writeNum ["g-alpha-i", gi] <> writeNum ["steu", steu]
<>
writeNum ["numu", numu] <> writeNum ["rho", , ""]<>
writeNum
["kappa",]];
Close [strm];
NotebookPrint [EvaluationNotebook[], nbDir <> sPDFFile];
(*Save nb file with graphs as PDF*)
sWinTt1 =": Done! See log file.";
Root Loci with U as the parameter:1st Torsion
Re[] Im[] Im[]

0.4
14 14
0.3 13 13
0.2 12 12
0.1 11 11
U U Re[]
200 400 600 800 1000 200 400 600 800 1000 0.1 0.2 0.3 0.4
Root Loci with U as the parameter:1st Bending

Re[] Im[] Im[]
U
200 400 600 800 1000 8.8 8.8
0.5 8.6 8.6
8.4 8.4
1.0 8.2 8.2
8.0 8.0
1.5 7.8 7.8
U Re[]
2.0 200 400 600 800 1000 2.0 1.5 1.0 0.5
Root Loci with U as the parameter:2nd Bending
Re[] Im[] Im[]

U 54.4 54.4
200 400 600 800 1000 54.2 54.2
0.005
0.010 54.0 54.0
53.8 53.8
0.015
53.6 53.6
0.020
53.4 53.4
0.025 U Re[]
0.030 200 400 600 800 1000 0.030 0.025 0.020 0.015 0.010 0.005
Root Loci with U as the parameter:3rd Bending
Re[] Im[] Im[]

U 37.6 37.6
200 400 600 800 1000
37.4 37.4
0.05 37.2 37.2
37.0 37.0
0.10 36.8 36.8
36.6 36.6
U Re[]
0.15 200 400 600 800 1000 0.15 0.10 0.05
Notes and Comments
It is humbling to note that in spite of all the theory developed, we still cannot answer
some of the simplest of questions. For example: which mode is going to flutter first
without carrying out a computer program for the given parameters? The dependence
of the flutter speed on the parameters is just a little too complicated. We do have,
however, a closed-form formula for the divergence speed which can give some idea
of the range of the flutter speed but not much more. This is further confounded by
the fact that in the case of axial flow treated in Chap. 10 the divergence speed is
simply not defined!
The Possio integral equation, which is the heart and soul of our theory, was
derived by Possio in 1938 [37] for the oscillating wing; it was customary at
that time to distinguish between the steady oscillatory motion and the transient
unsteady motion, a distinction that persists in the aeroelastic literature even today.
The Laplace Transform version that encompasses both was given in 2003 in [4].
It is not merely a matter of replacing i ! by . The Fourier transform integrals in the
original Possio version are not convergent whereas the Laplace transform integrals
are shown to be convergent and the Fourier transform version obtained by taking
limits as in modern theory [10, 41]. The well-known book by Fung [47] claimed
that the existence of a solution was proved but no reference was given. It was not
until 1976 that an unequivocal statement appeared in [49] in the negative but the
Fung assertion was believed by most of the aeroelasticians. In fact when I began
my research many of my colleagues in aeroelasticty would ask me, Why are you
doing this? It is all known! Moreover, a successful computational algorithm by
Rodden [36] has replaced the analytical version in practice even though no proof of
convergence has been given.
A radically new version that employs Fourier transforms in the space domain was
given in [4] with a simple explicit function (in contrast to the two-page statement
involving Hankel functions in [6]). In particular this shows the lack of analyticity
in M and k (the normalized frequency) thereby questioning many of the early
approximations in M and in k reported in [6] and references therein. In [4] it is
proved that there is a unique solution in Lp , 1 < p < 2 for each M for small
enough jkj. A time domain abstract version was given in 2007 in [5]. Finally the
nonlinear version was given in [14].
The basic idea is that the calculation of the pressure jump given the structure
normal velocity is embodied in the Possio equation and effectively replaces the
Euler field equation insofar as the structure dynamics is concerned.
We have thus an inputoutput problem characterized by the Possio equation,
the recurrent theme throughout this work.
The Possio equation and its various extensions including the nonlinear time
domain version are used systematically in this work, even for the incompressible
case in place of the classical theory in [6] which draws on the pioneering work of
Theodorsen. The equation was mostly ignored in the literature, with no mention
Notes and Comments 267
in the recent standard references [17] or [5], for example. It is ironic that Possio
was killed in the last Allied air raid on Turin, Italy, in World War II. Indeed even
his name would seem to be largely forgotten in his country of birth because of the
perception that he was a collaborator!
Chapter 6
Nonlinear Aeroelasticity Theory in 2D
Aerodynamics: Flutter Instability as an LCO
6.1 Introduction
In this chapter we return to the full nonlinear aeroelastic problem as stated in

Chap. 3, with the structure models both linear and nonlinear, described in Chap. 2
and the isentropic aerodynamics as treated in Chap. 3 with the flow tangency and
the KuttaJoukowsky boundary conditions. Recall that we use continuum models
without immediately approximating them by finite-dimensional models as in all the
current aeroelastic literature.
Our major result is a characterization of Flutter as a limit cycle oscillation with
the flutter speed as a Hopf bifurcation point determined by the linear equations,
obtained by linearizing the nonlinear equations about the equilibrium or rest
structure state and constant air flow. The point of departure is the inputoutput
point of view and the key role is played by the Possio equation which we need to
generalize to the nonlinear case, perforce in the time domain rather than the Laplace
domain. We limit the theory to 2D air flow because almost no results are available
for the Possio equation except in this case. For the same reason the angle of attack
is taken to be zero. Furthermore we consider only M such that 0 < M < 1: The
case M D 0 is treated in Chap. 10.
A basic assumption is that the structure displacements are neglible compared to
the air displacement in the same time. This means that the structure is essentially not
moving compared to the air, which is implicit in the statement of the fluid-structure
boundary conditions.
As in the time invariant case in Chap. 4, which we follow closely, we provide a
constructive existence theorem where the main tool is the power series expansion
in terms of the structure variables. This is unique to our approach and is consistent
with our view that our interest is in the structure dynamics and how it is affected by
the air flow primarily, and in the air flow per se only secondarily.

270 6 Nonlinear Aeroelasticity Theory in 2D Aerodynamics
6.2 The Aeroelastic Equations: Linear Structure Model
We begin with a statement of the full dynamics with the linear Goland structure
model. Thus we have:
R y/ C S .t;
mh.t; R y/ C EI h0000 .t; y/ D L.t; y/
Z b
D p.t; x; y/dx; 0 < y < `; (6.1)
b
R y/ C S h.t;
I .t; R y/ GJ 00 .t; y/ DM.t; y/
Z b
D .x a/p.t; x; y/dx
b
0 < y < `; (6.2)
with CF or FF end conditions.
p.t; x; y/ D 1 .t; x; y/: (6.3)
For each y; 0 < y < `, the 2D aerodynamic equations in x; z:
q.t; x; z/ D r.t; x; z/;

D @ 1
.t; x; z/ D D C r r; (6.4)
Dt @t 2
@2 @ @
C jrj2 C . 1/
@t 2 @t @t

1 2 jrj2
D a1 1 C
2
2
U jrj 2
r r ;
2a1 2
1 < x; z < 1; excepting z D 0; (6.5)
where we consider only:
U1
0 < M < 1I M D :
a1
(Omit M D 0 and M D 1.)

The boundary conditions are:
1. Flow Tangency.
6.2 The Aeroelastic Equations: Linear Structure Model 271
Allowing for flow discontinuity between the top and bottom of the wing and again
assuming wing displacement to be small compared to that of the air.
@ @1 h
.t; x; 0C/ D P y/ C .x a/P .t; y/
C .1/ h.t;
@z @z

@
C .t; x; 0C/.t; y/ ; jxj < b: (6.6)
@y
And

@ @1 P y/ C .x a/.t;
P y/
.t; x; 0/ D C .1/ h.t;
@z @z

@
C .t; x; 0/.t; y/ ; jxj < b: (6.7)
@x
2. KuttaJoukowski Conditions:
The pressure jump is zero off the wing:
.t; x/ D 0 jxj > b: (6.8)
The Kutta condition at trailing edge
.t; x/ D 0 as x ! b :
This completes the description of the problem.
The Aero-Structure Dynamics/InputOutput Problem
From a general viewpoint we may consider the fluid-structure interaction problem as

an inputoutput problem where the structure velocity wa .t; :/ (the downwash)
can be considered the input and the pressure jump p.t; :/ as the output. The lift and
moment are just linear functionals of p.t; :/.
Heuristically, we can then invoke the Duhamel principle, familiar in electro-
magnetic theory or more generally (see [1, 44]), as the response to a boundary
input in wave motion. Thus p.t; :/ must be a physically realizable response, albeit
nonlinear, extending the theory from the steady-state case in Chap. 4.
Power Series Expansion/Flow Decomposition
The far field potential is

1 .t; x; y; z/ D xU ;
where U is the speed parameter, U 0. We omit the subscript 1.
As in Chap. 4, we consider the solution potential .; t; x; y; z/ corresponding
to h.t; y/; .t; y/ where is a scalar-scale parameter, and make it 0 jj 1;
yielding the expansion:
1
X k
.; t; x; z/ D k .t; x; z/; (6.9)
k
kD0
where
0 .t; x; z/ D 1 .t; x; z/ D xU ;
@k
k .t; x; z/ D .0; t; x; y; z/; k 1;
@k
where we refer to k .:/ as the kth-order potential.

We calculate these by taking derivatives in the field equation with respect to .
We see that 1 .t; x; z/ satisfies the linear field equation:
.1 / D 0; 1 < x; z < 1; omitting the wing;
where
2
@2 @2 2
2 @ 2 @
./ D C 2U a1
2
1 M C a1 : (6.10)
@t 2 @t@x @x 2 @z2
It is interesting to note that the constant does not appear in this equation, and
hence if we are interested only in the linear equation we may take D 1:
More generally, for k ; k > 1, as in Chap. 4, we expand both sides of (6.5) in a
power series, obtaining:
0 1
1
X k 2 1 1 1 X1
@ k @ @ X X k j A
X k j @j
C qk qj C . 1/ k
k @t 2 @t j D0
k j j D1
k j @t
kD1 kD0 kD1
1
!0 0
1 X 1 1
11
X k 1 X i j qi :qj X i
@
D a1
2
k @1 C @ U
i AA
k 2a12 i j @x
kD1 i D1 j D1 i D1
X 1 X
1 X 1
1 @n @m @2 p @m @2 p
.nCmCp/ C
nD0 mD0 pD1
nmp @x @x @x 2 @z @x@z

@n @m @2 p @m @2 p
C C ; (6.11)
@z @z @z2 @x @x@z
where
@2 @2
D C ;
@x 2 @z2
qk D rk ;
@i @j @i @j
qi :qj D C :
@x @x @z @z
These of course reduce to the equations in Chap. 4 upon setting the time derivatives
to zero.
Theorem 6.1. The kth-order potential satisfies the linear nonhomogeneous equa-
tion
.k / D gk1 ;
where gk only involves potentials of order k, and g0 D 0.
0 1
@ @X X
k1 k1
@j
gk1 D Ck;j qkj qj A Ck;j kj
@t j D1 j D1
@t
X
k1
@kj
C .1 /U Ck;j j
j D1
@x
1 XXX 1 1 1
C k qi qj m; i Cj CmD k
2 i D1 j D1
i j m
kD1
XXX
k @n @m @2 p
C
nD1 mD1 pD1
nmp @x @x @x 2
@n @m @2 p @n @m @2 p

C C
@x @z @x@z @z @x @z@x

@n @m @2 p
C ; nCmCp Dk
@z @z @z2
X
k1
@kp @2 p @kp @2 p
2U Ck;p 2
C :
pD1
@x @x @z @z2
Proof. We go back to (6.12). Note that we can write

0 1 0 1
1 1 1 1 X1
@ @X X k j X k 2
@ @ X k j

qk qj A D 2U @ qk qj A :
k
C
@t j D0
k j k @x@t @t j D1
k j
kD0 kD1 kD1
And
X 1 X
1 X 1
1 .nCmCp/ @n @m @2 p @m @2 p
C
nD0 mD0 pD1

@n @m @2 p @m @2 p
C 2
C
@z @z @z @x @x@z
as
1
X X1
@2 p @1 @2 p @1 @2 p
U2 p 2U C pC1
pD1
@x 2 pD1
@x @x 2 @z @x@z
X 1 X
1 X 1
1 .nCmCp/ @n @m @2 p @m @2 p
C
nD1 mD1 pD1

@n @m @2 p @m @2 p
C 2
C :
@z @z @z @x @x@z
Hence (6.12) can be expressed:

0 1
1
X k 2 1
X k 2 1 X
X 1 m j
@ k @ k @ @
C 2U C qm qj A
k @t 2 k @x@t @t mD1 j D1 m j
kD1 kD1
1 X
X 1
m j @j
C . 1/ 4m
mD1 j D1
m j @t
1
X 1
X 1
X
k k
@2 k
2 kC1 @1 @2 k @1 @2 k
D 2
a1 4k U 2U C
k k @x 2 k @x @x 2 @z @x@z
kD1 kD1 kD1
X1 X 1 1 1 1
j k 1 X X X i j k
C .1 / j q1 qk C qi
j D1
j k 2 i D1 j D1 i j k
kD1 kD1
X 1 X
1 X 1
1
qj k .nCmCp/
nD1 mD1 pD1
nmp

@n @m @2 p @n @m @2 p @n @m @2 p
2
C C
@x @x @x @x @z @x@z @z @x @z@x
1 1
X X n p @n @2 p
@n @2 p
C 2U C : (6.12)
pD1
n p @x @x 2 @z @z2
kD1
Hence collecting the coefficients of k , we obtain
.k / D gk1 ; (6.13)
where
0 1
@ @X X
k1 k1
@j
gk1 D ck;j qkj qj A Ck;j kj
@t j D1 j D1
@t
X
k1
@kj 1
C .1 /U Ck;j j C k
j D1
@x 2
XXX 1 1 1
qi qj m ; i Cj CmDk
i D1 j D1
i j m
kD1
XXX
k @n @m @2 p @n @m @2 p
C 2
C
nD1 mD1 pD1
nmp @x @x @x @x @z @x@z

@n @m @2 p @n @m @2 p
C C
@z @x @z@x @z @z @z2
X
k1
@kp @2 p @kp @2 p
2U Ck;p C ; n C m C p D k:
pD1
@x @x 2 @z @z2
(6.14)
Note that gk depends only on j ; j k.

In particular for k D 1,
@ @1 @1
g1 D 2 jq1 j2 2 1 C 2.1 /U 1
@t @t @x

@1 @2 1 @1 @2 1
4U C :
@x @x 2 @z @z2
t
u
Next we deduce the corresponding boundary conditions.
Boundary Conditions
Flow Tangency

@ @
.; t; x; y; 0C/ D .h.t; y/ C .x a/.t; y//
@z @t
X1
@.; t; x; y; 0C/ @
C .t; y/ k k .t; x; 0C/;
@x @z
kD1

@
D .h.t; y/ C .x a/.t; y//
@t
1
X @
k k k1 .t; x; y; 0C/.t; y/:
@x
kD1
Hence

@ @ @
1 .t; x; 0C/ D .h.t; y/ C .x a/.t; y// U.t; y/ k .t; x; y; 0C/
@z @t @z
@
Dk k1 .t; x; 0C/.t; y/ for k 2;
@x

@ @ @
1 .t; x; y; 0/ D .h.t; y/ C .x a/.t; y// U.t; y/ k .t; x; y; 0/
@z @t @z
@k1 .t; x; y; 0/

Dk .t; y/ for k 2:
@x
Hence

@ @
1 .t; x; y; 0/ D .h.t; y/ C .x a/.t; y// U.t; y/; (6.15)
@z @t
@ @k1 .t; x; y; 0/
k .t; x; y; 0/ D k .t; y/ for k 2: (6.16)
@z @x
KuttaJoukowsky Conditions
It is important to note that we need the KuttaJoukowsky conditions in the nonlinear

case as well for uniqueness of solution.
For this we need to consider as before the acceleration potential first.
We have
@ 1
.:/ D C r r ;
@t 2
1
X @k 1
X 1 1
k
U2 k @k X X j k
.; t; x; y; z/ D C CU C qk qj :
k @t 2 k @x j D1
j k
kD1 kD1 kD1
With
@k .0; t; x; z/
k .t; x; z/ D ;
@k
we have the expansion
1
U 2 X k
.; t; x; z/ D C k .t; x; z/; (6.17)
2 k
kD1
@k X
k1
@k
k .t; x; z/ D CU C Ck;j qkj qj ; (6.18)
@t @x j D1
where Ck;j are the binomial coefficients.
1
X k
.; t; x; 0/ D k .t; x; 0/; (6.19)
k
kD1
.; t; x; 0/ D 0 jxj > b

implies that
k .t; x; 0/ D0 for jxj > b for every k:
Next define: (Kussner doublet function)
A .t; x/ D .t; x; 0/=U (6.20)
and for every k:

Ak .t; x; y/ D k .t; x; 0/=U: (6.21)
Then
@1 @1
.U /A1 .t; x/ D 1 .t; x; 0/ D CU : (6.22)
@t @x
And from the boundary conditions
@1 @1 .t; x; 0C/

D2 ;
@t @t
@1 @1 .t; x; 0C/
D2 :
@x @x
As in the time invariant case we consider next
Flow Decomposition
The next important step is to take advantage of the fact that (6.14) is a linear
nonhomogeneous equation with nonzero boundary conditions and decompose the
potentials as
k D L;k C 0;k ;
where 0;k satisfies the nonhomogeneous equation (6.14),
.0;k / D gk1 1 < x; z < 1 (6.23)
(where gk1 depends of course on j , j k 1) and there no discontinuities on

z D 0. Although L;k satisfies the homogeneous equation,
.L;k / D 0 (6.24)
except on z D 0 where it satisfies the boundary conditions given above. Note in

particular that 1 D L;1 .
Furthermore, from the field equation

@2 L;k @2 L;k 2
2 @ L;k
2
2 @ L;k
C 2U a1
2
.1 M / C a1 D 0;
@t 2 @t@x @x 2 @z2
we deduce that
L;k .t; x; z/ D L;k .t; x; z/ z > 0;
because we require that
@ @
L;k .t; x; 0/ D L;k .t; x; 0C/: (6.25)
@z @z
Hence
@ @
L;k .t; x; 0/ D L;k .t; x; 0C/: (6.26)
@x @x
Hence from (6.27)
2
@L;k
D 0:
@x
In as much as
2
@L;k @L;k .:; 0C/ @L;k .:; 0/
D C
@z @z @z

@L;k .:; 0C/ @L;k .:; 0/
;
@z @z
we have
2
@L;k
D 0 by .6:26/:
@z
Hence we define
@
k .t; x/ D L;k .t; x; 0/:
@z
Next we have the crucial result.

Theorem 6.2.
.U /Ak .t; x; y/ D k .t; x; y; 0/ D @L;k =@t C U @L;k =@x
for every k 1: (6.27)
Proof. From (6.18) we have
X
k1
k .t; x; y; 0/ D @k =@t C U @k =@x C Ck;j .qkj qj /:
j D1
And hence we need only to prove that
X
k1
Ck;j .qkj qj / D 0; (6.28)
j D1
or
.qkj qj / D 0:
Now the left side
D qkj .:; 0C/ qj .:; 0C/ qkj .:; 0/ qj .:; 0/: (6.29)
Let k denote @k =@zk denote @k =@x.

Then (6.29),
D kj .:; 0C/ j .:; 0C/ kj .:; 0/ j .:; 0/
C .kj .:; 0C/j .:; 0C/ kj .:; 0/j .:; 0//:
And
kj .:; 0C/ j .:; 0C/ kj .:; 0/ j .:; 0/
D . kj .:; 0C/ C kj .:; 0// j .:; 0C/
. j .:; 0/ C j .:; 0C// kj .:; 0/
D 0 by .6:25/;
and
.kj .:; 0C/j .:; 0C/ kj .:; 0/ j .:; 0//
D .kj .:; 0C/ kj .:; 0//j .:; 0C/ C .j .:; 0C/ j .:; 0//kj .:; 0/;
which by (6.26)D 0.
Hence (6.29) holds, proving the theorem. t
u
Remarks: The importance of (6.27) is, among others, that the formula for the
nonlinear case is the same as for the linear case. We have seen this already in the
time invariant case in Chap. 4 for zero angle of attack. In particular we use it to
calculate the pressure jump:
1
X
p D 1 I D k =k.@k =@t C U @k @x/:
kD1
Hence the boundary conditions satisfied by L;k are:
@ @L;k1 .t; x; 0/
L;k .t; x; 0/ D k .t; y/ for k 2; (6.30)
@z @x
Ak D L;k =U
D .1=U /@k =@t C @k =@x
and
1
X
AD Ak =k (6.31)
kD1
Note in particular that
k .t; x/ D 0; jxj > b for every k: (6.32)
To proceed further we need to determine the time domain functions Ak .:; :/ in X

which we show satisfy the linear time domain Possio equation.
The Linear Time Domain Possio Equation
Let us recall that we began with the field equation for 1 :
@2 1 =@t 2 C 2U @2 1 =.@t @x/ a1

2
.1 M 2 /@2 1 =@x 2 C a1 @ 1 =@z2
D 0
2 2
with the flow tangency boundary condition and the KuttaJoukowsky conditions.
The solution technique was to take the Laplace transform in the time domain and
the Fourier transform in the space domain and obtain the Laplace transform version
of the Possio integral equation. Showing that the solution is the Laplace Transform
of a time domain function can be nontrivialsee [10, 41]. Indeed the basic problem
of when a function analytic in a right-half-plane is the Laplace transform of a time
domain function is still largely an open question. For M D 0, for example, we had
to invoke a special result due to Sears.
Here we obtain the time domain version by formally inverting the Laplace
domain equation. This, it should be noted, would not be possible without the use of
the function a.M; s/ introduced in Chap. 5. We begin with the version:
p
2WO a .; :/ D .1 M 2 /PH.I C B.k//A.;
O :/; (6.33)
where the operator B.k/, it will be recalled, is defined by

Z 1
p
B.k/A D .kR.k; D/A/=. .1 M 2 // kR.ks; D/A a.M; s/ds;
2
Z 1
O x/ D
A.; et A.t; x/dt Re: > a ; jxj < b;
0
where A.:; :/ is in X . Here
O :/ is the function in Lp .R1 /W

R.k; D/A.;
Z x

O /d;
e U .x/ A.; 1 < x < 1;
1
whose inverse Laplace transform is
Z x Z t
. .x /=U /A.t ; /d d
1 0
Z tZ x
D . .x /=U /A.t ; /dd
0 1
Z t
D A.t ; x U/d: (6.34)
0
We note the mixing up here between the time and space variables.
O :/ is the Laplace transform of
Hence R.k; D/A.;
Z t
S.U/A.t ; :/d:
0
O :/. By the usual rules

Next we need to take the inverse transform of kR.k; D/A.;
the candidate would be
Z t
d=dt S.U/A.t ; :/d in 0 < t; (6.35)
0
which if A.t; :/ is in the domain of D for t > 0 and DA.t; :/ is in X can be defined
as:
Z t
D A.t/ C U PS.U t U/DA.; :/d; 0<t
0
and would be ok. But in our case, we simply cannot assume that A.t; :/ is in
the domain of D. Instead we require that A.t/ be absolutely continuous in t with
derivative in X .
We can then state the time domain Possio integral (TDP) as an equation in X :
Z b h
p
2=. .1 M 2 //wa .t; x/ D 1= 1=.x / A.t; /
b
Z t
p
@=@t ..A.t ; U//=. .1 M 2 //
0
Z 1 i
C a.M; s/A.t ; U=s/ds=s/d d;
2
jxj < b:
And the TDP in abstract form is:
h
p
wa .t; :/ D . .1 M 2 //=2PH A1 .t; :/
Z th
p
d=dt .PS.U.t //PA1 .; ://=. .1 M 2 //
0
Z 1 ii
C PS..U.t //=s/PA1 .; :/a.M; s/ds=s d; (6.36)
2
where the sense in which the integral is defined to take care of the singularity at
s D 0 needs elaboration:
The integral
Z tZ 1
S..U.t //=s/A1 .; :/a.M; s/ds=s d
0 2
is interpreted in the Cauchy sense:
Lim ! 0
Z tZ
S..U.t //=s/A1 .; :/a.M; s/ds=s d
0
Z tZ 1
M
C p S.U=s/A1 .t ; :/ds=s d
.1 M 2 / 0
Z tZ
M
C p S.U=s/A1 .t ; :/ds=s d;
.1 M 2 / 0 2
so that
Z t Z 1
1= S.U=s/A1 .t ; :/ds=s d
0 2
Z 1 Z t =s
D 1= ds 1= .S.U / S.U //A1 .t s; :/d
0 0
Z 1 Z 1
D HA1 .t/ 1= 1= .S.U / S.U //.A1 .t/ A1 .t s//dt ds
0 t =s
Z t
D PHA.t; :/ d=dt PH B.t /A./d;
0
where
Z 1
p
B.t/f D .S.U t/f /=. .1 M 2 // C S.U t=s/f a.M; s/ds=s:
2
Next let us denote the Possio equation by
.M /A D w
corresponding to the linear Possio equation:

Z b h Z t
p
1M =.2/ 2
1=.x / A.t; / @=@t .A.t ; U//
b 0
. Z 1
p
. .1 M 2 // C .a.M; s/ a.M; 0//A.t s; U/ds
1
Z 2 i
C a.M; 0/A.t s; U/ds d d
1
D wa .t; :/ (6.37)
with domain in X and range in X , with the Kutta condition in addition (implicit
everywhere, if not specified explicitly otherwise).
If we assume the time domain version has a solution, then of course we may take
the Laplace transform which will then satisfy the LDP.
Let us explore next how far we can go in terms of solving the time domain Possio
(TDP), inverting the Laplace Transform versions in Chap. 5.
Solving the Linear TDP Equation
We denote the solution of the Linear TDPwhen it has a unique solutionby A D

.M /wa where we have seen that in Sect. 5.5 the Laplace transform of the solution
has the form
p
AO D .I C PB.k/P C W .k//1 2=. .1 M 2 //T wO a : (6.38)
Or of the form (5.48)
p ! !1
.1 M 2 / 2
0 < M < 1 W AO D I C PH.B.k/ B.1// wO a : (6.39)
M M
We begin with the version (6.38).

Lemma 6.3. For A in Lp b; b
; 1 < p < 2: .I C PB.k/P C W .k//A is the
Laplace transform:
Z 1
D ek t d=dtP.B.t/ C W.t//Adt;
0
p
where PW.t/A is the function 1= ..b x/=.b C x//
h Z t
p
1=. .1 M 2 // A.b t C /d
0
Z tZ 1
p
C a.M; s/A.b .t /=s/ .=.2bs C //ds=..b x/s C /
0 0
p
C a.M; s/A..t /=s b/ ..2bs C /=/ds=..b C x/s C /
Z 2
p
C a.M; s/A..t /=s b/ ..2bs C /=/ds=..b C x/s C /d
1
Z 2
C A.t=s .2b C x//a.M; s/ds; jxj < b
0
and
Z 1
p
P B.t/PA D 1=. .1 M 2 //P S.t/PA C P S.t=s/PAa.M; s/ds=s :
2
Proof. To derive Z 1
W .k/A D P
ek t W.t/Adt
0
t
u
we use
p
1=. .1 M 2 //h .k/L .k/
Z 1 . h
p p
D ek t dt 1 . .1 M 2 // 1= ..b x/=.b C x//
0
Z b i
p
..t b C /=.t C b C //1=.t .x // A./d;
b
LC .ks; A/hC .ks; x/

Z 1 Z b Z t
D ek t A./d .t s.b C //1=
0 b 0
p p
..b x/=.b C x// .=.2bs C //d=..b C x/s C /dt
Z 1 Z b
p
D ek t A./d 1= ..b x/=.b C x//
0 b
p
..t s.b C //=.2bs C t s.b C ///1=..b C x/s C t s.b C //
Z 1 Z b
k t p
D e A./d 1= ..b x/=.b C x//
0 b
p
..t s.b C //=.t C s.b ///dt=.t C s.x //:
And following the notation in Chap. 5

Z 2
LC .ks; A/hC .ks; x/a.M; s/ds
0
Z 1 Z b Z 2
p
D ek t dt A./d 1= ..b x/=.b C x//
0 b 0
p
..t s.b C //=.t C s.b ///.a.M; s/ds/=.t C s.x //
Z 1 Z b
k t p
D e dt 1= ..b x/=.b C x// A./d
0 b
Z 2
p
..t s.b C //=.t C s.b ///.a.M; s/ds/=.t C s.x //;
0
and
Z 1
L .ks; A/h .ks; x/a.M; s/ds
0
Z 1 h Z b Z 1
p p
D ek t dt 1= ..b x/=.b C x// ..t sb C s/
0 b 0
. i
.t C sb C s//.a.M; s/ds/=.t s.x // A./d:
Using
Z 1
p p
h .ks; x/ D ek t 1= ..b x/=.b C x// .t=.2bs C t//dt=..b x/s C t/;
0
Z 1 Z b
L .ks; A/ D ek t dt .t s.b //A./d
0 b
and
Z b
j.ks/LC .ks; A/ D eks.bCx/ eks.bC/ A./d
b
Z b
D eks . x/A./d
b
Z 1 Z b
k t
D e dt .t s. x//A./d;
0 b
we have
Z 2
j.ks/LC .ks; A/a.M; s/ds
0
Z 1 Z b Z 2
D ek t dt .t s. x//a.M; s/ds A./d:
0 b 0
Hence W.t/A is the function

Z b Z 2
D .t s. x//a.M; s/ds A./d
b 0
Z b
p
C1= ..b x/=.b C x// A./d
b
Z 2
p
..t s.b C //=.t C s.b ///
0
p
.a.M; s/ds/=.t C s.x // C 1= ..b x/=.b C x//
Z b Z 1
p
..t sb C s/=.t C sb C s//.a.M; s/ds/=.t s.x //A./d
b 0
p p
C1=. .1 M 2 //1= ..b x/=.b C x//
Z b
p
..t b C /=.t C b C //1=.t .x //A./d:
b
Next for A in Lp b; b

,
Z 1
PB.k/PA D ek t P B.t/PAdt;
0
Z 1
p
PB.t/PA D 1=. .1 M 2 //PS.t/PA PS.t=s/PA a.M; s/ds=s:
2
Hence d=dt P.B.t/ C W.t//A is given by

Z t
d=dt P B.t /PA.; :/d
0
Z Z Z !
t b 2
C d=dt d . s. x//a.M; s/dsA.t ; /d
0 b 0
Z b Z 2
Cd=dt a.M; s/A.t s. x/; /dsd
b 0
Z tZ b Z 2
p p
C1= ..b x/=.b C x//d=dt .. s.b C //=. C s.b ///
0 b 0
.a.M; s/ds/=. C s.x //A.t ; /d

Z t Z b Z 1
p p
C1= ..b x/=.b C x//d=dt d .. sb C s/=. C sb C s//
0 b 0
.a.M; s/ds/=. s.x //A.t ; /d

p p
C1=. .1 M 2 //1= ..b x/=.b C x//d=dt
Z t Z b
p
d .. b C /=. C b C //1=. .x //A.t ; /d:
0 b
(6.40)
Hence
.I C PB.k/P C W .k//w.k;
O :/
Z 1 Z t
D ek t w.t; :/ C d=dt .P B. /P C W. //w.t ; :/d dt:
0 0
(6.41)
Hence the TDP can be expressed:

Z t
2
A.t; :/ C d=dt .P B. /P C W. //A.t ; :/d D p T wa .t; :/: (6.42)
0 1 M2
Theorem 6.4. The TDP (6.42) has at most one solution. Any solution of the TDP
(6.42) will satisfy the Kutta condition.
Proof. The Laplace transform of any solution satisfies the LDP which we have
shown has a unique solution for small enough (Theorem 5.16).
Furthermore we have shown that the Laplace transform satisfies the Kutta
condition. t
u
Finally, from the form (6.39):
p !
.1 M 2 /
IC O :/ D 2 w.;
PH.B.k/ B.1// A.; O :/;
M M
p
we have (strong) limit Re ! 1 I C . .1 M 2 / =M PH.B.k/ B.1/// D I .
Hence the time domain transform has a delta function at the origin for nonzero M .
(For M D 0 we also have the delta function derivative. Thus M D 0 is special, not
obtained as the 1imit for small M .) Hence, if the TDP has a solution for M > 0; it
is of the form
Z t
2
A.t; :/ D wa .t/ C P .M; t /wa .; :/d
; (6.43)
M 0
where the Laplace transform:

Z 1
p
ek t P .M; t/wdt D .I C . .1 M 2 //=M PH.B.k/ B.1///1 I
0
and P .M; :/ does not have a delta function at the origin. This is useful to us below
in isolating the non-circulatory terms in the aeroelastic structure equations.
It is interesting to consider what happens for k D 0, in (6.39). We have:
p !1
.1 M 2 /
IC PH.B.1// I
M
(because B.0/ D 0)
p
D M=. .1 M 2 //T I;
which checks with the solution in the time invariant case
@L;k1 .t; x; y; 0/
D k .t; y/: (6.44)
@x
Let us start with k D 2.

The potential 1 .:/ is given by specializing (5.18) to the typical section and
D0W QO
.; QO i !/erz=.k C i !/
i !; z/ D 1=2A.; z > 0;
where
p
r D .M 2 k 2 C 2M 2 ki ! C .1 M 2 /! 2 /:
Hence the LaplaceFourier transform of .@L;1 .t; x; y; 0C//=@x is
D 1=2AQO1.; i !/i !=.k C i !/
D 1=2.1 k=.k C i !//AQO1 .; i !/
D 1=2.1 =. C i !U //AQO1 .; i !/:
Hence
Z t
.@L;1 .t; x; 0C//=@x D 1=2 A1 .t; x/ d=dt A1 .t ; x U/d ;
0
where we have used

Z 1

1=. C i U !/ D 1=U e U x ei !x dx:
0
Hence Z t
@1 =@x D A1 .t; x/ C d=dt A1 .t ; x U/d: (6.45)
0
For k > 1: we have again:
1 @
Ak .t; x/ D L;k C L ; jxj < b; (6.46)
U @x
we omit y being now only a fixed parameter.

Or
@k =@t D U @k =@x UAk .t; x/; 0 < t; jxj b: (6.47)
Or, solving the equation, we have

Z t
k D U S.U/Ak .t /d
0
and Z t
@k =@x D Ak .t; x/ C d=dt Ak .t ; x U/d: (6.48)
0
In (6.47), because
A.t; x/ D 0 for jxj > b;
we have that the integral is zero for U t > .b C x/ and hence for t > .2b/=U
Z .bCx/=U
@k =@x D Ak .t; x/ C d=dt Ak .t ; x U/d: (6.49)
0
We omit the subscript L from now on. We start with
QO k .; i !; z/ D erz =r QO

.; i !; 0C/ z > 0; (6.50)
where
p
r D .M 2 2 C 2M 2 i! C .1 M 2 /! 2 /;
D =U
and
QOk .; i !; z/ D erz =r QO k .; i !; 0/ z < 0; (6.51)
where
@=@zOk .; x; z/ D O k .; x; z/:
Hence with
@=@x Ok .; x; z/ D Ok .; x; z/;
we have
QOk .; i !; 0C/ D i !=r QO k .; i !; 0C/; (6.52)
QOk .; i !; 0/ D i !=r QO k .; i !; 0/; (6.53)
which generalize (6.61) and (6.62).

Again:
Ak .t; x/ D 1=U @L;k =@t C @L;k =@x jxj b:
The transform version of which using (6.52) and (6.53) yields:

AMOk .; i !/ D . C i !/ QOk .; i !; 0C/ QOk .; i !; 0/

D . C i !/=r MO k .; i !; 0C/ C MO k ; .; i !; 0/ :
Or
1 r MO
Ak .; i !/ D 1=2 MO k .; i !; 0C/ C MO k .; i !; 0/ : (6.54)
2 C i!
For k > 1, we have to respect the possible discontinuity in k .t; x; z/ at z D 0.

Equation (6.54) is then the version of the linear Possio equation valid for k 2,
recalling that .1 r/=.2 C i !/ is the multiplier corresponding to the time domain
operator .M / on into .
Hence we have the time domain equation in
.M /Ak D gk ;
gk .t; x/ D 1=2. k .t; x; 0C/ C k .t; x; 0// jxj < b:
The right side by the boundary condition (6.31)
D 1=2.k/.t/.k1.t; x; 0C/ C k1 .t; x; 0C// for jxj < b:
The presence of the factor .t/ which depends on t makes use of the Laplace
transform difficult unlike the time invariant case in Chap. 4 where is a constant.
Hence we need to work in the time domain from now on. But first we have the
frequency domain relations.
By (6.52) and (6.53)
r= i ! MOk1 .; i !; 0C/ D MO k1 .; i !; 0C/: (6.55)
Similarly
.r/= i ! MOk1 .; i !; 0/ D MO k1 .; i !; 0/: (6.56)
Hence, adding, we have:

r= i ! MOk .; i !; 0C/ MOk .; i !; 0/ D MO k .; i !; 0C/ C MO k .; i !; 0/ ;
(6.57)
where we note that
k .t; x/ D .k .t; x; 0C/ k .t; x; 0//

D0 for jxj > b:
And by (6.54)
i ! r MO i ! MO
MOk D Ak .; i !/ D Ak .; i !/: (6.58)
r C i! C i!
Hence
AMOk .; i !/:

MOk D 1 C
C i!
Or in the time domain:
Z t
k .t; x/ D Ak .t; x/ C d=dt Ak .t ; x U/d; jxj b;
which is simply the solution of the partial differential equation (6.31).

We now define the time domain operator on the subdomain of which is
absolutely continuous in t, with derivative in , into by
Z t
f D gI g.t; x/ D f .t; x/ d=dt f .t ; x U/d; jxj < b:
0
Then we have
k D Ak : (6.59)
Note that
f D 0 implies f D 0:
Hence we can define the inverse 1 by
Z x
1 d
1 f D gI g.t; x/ D f .t; x/ f .t; s/ds; (6.60)
U dt b
on the same domain as that of and in particular:

Z x
Ak .t; x/ D k .t; x/ C 1=U d=dtk .t; s/ds; jxj < b;
b
where the necessary differentiability in t of k .t; s/ is assumed.

Also we need now to define the linear operator with domain and range in
f D hI h.t; x/ D .t/f .t; x/; jxj < b where .t/ is the given torsion angle that
does not depend on the chord variable x.
We only need to consider the case where the torsion angle is bounded (actually
< 1) that is actually linear bounded.
Then we have
2.M /Ak D kP k1 ; (6.61)
where we use
k .t; x/ D 1=2.k .t; x; 0C/ C k .t; x; 0//
and similarly we use the notation:
k .t; x/ D . k .t; x; 0C/ C k .t; x; 0//:
Again, subtracting (6.42), from (6.41) we have:

MOk1 .; i !; 0C/ C MOk1 .; i !; 0/

D i !=r MO k1 .; i !; 0C/ MO k1 .; i !; 0/ : (6.62)
The boundary conditions yield
k1 .t; x/ D .k 1/.t/k2 .t; x/; jxj < b;
k1 .t; x/ D .k 1/.t/ k2 .t; x/; jxj < b:
Now i != r is a Mikhlin multiplier for each . In fact we have the follow-

ing; see [55].
Lemma 6.5.
Z 1
i !=.r/ D O
ei !x L.; x/dx; 1 < ! < 1;
1
where
O
L.; O
x/ D L.; x/;

O M M
L.; x/ D K1 x x>0
.1 M 2 /3=2 1 M2
and the inverse Laplace transform

" 2 !#
M .
d p M x
L.t; x/ D t t
2
;
.1 M / dt
2 2 1M U
2
M x M x
t> D0 0<t < x > 0;
1 M2 U 1 M2 U
L.t; x/ D L.t; x/:
Let L.t/ denote the bounded linear operator on Lp .R1 / into itself defined by
Z 1
L.t/f D gI g.x/ D L.t; x s/f .s/ds; jxj < 1:
1
And let L.M / be the corresponding operator on into defined by

Z t
L.M /A D gI g.t; :/ D PL.t /A.; M /d: (6.63)
0
Then we have:
k1 D L.M /.k 1/k2 : (6.64)
Next by (6.58) and (6.59),
i ! MO i!
Ak .; i !/ D M k .; i !; 0C/ C MO k .; i !; 0/ :
C i! r
Or

AMOk .; i !/ D MO k .; i !; 0C/ C MO k .; i !; 0/ :
r

C i!
Hence 2.M /Ak D P k D kP k1 which by (6.64) D kL.M /.k
1/k2 . Hence using (6.61)
D k.k 1/L.M /Ak2 :
Hence we obtain
2.M /Ak D k.k 1/L.M /Ak2 :
Or under the assumption of unique solution to the Possio equation:
Ak D k.k 1/1=2.M /L.M /Ak2; (6.65)
which is our recursive relation.

This generalizes the time invariant case in Chap. 4, where 1=2.M /L.M /
corresponds to multiplication by
T PH..y/2 /=.1 M 2 / D ..y/2 /=.1 M 2 /:
An immediate inference from (6.65) is that
A2k D 0
and hence
A2k D 0: (6.66)
Let
1
J D .M /L.M /: (6.67)
2
Then (6.51) yields:
A2nC1 =.2n C 1/ D J n A1
and finally
1
X 1
X
AD Ak =k D J k A1 : (6.68)
kD0 kD0
Assuming that
jjJ jj < 1: (6.69)
Or
Sup j.t; y/j
y
is sufficiently small we have
A D .I J /1 A1 : (6.70)
Of course .t; y/ is determined as a solution of the aeroelastic equation, but in

practice it is indeed much less than one radian, so the convergence holds.
We can express (6.70) as the nonlinear version of the time domain Possio
equation.
6.3 The Nonlinear Possio Integral Equation
This is an equation in Lp b; b

; 1 0 for determining the pressure
doublet function A.t; x; y/, given the structure state variables, h.t; y/; .t; y/. In
other words it replaces the aerodynamics insofar as the structure dynamics is
concerned in isentropic flow. It is the nonlinear inputoutput relation for the system,
the structure dynamics in air flow.
6.3 The Nonlinear Possio Integral Equation 297
We begin with (6.70),

A JA D A1 :
Hence
.M /A .M /JA D wa ;
where
.M /J D L.M /:
Hence
.M /A 1=2L.M /A D wa : (6.71)
Then the nonlinear time domain Possio equation is given by
p Z Z t
.1 M 2 / b
1 @ p
A.t; / .A.t ; U//=. .1 M 2 //
2 b x @t 0
Z 1
C .a.M; s/ a.M; 0//A.t s; U/ds
1
Z 2
C a.M; 0/A.t s; U/ds d d
1
Z tZ b
1=2.t; y/ L.t ; x s/.; y/D.; s/dsd
0 b
C d=dt.h.t; y/ C .x ab/.t; y// C U.t; y/ D 0; (6.72)
where
Z t
d
D.t; x/ D A.t; x/ A.t ; x U/d (6.73)
dt 0
and
. .
d t p
L.t; x/ D.M / ..1 M 2 /2 / q t . .t 2 .M 2 x 2 /
dt t 2 Mx

. Mx
..1 M 2 /2 U 2 ///; for t > ; and D 0
U.1 M 2 /
Mx
0<t < x>0
U.1 M 2 /
L.t; x/ D L.t; x/:
Note that complicated as it is, this completely bypassesthe aerodynamic flow

equationthe nonlinear Euler equation. We are not concerned with the air flow
any more. Note also that the nonlinearity in .:; :/ is physically realizable.
Remarks: We have existence and uniqueness for the solution of (6.72) assuming the
regular Possio equation has a unique solution and the torsion angle is small enough.
The solution is given by (6.53).
We use the qualifier nonlinear to indicate that the solution is no longer linear
in the structure state variables, h.t; y/; .t; y/.
6.4 Nonlinear Aeroelastic Dynamics
Having determined the pressure jump, we can now go on to full nonlinear aeroelastic
dynamics: specialized to the typical section, zero angle of attack, 0 < M < 1, and
linear structure dynamics. We show first that the nonlinear aeroelastic dynamics
formulates as a nonlinear convolution/evolution equation in a Hilbert space, in fact
in the energy space we have already discussed in the previous chapters.
We begin with the series (6.68):
1
X
A D .M /wa C J k .M /wa : (6.74)
kD1
We have already studied the first term leading to the linear convolution evolution
equation in Chap. 5 for nonzero M . Now we have an additional component. We
verify first that there are no additional non-circulatory terms.
Lemma 6.6. There are no non-circulatory terms in
1
X
J k .M /wa : (6.75)
kD1
Proof. We have only to note in
J D 1=2.M /L.M /; (6.76)
the Laplace transform of the kernel in L.M / given by
O
L.k; :/ ! 0 as Re: k ! 1
and therefore there are no delta functions in L.t; :/ and hence J .M /wa has no
non-circulatory components and neither does the sum in (6.76). u
t
6.4 Nonlinear Aeroelastic Dynamics 299
Lift and Moment Calculation
We next calculate the lift and moment corresponding to A(.,.) given by (6.75).
Here we need to bring back the dependence on the structure span variable y.
Thus we have:
wa .t; x; y/ D f1 .x/a1 .t; y/ C f2 .x/a2 .t; y/: (6.77)
Similarly we need to consider the functions of the form: .M /wa as a function of

x and y.
Hence we need to introduce the space C of functions of the form:
f .x; y/ jxj < bI 0<y<`
with range in the scalar field such that for each y, 0 < y < `,
f .:; y/ is in Lp .b; b/:
And
Z `
.jjf .:; y/jjp /2 dy < 1;
0
where jj:jjp denotes the norm in Lp .b; b/. This is a Banach space with norm:
s
Z `
.jjf .:; y/jjp /2 dy:
0
Let L1 D L1 .0; 1/I HE

denote the space of functions such that f .t; :/ is in HE
for each t 0 and
RT
1: 0 jjf .t; :/jjE dt < 1I for each T; 0 < T < 1
and Laplace transformable
R 1 t
2: 0 e jjf .t; :/jjE dt < 1 for > a 0.
We note that the function wa .t; :; y/ defined in (6.4) is in L1 . Define now on
Lp .b; b/:
Z b
LL .f / D U1 f .x/dx;
b
Z b
LM .f / D U1 .x ab/f .x/dx:
b
Then for f .:/ in C , the functions
LL .f .t; :; y//LM .f .t; :; y// 0 < tI 0 < y < `
are in L1 .
Recall now the structure state vector

x .t; y/
Y .t; y/ D ; 0<y<`
xP .t; y/
as an element of the energy space HE for each t 0, and also in L1 . Then
wa .t; x; y/ D f1 b1? Y .t; y/ C f2 b2? Y .t; y/; (6.78)
where b1 , b2 are column vectors:
b1? D 0; U; 1; ab
I b2? D 0; 0; 0; 1
:
And hence
.M /wa D .M /.f1 b1? C f2 b2? /Y .:; y/
;
which yields a linear bounded operator denoted L12 on L1 into :
.M /wa D L12 Y .:; :/: (6.79)
Let
g.t; x; y/; jxj < b; 0 < y < `
denote the function L12 Y . Then .M /wa is the function
Z t
h.t; x; y/ D g.t; x; y/ d=dt g.t ; x U; y/d;
0
where
Z t
d=dt g.t ; x U; y/d
0
Z bCx
U bCx
D P ; x U; y/d;
g.t t>
0 U
Z t
bCx
D P x U.t /; y/d ;
g. ; t>
t bCx
U
U
and can be neglected for large t, compared to the first term. And in as much as our
interest is primarily in the asymptotic response we omit it, and thus in what follows
we set .M / D .M /; so that
1
Jg D .M /L.M /L12Y .:/;
2
where now
.t/ D .t; y/ D b ? Y .t; y/; b D col0; 1; 0; 0
and
1
D
.M /.L.M /L12Y .:// (6.80)
2
from now on. With this qualification let us evaluate Jg. We have: L.M /g is the
function
Z t
h.t; :; y/ D .t; y/ L.t /.; y/g.; :; y/d in Lp .b; b/
0
for each y which we can express as

Z t
D B2 .Y .t; y/; Y .; y//L.t /g.; :; y/d;
0
where B2 .Y1 ; Y2 / is the bilinear functional
b ? Y1 Y2 b:
Next we need to consider .M /h for h in C

We have already used this for h D wa . Recall (6.67) .M /h is the function:
Z t
2
hC P .M; t /h.; :/d;
M 0
where the first term is of importance in the first term in (6.75).

Hence we find it convenient now to express it as
Z t
'.M; t /h.; :y/d
0
allowing for a delta function in '.M; t/. Thus Jg is the function

Z t Z
1
'.M; t / B2 .Y .; y/; Y . ; y//L. /g. ; :; y/d d (6.81)
2 0 0
valid for large t, because we are taking as the identity, and is recognized as a
second-degree polynomial in Y .:; :/ the structure state.
Also
Z Z
1 t
J .M /wa D '.M; t / B2 .Y .; y/; Y . ; y//L. /
2 0 0
Z
'.M; s/wa .s; :; y/d d: (6.82)
0
And the corresponding lift

Z t Z
1
LL .Jg/LL '.M; t /B2 .Y .; y/; Y . ; y// L. /g. ; :; y/d d
2 0 0
can be expressed as
Z tZ
1
D LL .'.M; t /B2 .Y .; y/; Y . ; y//L. /g. ; :; y//d d:
2 0 0
(6.83)
And
LL .J .M /wa .:; :; y//

1
LL .M /L.M / .M /wa.:; :; y/ ;
2
where .M /wa .:; :; y/ is the function (fixing M; 0 < M < 1; in what follows in
this chapter)
Z
g0 . ; :; y/ D '. s/.f1 b1? C f2 b2? /Y .s; y/ds; (6.84)
0
so that (6.83) becomes

Z tZ Z
1
LL .'.t /B2 .Y .; y/; Y . ; y//
0 0 0 2
L. /'.M; s/.f1 b1? C f2 b2 //Y .s; y/dsd d:
More generally:
LL .J k g0 /; k 1; (6.85)
Z tZ
1
LL .'.t /B2 .Y .; y/; Y . ; y//L. /gk1 . ; :; y//d d (6.86)
2 0 0
with g0 .:; :; :/ given by (6.84).

Hence we can express:
gkL .t; y/ D LL .J k .M /wa .:; :; y//
as
Z tZ 1 Z 2k
1
:: LL '.t 1 /b ? Y .1 ; y/b ? Y .2 ; y/L.1 2 /
2k 0 0 0
'.2 3 /b ? Y .3 ; y/b ? Y .4 ; y/L.3 4 /
::::::::::::::::::::::::
'.2k2 2k1 /b ? Y .2k1 ; y/b ? Y .2k ; y/L.2k1 2k /

'.2k 2kC1 /.f1 b1? C f2 b2? /Y .2kC1 / d1 : : : :d2kC1

:::::::::::::::::::::::: (6.87)
Similarly we can calculate the moment
gkM .t; y/ D LM .J k .M /wa .:; :; y//

Z tZ 1 Z 2k
1
as D k :: LM .'.t 1 /B2 .Y .1 ; y/; Y .2 ; y//L.1 2 /:
2 0 0 0
'.2 3 /B2 .Y .3 ; y/; Y .4 ; y//L.3 4 /

::::::::::::::::::::::::
'.2k2 2k1 /B2 .Y .2k1 ; y/; Y .2k ; y//L.2kl 2k /
'.2k 2kC1 /.f1 b1? C f2 b2? /Y .2kC1 //d1 : : : d2kC1 : (6.88)
This completes the calculation of the lift and moment. So we can proceed to the
following.
The Nonlinear Convolution/Evolution Equation
We can now state the nonlinear convolution/evolution equation, generalizing the

linear case in Chap. 5, using the same notation therein.
We state this in the form of a theorem.
Theorem 6.7. The dynamics of the structure under aerodynamic loading can be
expressed as
YP .t/ D AY .t/ C N .t; Y .// t > 0; (6.89)
where Y .t/ is the structure state at time t, N .t; Y .:// defines a physically realizable
nonlinear convolution, and for Y .:/ such that for each T; 0 < T < 1,
j.t; y/j D jb ? ; y.t; y/j
is
m.T / < 1; 0 t T; 0<y<` (6.90)
is defined by the Volterra expansion: valid for 0 < t < T
Z t
N .t; Y .:// D L.t /Y ./d (6.91)
0
1 Z t Z
X 1 Z 2k
C L2k .t 1 ; 1 2 ; :; 2k 2kC1 I
kD1 0 0 0
Y .1 /; :Y .2kC1 /d1 : : : d2kC1 ;
where
L2k .t1 ; t2 ; ::t2kC1 ; Y1 ; y2 ; ::Y2kC1 /; Yi in R4

0
D ; (6.92)
M1 F2k .t1 ; t2 ; t2kC1 I Y1 ; Y2 ; Y2kC1 /
where
F2k .t1 ; t2 ; :t2kC1 I Y1 ; Y2 ; :Y2kC1 /

2 3
LL .'.t1 /B2 .Y1 ; Y2 /L.t2 /:'.t3 /B2 .Y3 ; Y4 /L.t4 /
6 7
6 7
6 7
6 '.t2k1 /B2 .Y2k1 ; Y2k /L.t2k / 7
6 7
1 6 6
'.t2kC1 /.f1 ; b1? C f2 ; b2? /Y2kC1 7
7:
D k6 (6.93)
2 6 LM .'.t1 /B2 .Y1 ; Y2 /L.t2 / '.t3 /B2 .Y3 ; Y4 /L.t4 / 7 7
6 7
6 7
6 7
4 '.t2k1 /B2 .Y2k1 ; Y2k /L.t2k / 5
'.t2kC1 /.f1 ; b1 C f2 ; b2 /Y2kC1
? ?
Proof. In what follows we fix T and so abbreviate m.T / to simply m. We start with
the linear part of (6.90) given in Sect. 5.6:
Z t
YP .t/ D AY .t/ C L.t /Y ./d; (6.94)
0
where

x .t/
Y .t/ D ;
xP .t/

m S
MD ;
S I

0 I
AD :
M1 .A C K/ M1 D
A; K, and D are as defined in Sect. 5.6.

0
L.t/Y D for Y in R4 :
M1 F .t/B Y
The nonlinear part is determined by the lift and moment calculated for the pressure
doublet A.:; :/ determined as a solution of the nonlinear Possio equation (6.73).
Under condition (6.90) we can use the expansion (6.75) for the pressure doublet
and our calculation of the corresponding lift and moment above.
This yields the kth term for k 1 in the Volterra expansion (6.90):
1
X
N .t; Y .:// D gk .t; Y .://; (6.95)
kD1
Z tZ 1 Z 2k
gk .t; Y .:// D ::: L2k .t 1 ; 1 2 ; : : : 2k 2kC1 I (6.96)
0 0 0
Y .1 /; : : : .2kC1 //d1 : : : d2kC1 ; (6.97)
where
L2k .t1 ; t2 ; : : : t2kC1 I Y1 ; Y2 ; : : : Y2kC1 /; Yi in R4

0
D
M1 F2k .t1 ; t2 ; : : : t2kC1 I Y1 ; Y2 : : : Y2kC1 /
and
F2k .t1 ; t2 ; : : : t2kC1 I Y1 ; Y2 ; Y2kC1 /

2 3
LL .'.t1 /b ? Y1 b ? Y2 L.t2 / '.t3 /b ? Y3 b ? Y4 /L.t4 /
6 7
6 7
6 7
6 ? ?
'.t2k1 /.b Y2k1 b Y2k /L.t2k / 7
6 7
1 6 '.t2kC1 /.f1 b1 C f2 b2 /Y2kC1
? ? 7
D k6 7
2 6 LM .'.t1 /b Y1 b Y2 L.t2 / '.t3 /b Y3 b Y4 L.t4 / 7
6 ? ? ? ?
7
6 7
6 7
6 7
4 '.t2k1 /b ? Y2k1 b ? Y2k L.t2k / 5
'.t2kC1 /.f1 ; b1? C f2 ; b2? /Y2kC1
and

0
gk .t; y/ D ; (6.98)
M1 F2k .t; y/
where

F2 kL .t; y/
F2k .t; y/ D
F2 kM .t; y/
Z tZ 1 Z 2k
1
F2kL .t; y/ D k :: LL .'.t 1 /b Y .1 ; y/b Y .2 ; y/L.1 2 /:
2 0 0 0
'.2 3 /b Y .3 ; y/b Y .4 ; y/L.3 4 /

::::::::::::::::::
'.2k2 2k1 /
B2 .Y .2k1 ; y/; Y .2k ; y//L.2k1 2k /
'.2k 2kC1 /.f1 b1 C f2 b2 /Y .2kC1 //d1 : : : : d2kC1 ;
Z tZ 1 Z 2k
1
F2kM .t; y/ D k :: `M .'.t 1 /b Y .1 ; y/b Y .2 ; y/L.1 2 /:
2 0 0 0
'.2 3 /b Y .3 ; y/b Y .4 ; y/L.3 4 /

::::::::::::::::::
'.2k2 2k1 /
b Y .2k1 ; y/b Y .2k ; y/L.2k1 2k /
'.2k 2kC1 /.f1 b1 C f2 b2 /Y .2kC1 //d1 : : : : : : d2kC1 :
To prove convergence of the series in (6.91), we need to get bounds for the terms.
Let us look at the case k D 1. We have
Z t
1
F2L .t; y/ D `L '.t 1 / b Y .1 ; y/v1 .1 ; :; y/d1 ;
2 0
where
Z t
v1 .t; :; y/ D L.t s/b Y .s; y/v0 .s; :; y/ds;
0
Z t
v0 .t; :; y/ D '.t s/.f1 . : /b1 C f2 . : /b2 /Y .s; y/ds
0
and
Z t
jjv1 .t; :; y/jj m l.t s/a0 .s; :; y/ds;
0
1
jF2L .t; y/j jj`L jjm2 .a.:/ l.:/ v1 .:; y//.t/; (6.99)
2
where denotes convolution, and
.t/ D jj'.t/jj t 0;
l.t/ D jjL.t/jj t 0;
Z t
a0 .t; y/ D .t s/jj.f1 b1 C f2 b2 /Y .s; y/jjp ds; t 0:
0
And more generally

1
F2kL .t; y/ k jj`L jjm2k ..:/ l.://k a0 .:; y/.t/
2
with a similar estimate for the moment.
Hence
1
X X 1
1
F2kL .t; y/ jj`L jjm2k .a.:/ l.://k .t/ a0 .t; y/; (6.100)
2k
kD1 kD1
where super k denotes the k-fold convolution.

We want to show that the right side is finite. Note that
X
N
1
SN .t/ D jj`L jjm2k .a.:/ l.://k .t/
2k
kD1
is a nonnegative monotone increasing sequence in N . t

u
Lemma 6.8. The series
1
X 1 2k
m ..:/ l.://k .t/ (6.101)
2k
kD1
converges a.e. in t 0, uniformly in y, 0 < y < `.

Proof. Let
.t/ D ..:/ l.://.t/;

Z 1
O
./ D e t v.t/dt 0:
0
The transform goes to zero as ! 1, thus we see that for sufficiently large
1 2
0< O
m ./ < 1:
2
Now
Z T Z 1 X
N
1 2k
e t SN .t/dt e t SN .t/dt D O
m ..// k
;
0 0 2k
kD1
which converges as N ! 1 and as N ! 1 ! the finite limit
X1
1 2k 1 1
D O //k D m2 ./
m .. O < 1:
kD1
2 k 2 1 2 m2 ./
1
O
We can then apply Fatous lemma to the sequence
X
N
1 2k
e t m .a.:/ l.://k .t/
2k
kD1
to see that it converges to a finite limit a.e. in t, 0 < t < T , because the integrals
converge. t
u
P
Hence 1 kD1 jF2kL .t; y/j converges for every t, rather than a.e, because of the
convolution. With similar results for the moment, we see that we have convergence
of the series (6.16):
1
X
gk .t; Y .:// D N .t; Y .://
kD1
in the energy space.

However, in this generality there is no implication that the function is bounded
in 0; 1
. Indeed one has only to take 12 m2 v.t/ D 1 which yields the limit: et .
Series Solution
Next we develop a constructive solution of the initial value problem for the aeroe-
lastic equation (6.89) under some restrictive conditions that imply, in particular, the
torsion angle is small enough.
We can use the state space theory developed in Chap. 5 which would lead to the
theory of nonlinear semigroups (see [45]) but to minimize abstraction we take a
more direct and constructive route, especially because our main interest is in the
steady-state response.
Let S.t/ t 0 denote the semigroup generated by A. Let Y .0/ denote the initial
structure state. Then the method of variation of parameters yields:
Z t
Y .t/ D S.t/Y .0/ C S.t /N .; Y .://d; t 0; (6.102)
0
Z t
Y .t/ S.t /N .; Y .://d D S.t/Y .0/ t 0: (6.103)
0
The solution then can be expressed as a Volterra series [8], which will then be our
constructive solution.
Here, however, we can take advantage of the fact that we have already con-
structed the solution to the linear equation (6.15). Thus
Z t
YP .t/ D AY .t/ C L.t /Y ./d C u.t/; (6.104)
0
where
Z t
u.t/ D N .t; Y .:// L.t /Y ./d
0
and we denote the right side by:

1 Z t Z
X 1 Z 2k
Nnl .t; Y .:// D L2k .t 1 ; 1 2 ; :; 2k 2kC1 I
kD1 0 0 0
Y .1 /; :Y .2kC1 //d1 :: d2kC1 ; (6.105)
where the subscript nl stands for nonlinear.

The initial value problem for the linear equation
Z t
YP .t/ D AY .t/ C L.t /Y ./d
0
has the solution

Y .t/ D W .t/Y .0/;
where W (.) is given by (5.15) and the solution of (6.102) can be expressed:
Z t
Y .t/ D W .t/Y .0/ C W .t s/Nnl .s; Y .://ds: (6.106)
0
This is again a Volterra equation.

However, to get a constructive solution we need to constrain the initial condition.
We first consider the case where the linear system is (strongly) stable.
Theorem 6.9. The Volterra equation (6.105) has the constructive Volterra series
solution for each T; 0 t < T , given by
1
X
Y .t; y/ D Y0 .t; y/ C Yk .t; y/ (6.107)
kD1
uniformly in y.
where
Y0 .t/ D W .t/Y .0/ t 0;

Z t
Y1 .t/ D W .t s/Nnl .s; Y0 .://ds;
0
Z t
Yk .t/ D W .t s/Nnl .s; Yk1 .://ds k1 (6.108)
0
for
m0 D maxjY .0; y/j; 0<y<` (6.109)
small enough.
Proof. We use the standard technique for solving Volterra equations [8] which
yields the series in (6.106) and the problem is to prove convergence.
For this we need to relate the properties of Yk .:/ to those of Yk1 .:/. Hence we
start with the properties of Y1 .:/ in terms of the function
Y0 .t; y/ W
Y0 .t; :/ D W .t/Y .0; :/;

1
X
Nnl .t; Y0 .:// D gk .t; Y0 .://: (6.110)
kD1
We go back and work with (6.98).

Z t
v0 .t; y/ D '.t s/.f1 b1 C f2 b2 /Y0 .s; y/ds t 0; 0 < y < `;
0
Z t Z s

v1 .t; y/ D '.t s/b Y0 .s; y/ L.s /b Y0 . ; y/v0 . ; y/d ;
0 0
Z t Z s
vk .t; y/ D '.t s/b Y0 .s; y/ L.s /b Y0 . ; y/vk1 . ; y/d ; k 1:
0 0
(6.111)
Note that in this breakdown

1
F2kL .t; y/ D LL .vk .t; y// k 1;
2k
1
F2kM .t; y/ D LM .vk .t; y// k 1:
2k
And of course because we are fixing Y .:/ to be Y0 .:/, we may define

0
gk .t; y/ D 1 :
M F2k .t; y/
From Y0 .t/ D W .t/Y0 using the Greens function form of solution:

Z t
Y0 .t; y/ D W .t; y; s/Y0 .s/ds;
0
we have that
m.Y0 .:/; T / D MaxjY0 .t; y/j WT m0 ; 0 < y < `; 0 < t < T;

where
WT D MaxjjW .t; y; s/jj 0 T; 0 y; s `;

m.Y0 .:/; T / D MaxjY0 .t; y/j; 0 < y < `:
Denote m.Y0 .:/; T / by m.

Then
Z t Z s
jv1 .t; y/j m 2
jj'.t s/jjds jjL.s /jjd ;
0 0
Z
jj '. /.f1 b1 C f2 b2 /Y0 .; y/djj; 0t T
0
m2 .:/ l.:/ c.:; Y0 .:/; y/.t/;
with denoting convolution as before and
.t/ D jj'.t/jj;
l.t/ D jjL.t/jj;
Z t
c.t; Y0 .:/; y/ D .jj'.t s/f1 jj jb1 Y0 .s; y/j
0
C jj'.t s/f2 jj jb2 Y0 .s; y/j/ds:
Then for k 1, and t T ,
jvk .t; y/j m2k .:/ l.:/

k c.:; Y0 .:/; y/.t/:
Hence
XN X N

vk .t; y/ vk .t; y/
kD1 kD1
! !
X
N
k
< m .:/ l.:/
2k
c.:; Y0 .:/; y/ .t/ t T; (6.112)
kD1
where
Z t

c.t; Y0 .:/; :/ ..t s/jjf1 jjj b1Y0 .s; y/j C a.t s/jjf2 jjj b2Y0 .s; y/j/ds
0
Z t
f .t s/jY0 .s; y/jds;
0
f D max.jjf1 jj; jjf2 jj/ (which depends on U): (6.113)

Let
X
N
N D m2k .:/ l.:/
k :
kD1

Z 1 X
N
O
et N .t/dt D m2k O ./k ./k :
0 kD1
Assume now that m0 is small enough so that
O D r < 1:
m2 O .0/l.0/
Then N .t/ converges monotonic increasing to a function we denote by .t/; t

0, whose Laplace transform is
O
m2 O ./l./ r
which is bounded by
O
1 m O ./l./
2 1r
and .t/ is nonnegative and goes to zero as t ! 1, and

Z 1
r
.t/dt < : (6.114)
0 1r
Hence
X1 Z t

vk .t; y/ f . /.t /Y0 . ; y/ d (6.115)
kD1 0
r
m O .0/ 0 < t < T:
1r
Hence it follows that

r
Nnl .t; Y0 .:/; y/ < const. f O .0/m 0<t <T (6.116)
1r
and
p r
jjNnl .t; Y0 .://jj < ` const. f O .0/m: (6.117)
1r
Next Z t
Y1 .t/ D W .t s/Nnl .s; Y0 .://ds 0 < t; (6.118)
0
which is recognized as the response of the linear system to the input
Nnl .t; Y0 .:// 0 < t:
To determine Y1 .t/ as a function of y we use the Greens function formula

Z ` Z t
Y1 .t; y/ D W .t ; y; s/Nnl .; Y0 .:/; s/dds; (6.119)
0 0
where
jjW .t; y; s/jj < w < 1 for 0 < t < T; and 0 < y; s < `:
Hence
Z ` Z t
jY1 .t; y/j < w jNnl .; Y0 .:/; s/jdds (6.120)
0 0
Z ` Z T
<w jNnl .; Y0 .:/; s/ jdds 0 < t < T:
0 0
Or
r
m.Y1 .0/; T / w ` T const. f O .0/ m.Y0 .:/; T /:
1r
Then it follows that
r k
m.Yk .0/; T / w ` T const. f O .0/ m .Y0 .:/; T /;
1r
where the constant does not depend on Y0 .:/

Next we constrain r so that
r
w ` T const.f O .0/ D 2 < 1; (6.121)
1r
so that
jY1 .t; y/j < 2 m for t < 1:
And iteratively:
jYk .t; y/j < 2k m for every k for t < T:

6.5 Stability 315
And hence
1
X
Yk .t; y/ converges for t < T
kD1
as required. t
u
6.5 Stability
Our main interest is the stability of the solution (6.106) and its dependence on the
far field speed. Assuming zero angle of attack, as we do, the far field flow velocity
is q1 D Ei U1 and stability thus depends on U1 , which in this section we simply
denote U , the speed parameter.
This presents an excellent illustration of Hopf bifurcation theory (see [4, 7, 37]),
although our presentation is independent, not invoking that theory. In fact a direct
application is not possible, in the sense that we have a single mathematical theorem
we can quote from which the result follows.
We start with the steady-state solution of (6.97) given in Chap. 4: Y .t/ D 0 valid
for all speeds, noting that N .t; Y .:// D 0 for Y .:/ D 0.
Linearizing the equation about the zero structure state we get the linear equation:
Z t
YP .t/ D AY .t/ C L.t /Y ./d: (6.122)
0
The stability theory for this equation has been treated in detail in Chap. 5 and is
determined in terms of the aeroelastic modes.
Stability of the Linearized System: 0 < U < UF
The linear system (6.1) is modally stable for all speeds U such that 0 < U < UF
where UF is the flutter velocity. Each modeexcept for the zero frequencydecays
to zero in time.
There are stronger notions of stability. A linear system is strongly stable if
every initial state decays to zero; the elastic energy goes to zero.
Lemma 6.10. For U < UF the linear system (6.120) is strongly stable.
Proof. Actually we prove that the response can be characterized in terms of
semigroup theory.
Semigroup on Riesz Space:
Recall the time domain solution of the linearized equation (6.1) for M > 0 given by
X
W .t/Y D ek t Pk Y C Q.t/Y; (6.123)
where Q.t/Y is superstable and can be omitted from consideration. Because it

depends on the speed U , let R.U / denote the subspace generated by the Riesz
vectors fYk g; we call it the Riesz subspace, which is then also the span of the vectors
fZk g. Then W .t/ maps R.U / into itself. Call the restriction of W .t/ to R by WR .t/.
t
u
SubLemma 6.10. WR .t/ t 0, defines a C0 semigroup over R. It is a contraction
with a compact resolvent. For U < UF it is strongly stable. Furthermore, it has the
representation
X X
WR .t/Y D ek t Pk Y I D WR .t/ Y D ek t Pk Y:
The generator denoted AR has the representation

X
AR Y D k Pk Y
and the resolvent R.; AR / has the representation

X Pk Y
R.; AR /Y D for k :
k
k
Proof of Sublemma. We only show the semigroup property of WR .t/. And this is
immediate.
X X
WR .t/WR .t/Y D ek t Pk ej s Pj Y
X
D ek t Pk ek s Pk Y
X
D ek .sCt / Pk Y
D WR .s C t/Y:
And the representations are immediate. Furthermore,

Xh X i
WR .t/Y; WR .t/Y
D ek t Pk Y; ej t Qj Y
X
D e2k t Pk Y; Qk Y
X
Pk Y; Qk Y
D Y; Y
for every t 0.
6.5 Stability 317
Given 2 > 0, we can find N2 such that the tail:

X
e2k t Pk Y; Qk Y
; k > N2 is < 2:
And if all k < 0, we can find t2 such that

X
e2k t Pk Y; Qk Y
; k < N2 is < 2;
which is enough to prove strong stability for the case U < UF . t

u
Stability of Nonlinear System
Let us see what we can say about stability in the nonlinear case for (6.104).
We say that the nonlinear system is strongly stable if every initial condition
bounded in the norm by some nonzero constant decays to zero.
Theorem 6.11. For U < UF , the nonlinear system is strongly stable for a small
enough initial condition, by which we mean that jY .t; y/j ! 0 for all maxy jY0 .y/j
small enough.
Proof. For U < UF the linear system is strongly stable. In particular
jjW .t/Y0 jj ! 0 as t ! 0:
This is a new condition that we exploit in extending Theorem 6.7. Thus choose
m D maxjY0 .y/j; 0 y `; < 1:
Then
maxjY0 .t; y/j; 0 y `; m < 1 for all t > 0:
Next choose m so that
Z 1 Z 1
m 2
jj'.t/jjdt jjL.t/jjdt < r < 1:
0 0
Note that jY0 .t; y/j m for all t, and in fact ! 0 as t ! 1 even if not
exponentially.
Hence we can replace T by infinity in Theorem 6.7, and say more.
Thus in the same notation as there we have
r
jNnl .t; Y0 .:/; y/j < const. f O .0/m for all t; 0<t <1 (6.124)
1r
and further goes to zero as t ! 1. Next we choose

r
const. f O .0/ D 2 < 1:
1r
And the iteration:
jYk .t; y/j < 2k jY0 .t; y/j:
Hence
1
X
Y .t; y/ D Yk .t; y/;
kD0
where
1
jY .t; y/j jY .t; y/j ! 0 uniformly in y; as t ! 1:
12
t
u
Next let us consider U D UF .
6.6 Limit Cycle Oscillation
At U D UF at least one mode will have zero real part.
d.M; 1 ; U / D 0I Re 1 D 0:
However, there can be more than one despite popular belief to the contrary.
Remember that when we say the kth mode flutters, the frequency Imk .UF / is quite
different from Imk .0/. So Imk .UF / can be the same for different values of k. But
then for how many values? How many modes can flutter at the same flutter speed?
Indeed it may be nonfinite as, for example, for zero speed, if S is zero. Note that
these are questions that no computer program can answer.
Theorem 6.12. The number of modes that flutter at a given speed U > 0 is finite.
Proof. Suppose now that the number of modes that flutter is infinite. Then denoting
them fi !k g; !k > 0, we must have that
!k ! 1; k ! 1I k D i !k ;
in as much as they cannot have a finite accumulation point. Thus we have that
Z UF
@k
k .UF / D 0 D k .0/ C dU D 0:
0 @U
6.6 Limit Cycle Oscillation 319
Because
k .0/ D 0; we have that

Z UF
@n
dU D 0 for every n; and n ! 1; (6.125)
0 @U
where we can use the formula:

P4
@d @wi
@n i D1
@wi @U
D : (6.126)
@U P4 @d @wi
i D1
@wi @
Now as !n ! 1, so does n . And taking limits in (6.124), and using our previous
estimates for M D 0 and M nonzero, we can see that the limit is a constant.
But it is negative at U D 0, as we have shown in Sect. 5.6, and hence we reach
a contradiction showing that the number of fluttering modes for any speed cannot
be infinite. u
t
Remarks: Note the implication of this result: as the mode number increases they
eventually stop fluttering!
Let us first consider the case where exactly one mode has zero real part and all others
have strictly negative real parts. Hence let i ! D i !F ; where ! is the fluttering mode
@
d.M; i !; UF / D 0; D i ! > 0;
@U
Yi ! .y/ the mode shape. Let P1 denote the corresponding projection.

At U D UF , let us denote W . : / by WF . : /. Then
1
X
Y .t/ D Y0 .t/ C Yk .t/; (6.127)
kD1
Y0 .t; y/ D e Yi ! .y/;
i !t
Z t
Y1 .t/ D WF .t s/Nnl .s; Y0 . : //ds;
0
Z t
Yk .t; y/ D WF .t s/Nnl .s; Yk1 .://ds;
0
1
X
Nnl .t; Y . : // D gk .t; y/;
kD1

0
gk .t; y/ D ;
M1 Fk .t; y/

1 LL .vk .t; y//
Fk .t; y/ D k ;
2 LM .vk .t; y//
where
Z t
v0 .t; y/ D '.t s/.f1 b1 C f2 b2 /Y0 .s; y/ds t 0; 0 < y < `;
0
Z t Z s

v1 .t; y/ D '.t s/b Y0 .s; y/ L.s /b Y0 . ; y/v0 . ; y/d :
0 0
Hence v1 .t; y/ asymptotically, in steady state,
O !/'.i
O !/L.i
D e3i !t '.i O !/.f1 . : /b1 C f2 . : /b2 /Yi ! .y/.b Yi ! .y//2 :
And more generally:
vk .t; y/ D ei.2kC1/!t .'.i O !//k : '.i

O !/L.i O !/
.f1 . : /b1 C f2 . : /b2 /Yi ! .y/.b Yi ! .y//2k ; (6.128)
where
k
jj '.i O !/ .b Yi ! .y//2k jj
O !/L.i
k
jj'.i
O !/jj jjL.i O
O !/jj jb Yi ! .y/j2 D .O .0/l.0// k
:
Let
m D max 0 < y < ` jb Yi ! .y/j

and we assume
O
r D m2 O .0/l.0/ is < 1: (6.129)
Hence the series
1
X
v.t; y/ D vk .t; y/
kD1
D e3i !t '.i O !/.I e2i !t .b Yi ! .y//2 '.i

O !/L.i O !//1
O !/L.i
O !/.f1 . : /b1 C f2 . : /b2 /Yi ! .y/:

'.i
Note that this is made up of harmonics of the form .3 C 2n/!:

6.6 Limit Cycle Oscillation 321
Approximation
For m small enough we may approximate the sum just by the first term, just the third
harmonic. Denote the approximation by va .:/:
va .t; y/ D e3i !t '.i O !/'.i

O !/L.i O !/.f1 .:/b1
C f2 .:/b2 /Yi ! .y/.b Yi ! .y//2 (6.130)
and hence within this approximation
Nnl .t; Y0 .:// D e3i !t g1 .y/; (6.131)
where

0
g1 .y/ D
M1 Fl .y/

1 `L .va .0; y//
F1 .y/ D
2 `M .va .0; y//
va .0; y/ D '.i O !/'.i

O !/L.i O !/.f1 . : /b1 C f2 . : /b2 /Yi ! .y/.b Yi ! .y//2 : (6.132)
Note that
r
jg1 .y/j const: O !/.f1 . : /b1 C f2 . : /b2 /Yi ! .y/j
j'.i
1r
r
const: O .0/f jYi ! .y/j; (6.133)
1r
where f is given by (6.112). We use (6.130) in what follows. Then

Z t
Y1 .t/ D WF .t s/Nnl .s; Y0 .://ds
0
and the steady-state response

Z 1
Y1 .t/ D e 3i !t
WF .s/ei 3!s ds g1 .:/
0
D e3i !t WO F .3i !/g1 .:/
D e3i !t Y1 .0/; Y1 .0/ D WO F .3i !/g1 .:/; (6.134)

which is the same form as Y0 .t; :/; and

r
Y1 .0; y/ const. O .0/f Y0 .0; y/: (6.135)
1r
And similarly by iteration we have
Yk .t; y/ D e3i !t Yk .0; y/; (6.136)
where
k X
1
r
Yk .0; y/ const. O .0/f Yk .0; y/: (6.137)
1r
kD1
Hence
1
X 1
X
Y .t; y/ D e3i !t Yk .0; y/ C ei !t Yk .0; y/; (6.138)
kD1 kD1
where the series converges if r is small enough so that

r
const. O .0/f D < 1;
1r
and
X1

Yk .0; y/ < Yi ! .y/: (6.139)
1
kD1
We have thus a first approximation to the response at the flutter given by (6.13)
which is thus the LCO.
It is periodic with the same period as that of the flutter mode frequency predicted
by the linear model, with a predominant third harmonic. It is thus an illustration of
the Hopf bifurcation theory.
Higher-order approximation would involve the .3 C 2n/ harmonics. If we have
more than one mode fluttering, we would then have to consider .3 C 2n/ harmonics
of each mode and intermodulation, a messy calculation at best!
But the main point is again that we have a response which is periodic in the steady
state with harmonics of the fundamental frequency given by the linear model.
The LCO Amplitude
We define the amplitude of the LCO as the (RMS value):

s
Z T
Lim T ! 1 .1=T / jjY .t; y/jj2 dt;
0
6.7 The Air Flow Decomposition Theory 323
AWGLELS ALPHA
60 3
40 2
20 1
0 0
20 1
40 2
60 3
11:01:05 11:01:10 11:01:15 11:01:20 11:01:25 11:01:30
Fig. 6.1 Flutter LCO
which within the approximation (6.12) is given by the Parseval formula for the
Fourier series:
v !2
u 2
u X1
t
LCO RMS amplitude D Yi ! .y/ C Yk .0; y/ (6.140)
kD1
s
2

Yi ! .y/ 1 C : (6.141)
1
In particular we see that for initial amplitudes small enough, the LCO amplitude is
proportional to the initial amplitude. Further investigation is required to determine
whether this continues to be true for larger initial amplitudes.
Figure 6.1 shows the incidence of flutter LCO encountered in a specially
designed Flight Test; see [108] for more details.
Finally we come to the solution to the potential field equation, the air flow.
6.7 The Air Flow Decomposition Theory
As we have noted, our primary concern is the stability of the structure. The air
flow per se is of secondary importance. However, as we have seen, we are able to
characterize the flow as well. We consider only 2D flow with zero angle of attack.
From Sect. 6.2, we have the decomposition of the kth-order potential
k D L;k C 0k
and
1
X
D k =k
kD0
recalling (6.134). We define

1
X
L D L;k =k
kD1
1
X
0 D 0;k =k
kD0
The sense in which the series converge is specified below.

The solution of
.0;k / D gk1
following ([45, p. 692]) is given by
Z tZ 1 Z 1
0;k .t; x; z/ D F .t ; x ; z /gk1 .; ; /ddd;
0 1 1
t > 0I 1 < x; z < 1; (6.142)
where the kernel

p p 2
F .t; x; z/ D 1= 2 1 M 2 1= t Ux=c12

1= 1 M 2 x 2 =c12 C z2 =c22 ;

c12 D a1
2
1 M2 ;
c22 D a1
2
:
And defining
1
X
gD gk =k
kD1
where the convergence is in L2 .R2 /, we have

Z tZ 1 Z 1
0 .t; x; z/ D F .t ; x ; z /g.; ; /ddd;
0 1 1
t > 0I 1 < x; z < 1: (6.143)

6.7 The Air Flow Decomposition Theory 325
For L;k we extend Theorem 5.3 to nonzero M . Thus specializing (5.18) to

the typical section case and zero angle of attack, we need to take the inverse
Laplace/Fourier transform of
p
1=.k C i !/ez k 2 M 2 C2kM 2 i !C! 2 .1M 2 /
AQOk .; i !/; z > 0:
Now the inverse Laplace/Fourier transform of
1=.k C i !/AQOk .; i !/
is Z t
k .t; x/ D Ak .t ; x U/d; 1 < x < 1:
0
The inverse Fourier transform of
p
ez k M C2kM i !C! .1M /
2 2 2 2 2
is
p p
L.t; x; z/ D 1= 2 .1 M 2 / 1= ..t Ux=c12 /2 r 2 / ;
where
r 2 D .1 M 2 /.x 2 =c12 C z2 =c22 /:
Hence we have
Z tZ bCUt
L;k .t; x; z/ D L.t ; x ; z/k .; /dd: (6.144)
0 b
And hence by (6.143)

Z tZ bCUt
L .t; x; z/ D L.t ; x ; z/.; /dd; (6.145)
0 b
where
Z t
.t; x/ D A.t ; x U/d
0
.
p p
L.t; x; z/ D 1=.2 .1 M 2 //1=. ..t Ux c12 /2 r 2 //; (6.146)
t > 0; 1 < x < 1: (6.147)
Thus finally we have for the time domain potential flow decomposition
.t; x; z/ D L .t; x; z/ C 0 .t; x; z/; (6.148)

where L .t; x; z/ provides the lift, has no discontinuities in the velocity, and can be
linearized, and the linearized equation determines the LCO speed and period. We
are not interested in 0 .t; x; z/ because it does not affect the structure stability. It
cannot be linearized as observed in [86], and may have discontinuities in the flow
velocity. But whether there are shocks (see, e.g., [14] for definition of shocks in
the flow) is not settled by this theory. Shocks imply a change in entropy [14] and
our key assumption is that the flow is isentropic. Also, even if there are shocks, it is
not clear [86] that they affect the structural stability, not withstanding the generally
held beliefs by aeroelasticians.
Notes and Comments
This chapter illustrates the need for analytical theory. For example, a precise
definition of the crucial concept of flutter speed is not possible by numerical
computation, which is probably why there is no formal definition in all of the
aeroelastic literature.
The continuum formulation and solution of the aeroelastic equations in isentropic
flow appears here for the first time. The structure dynamics is assumed to be linear
and although the extension to the nonlinear case would be of interest, the extra
complications it would bring in even in the domain specifications would be too
much without proportional gain.
And of course our main interest is in the nonlinear aerodynamics anyhow. We
have also limited the treatment to the case of zero angle of attack because the
extension to the nonzero case would involve too much effort without justifying the
returns. Our main objective here is in showing the Hopf bifurcation and the LCO.
We have provided the means, however complex, for calculating the LCO amplitude.
This is of course also a tedious process in CFD requiring time marching.
The existence of shocks in isentropic flow continues to be controversial. It is
indeed established in 1D flow (the Riemann problem, see, e.g., [4]). However, there
is no mathematical proof (yet) in 2D or higher dimensions [private communication:
A. Chorin, 2010]. And furthermore, ours can be termed a singular case because of
the finite boundary.
Chapter 7
Viscous Air flow Theory
7.1 Introduction
In this chapter we extend the theory allowing for nonzero viscosity. The field
equations then are the NavierStokes equations, a subject in fluid dynamics of
intense research activity, but still replete with many open problems. Our interest
here is again on the effect of viscosity on the wing-structure response rather than
the flow itself. The extant aeroelastic literature on this aspect is all computational
[82, 83, 93].
7.2 The Field Equation/Conservation Laws
We begin as in Chap. 3, with the governing conservation laws for fluid flow in
their differential form, but now including viscosity. We follow the notation there
for pressure, density, and other thermodynamic variables, with q.t; x; y; z/ in
particular denoting the fluid velocity.
1. Conservation of mass
@
C r .q/ D 0: (7.1)
@t
2. Conservation of momentum (NavierStokes equation)

@q 2
C .q: r/q D q rp C C r.r q/; (7.2)
@t 3
where
is the coefficient of viscosity. .1:78106 (for air)).
is the second coefficient of viscosity; more on this later.
q is the 3 1 vector with components.
r rqi i D 1; 2; 3, and qi are the components of q.

328 7 Viscous Air flow Theory
Similarly .q:r/q is the 3 1 vector with components
q:rqi i D 1; 2; 3;
E 3;
q D Ei q1 C jEq2 C kq
Ei ; jE; kE being the unit orthonormal vectors.

And finally the law that is of crucial importance in our theory:
3. Energy-Flux Equation
As noted by Meyer [14], this is no more than the first law of thermodynamics in
a Eulerian version.
The energy per unit volume
p
E.t/ D cv T D
. 1/
and the flux of energy per unit mass
X 3 X 3
DE
D pr q C i k ei k C T; (7.3)
Dt i D1 kD1

2
i k D i k r:q C 2ei k ;
3

1 @qk @qi
ei k D C
2 @xi @xk
with the notation x1 D x; x2 D y; x3 D z; i k is the Kronecker delta and
3 X
X 3
D i k ei k
i D1 kD1
is called the dissipation function and we have the thermodynamic relation [14]
DS
T D C T; (7.4)
dt
where S is the entropy. (assumed constant) is the coefficient of thermal
diffusivity 1; 000 ; D cp =cv 1:4. The flow is incompressible if the
flow divergence
r:q D 0:
These are the field equations. But what distinguishes the aeroelastic problem are
the boundary conditions. We consider again the finite plane structure model as in
Chap. 3. The big difference from the nonviscous flow is:
7.2 The Field Equation/Conservation Laws 329
1. Zero Slip at the boundary:

Dz.t; x; y/
q.t; x; y; 0C/ D kE ; jxj < b; 0 < y < `; (7.5)
Dt
where z.. , . , . / is the structure displacement along the z-axis, as in Chap. 3. In
other words, there is no slip between the air and the wing motion, the latter being
normal to the wing plane in our beam model. This is of course more restrictive
than just flow tangency, as in the nonviscous model.
2. The KuttaJoukowsky conditions remain the same as described in Chap. 3.
However, whether the Kutta condition is needed is examined later. The far field is
Ei U where we omit the subscript infinity. Our main interest again is the structure
stability as a function of U > 0.
The basic relevant references for viscous flow for our needs are: Meyer [14];
LandauLifschitz [12], SchlictingKersten [15], 2003 edition of the 1965 classic;
ChorinMarsden [4], R. Temam [19]; and OleinikSamokhin [13].
The main difficulty here starts with the basic question of existence and unique-
ness whatever the function space and sense (weak) of convergence chosen for the
flow solution.
Of the huge readily available literature on NavierStokes equations, it is fortunate
that we are only concerned with aspects of the theory that affect the boundary-
structure stability.
It should be noted that if we set D 0P0 D 0 in the NavierStokes equation
(7.2), it does formally reduce to the Euler equation (3.2) we start with. But the
further condition of isentropy is required to reach the full-potential equation, and
the fluid-structure boundary conditions imposed are of course quite different.
Incompressible Flow
Furthermore, we make the simplifying assumption that the flow is incompressible

in the sense that the flow divergence can be neglected
r: q D 0:
This in particular allows us to sidestep the question of including the second

coefficient of viscosity. As a result, the NavierStokes equation (7.2) simplifies to:
@q
C .q:r/q q C rp D 0 (7.6)
@t
and the energy-flux equation to:
X 3 X3
DE
D 2 ei k 2 C T; (7.7)
Dt i D1 kD1

1 @qk @qi
ei k D C (7.8)
2 @xi @xk
and we do not need to worry about .

Before we can consider the stability of the structure we need to consider the
following.
The Static/Steady-State Equation
The static NavierStokes equation simplifies to
.q:r/q C q D 0 (7.9)
and the energy-flux equation to
3 X
X 3
2 ei k 2 C T D 0: (7.10)
i D1 kD1
But this condition by (7.4) is equivalent to saying that the entropy is constant in time.
Hence in the time-invariant case this equation is automatically satisfied because
entropy is taken to be time invariant. Also the continuity equation (7.1) using the
incompressibility condition yields
q:r D 0;
so that we may take to be constant. Thus (7.9) holds with a constant. The surprise
here is that the static solution for the nonviscous case, where the structure is at rest
and the flow vector is a constant, does not hold here because of the no-slip boundary
condition that the structure and the air move together so that the air velocity has
to be zero over the structure. Hence the flow cannot be a constant if the far field
velocity is nonzero.
Typical Section/Zero Angle of Attack
To reduce complexity further we now consider a flexible wing so that we may

specialize to the typical section/zero angle of attack, as in the previous chapters.
Thus we have the field equation
.q:r/q q D 0;
where is a constant.
The 2D NS in the xz-plane can now be written out as
2
@q1 @q1 1 @p @ q1 @2 q1
q1 C q3 C Dv C q1 .x; 0/ D 0; (7.11)
@x @z @x @x 2 @z2
2
@q3 @q3 1 @p @ q3 @2 q3
q1 C C q3 C Dv C ; (7.12)
@x @z @z @x 2 @z2
@q1 @x
r:q D C q3 D 0;
@x @z

v D :

Next we come to the all-important flow-structure boundary conditions.
Boundary Conditions
q3 .x; 0/ D 0 D q1 .x; 0/ jxj < b:
Far Field
q3 .x; 1/ D 0 D q3 .1; z/;
q1 .x; 1/ D U D q1 .1; z/:
The entropy S is given by ([14], p. 144)
pe .s=cv / D constant :
So the first task is to determine the static solution. At the present time the only way
is to invoke the following celebrated theory.
Prandtl Boundary Layer Theory
Proposed by Prandtl in 1904 [15] for fluids of small viscosity (vanishingly small
and air qualifies with . 10/6 ); but more important, of large Reynolds number;
see [4]. The concept of the boundary layer is to subdivide the flow region into
two parts: a thin-boundary layer at the wing boundary where the viscosity must
be taken into account where the no-slip boundary condition holds, and the outer
layer which is the bulk of the region where we may neglect viscosity and assume
isentropic flow. For the matching problems here see [3, 4].
A word of caution! It should be noted right away that Oleinik writing in 1999
includes this problem in [13] in the section on Some Open Problems: Is it
possible to give a strict mathematical justification of this procedure and find the
limits of applicability of the Prandtl equations?
In other words, there is no mathematical justification for this procedure, yet.
The Prandtl Limit Equations and the Blasius Solution
Here we follow LandauLifschitz [12]. The Kutta condition is not invoked. In the
usual formulation, as in [12], the boundary is allowed to be infinite. But here we do
not have that luxury; the boundary is finite. There is another additional complication.
In reference to (7.2) we need to make additional simplifications in the inner layer:
@p @p
D0D
@z @x
or the pressure is a constant, along with the density. We set

p
vD :

Next we assume:
@2 q3 @2 q3
D 0 P
0 D 0:
@z2 @x 2
Hence the NavierStokes equation (7.2) becomes
@q1 @q1 @2 q1
q1 C q3 Dv 2 ; (7.13)
@x @z @z
@q3 @q3
q1 C q3 D 0;
@x @z
which using
@q1 @q3
C D0
@x @z
becomes
@q3 @q1
q1 D q3 : (7.14)
@x @x
Here we can state the next theorem.
Theorem 7.1 (Oleinik [13]). The field equation:
@q1 @q1 @2 q1
q1 C q3 Dv 2 ; (7.15)
@x @z @z
@q1 @q3
C D0 (7.16)
@x @z
in the strip D f z > 0I b < x < b g, with the boundary conditions
q1 .x; 0C/ D 0P0 q3 .x; 0C/ D 0;

q1 .x; 1/ D u; b < x < b
has a unique solution such that
@q1 @2 q1
and
@z @z2
are continuous and bounded in .
@q1 @q3
;
@x @z
are continuous and bounded in any finite part of .

We do not have a constructive solution technique. We do have the following
however.
Theorem 7.2 (Blasius [cf. 12, 14, 15]). If the chord is infinite, that is, we have in
our notation b < x < 1, the solution is unique and the limit value as z goes to
infinity r
vU
q3 .x; 1/ D 0:43 b < x < 1: (7.17)
xCb
Proof. See [14]. t
u
We use this as the boundary condition to calculate the steady-state potential flow in
the outer layer.
Thus we have the steady-state problem as in Chap. 4, (4.1), now specialized
to typical section and zero angle of attack. The field equation for the potential
.x; z/ is:
" 2 2 !#
2 . 1/ 2 @ @
0D a11C 2
U . /
2a1 @x @z

1 @ @ 2 @ @ 2
jr j C jr j ; (7.18)
2 @x @x @z @z
@2 @2
D C ;
@x 2 @z2
2 2
2 @ @
jr j D C
@x @z
with the boundary condition:
@ .x; 0/
D w.x/; jxj < b;
@z
where w./ is given by (7.9), and the far field condition that
q1 D i U
and the KuttaJoukowsky conditions.

This problem has already been treated in Chap. 4. The solution for the potential
as calculated there is given by
1
X 1
.x; z/ D xU C k .x; z/; (7.19)
k
kD1
where
0 Z p 1
0 @ 1 1 M2 b A.
/d
l
B z 2 2 2 2 C
B @x C B b z .1 M / C .x
/ C
@ @ A D U B p Z C; (7.20)
l @ 1M 2 1
B.
/ A
@z z
2 2 2 2
d
1 z .1 M / C .x
/
s s
Z b
0:86 bx 1 1
A.x/ D p d
; jxj < b; (7.21)
1 M2 bCx b b

x
Z b
1 A.
/
B.x/ D d
; jxj < 1: (7.22)
b x
For k 2:
Z 1 Z 1
k .x; z/ D L.x
; z /gk1 .
; /d
d;
1 1
x; z in R2 .4:82/;
2 k1 j
Y 1 XX k X @ j m @ m
gk1 D cj kj Cmj
2 i D1 j D1 mD0
@xi @xi
k2 X
X 2 X 2 1
@2 k1 X 1 @ 1m @ m
C c ;
i D1 j D1
lD1
x1 D x x2 D z: (7.23)
Linearization
Next we linearize the full-potential equation (typical-section theory) about this

steady-state solution given by (7.11). Denoting it by 0 .x; z/ and . ; t; x; z/
the solution corresponding to .t/; h.t/ with
.0; t; x; z/ D 0 .x; z/:
Then we have the expansion:

1
X k
. ; t; x; z/ D 0 .x; z/ C k .t; x; z/;
k
kD1
where
@k .t; x; z/
k .t; x; z/ D :
@ k
We are only (!) interested in 1 .t; : ; :/.
Hence differentiating

@2 . / @ 2 2 1 2 2 @
C jr . /j D a1 1 C U 1 jr . /j . / . /
@t 2 @t 2
a1 @t
jr . /j2
r . / r (7.24)
2
with respect to and setting it equal to zero, we obtain:
@2 1 @ 2 1 2
C 2 r 1 r 0 D a1 .1 C 2 .U1 jr 0 j2 // 1
@t 2 @t a1

@
. 1/ 2r 1 r 0 C 1 0
@t
jr 0 j2
r 1 r r 0 r.r 1 r 0 /: (7.25)
2
The flow-tangency boundary condition is now

@ 1 .t; x; 0/ P y/ C @ 1 .x; 0/ .t; y/
P y/ C .x ab/.t;
D h.t; (7.26)
@z @x
and we have used the fact that
k .x; 0/ D 0 k 2:
Plus: the KuttaJoukowsky conditions as in Chap. 3.

To proceed further we have to construct the whole linear theory in Chap. 5 for
(7.25), where now we face the new difficulty that the coefficients depend on the
space variables. So we stop here in this version of the book, leaving it as part of
suggested future work.
Chapter 8
Optimal Control Theory: Flutter Suppression
8.1 Introduction
We have examined the problem of feedback control of structures for stability

enhancement in Chap. 2. In this chapter we return to this problem for aeroelasticity
where the structure is subject to aerodynamic loading. An important point to be
noted here is the dependence of stability properties on the speed parameter. Hence
the control action will need to depend on the speed as well.
Our continuum model is as described in Chap. 3. As we have shown in Chap. 6,
stability is determined completely by the linearized model linearized about the
steady-state solution. Hence it follows that the controller need be based only on
this linear model, an important conclusion of the theory.
Hence we consider only the linear model in Chap. 5. In particular we limit
ourselves to isentropic flow.
As is now well established, to begin with for control design we need a state-
space model. The complexity here is that the state is no longer finite-dimensional.
For finite-dimensional approximations see [17, Chap. 12].
Fortunately we have developed a state-space model in Chap. 5 which we now
invoke. Thus we have:
P
Z.t/ D A. Z.t/;
Y .t/ D PZ.t/:
We need to add to this the control actuator model. Whatever it is, however, it has to
be on the structure. This involves a sensor that inputs to the actuator, a compensator.
Thus for M D 0, we have the control model:
YP .t/ D AY .t/ C C x.t/ C Bu.t/;

P
x.t/ D Ac x.t/ C LYP .t/: (8.1)

338 8 Optimal Control Theory: Flutter Suppression
Or, going back to the convolution/evolution form:

Z t
YP .t/ D AS Y .t/ C T Y .t/ C L.t /YP ./d C Bu.t/; (8.2)
0
where
0
Bu.t/ D (8.3)
Bu.t/
and thus the control may be force or moment, and
u./ L2 Hu ; 0; T 0 < T < 1;
where Hu is a separable Hilbert space, and B is a linear bounded operator into H ,

following the notation in Chap. 2.
For M 0, we have the modification of the convolution term (cf. 5.4.)
Z t
YP .t/ D AY .t/ C L.t /Y ./d:
0
But this can also be subsumed in the state equation in terms of the full
state Z.:/:

P 0
Z.t/ D A. Z.t/ C : (8.4)
Bu.t/
The first question is whether this class of controls, and in particular the canonical
feedback control of Chap. 2: B PZ.t/ can yield controllability in the Z-state
space.
Here we can use the explicit semigroup solution
0 Z t 1
S .t/x C S .t /LYP ./d
Sz .t/Z D @ c 0
c A;
Y .t/
Z t
Y .t/ D W .t/Y C W .t /.CS c ./x L./Y /d;
0
x
ZD
Y
for the solution to the control equation for zero initial conditions:
Z !
t
0
Z.t / D Sz .t / d
0 Bu. /
0Z t Z t Z 1
d
d Sc .t /L W . /Bu. / W . s/L.s/Bu. /ds A
D@ 0 0R d R R 0 :
t t
0 W .t /B u. /d 0 0 W . /L . /d B u.t /d
8.1 Introduction 339
ControllabilityStabilizability
As we have seen in Chap. 2, controllability requires that

[[
0
D Sz .t/
u t >0 Bu
must be dense in the state-space, where

0Z t Z 1
d
B Sc .t /L W ./Bu W . /L./Bud d C
0
Sz .t/ DB @
0 d 0 C:
A
Bu Rt
W .t/Bu 0 W .t /L./Bud
For M 0,
0Rt 1
Sc .t /L./d
0 0
Sz .t/ D@ A;
Bu .t/
.t/ D W .t/Bu.t/:
Let

Xi
i D
Yi
denote the eigenvector of the generator A. corresponding to the eigenvalue i . These
are the aeroelastic modes, as we have seen, with Yi the structure mode shape.
As we show in Chap. 5 they are confined to a bounded real part strip in the
complex plane with a finite number with positive real part for each speed.
Let us recall from Chap. 2 our key result in control theory, Theorem 2.1, and our
canonical candidate for a robust stability enhancing controller.
Here, however, the state space is no longer a Hilbert space. So it does not apply
directly.
Theorem 8.1. Suppose for some i,

Yi;1
B Yi;2 D 0 where Yi D ; and Re i 0:
Yi;2
Then the state space is not controllable and the canonical robust feedback
control
u.t/ D B PZ.t/
cannot stabilize the system.
Proof. This control yields the state equation:

P 0
Z.t/ D A. B P Z.t/
B
and

0
P A. B P i D i Yi BB Yi :

B
But
B Yi D B Yi;2 D 0:
Hence

0
A. B P i D A. i D i i :
B
Or this unstable mode is not stabilized.

Remark: This result is not surprising because a mode cannot be affected if the
corresponding mode shape leaves no trace on the control input. In fact we have:
Z t
Mx.t/
R C Dx.t/
P C Kx.t/ C Ax.t/ D F ./x.t /d:
0
Then the question is: Is it possible to find a controller that has a trace of every mode?
For which, in other words,
B Pi 0
or
B Yi;2 0: (8.5)
Or is there a ui in the control space such that Yi;2 ; Bui 0 for every i ?
This requires that we calculate the mode shapes, and take note of the dependence
on the far field speed. Suppose then we can find B and u such that
B Pi 0 for any i:
Theorem 8.2. Suppose
B Pi 0 for every i:
Then the feedback control u.t/ D B PZ.t/ will enhance the stability of the
structure response, if we can assume that the eigenfunctions are little changed in
closed loop.
Proof. At the simplest level, the argument would be this. In practice, the control
effort is not large enough to change the mode shapes. Hence the eigenfunctions i
will remain the same as without control. Hence the eigenvalues of the generator are
given by

0
A. B P i D i i
B
and hence

0
i jjPi jj D P A.
2
B P i ; Pi
B

D P A. i ; Pi jjB Pi jj2 ;
jjB Yi jj2
Re i D Re i ;
jjYi jj2
which is strictly less than Re i . Hence the damping of every mode is increased. In
particular the structure response is more stable.
Note, however, that (cf. Chap. 2)
jjB Yi jj2
goes to zero as i ! 1:
jjYi jj2
And in any event we may not be able to make every mode stable: we cannot
necessarily make Re i negative unless Re i is zero.
Now let us consider the more general proof.
Let i denote the eigenvectors of the controlled system stability operator so that

0
A.i D i i :
BB Pi
Then
P A. i BB Pi D i Pi :
Suppose
BB Pi D 0:
Then
A. i D i i ;
so that i is an eigenvector of A. , which would contradict our hypothesis. But then
P A. i BB Pi D i Pi
and the assumption about eigenfunctions needed is that

P A. i ; Pi D i :
In any event it would appear that
u.t/ D B PZ.t/ (8.6)
is a good candidate feedback control law for augmenting stability, in particular for
increasing flutter speed.
Flutter Suppression
Note that flutter instability cannot be totally avoided; the system will never be stable.
The terminology flutter suppression is usually used in this sense of increasing the
flutter speed.
In particular this means that there is no optimal control! Also here is an area
where the control based on discretized system models may be misleading because
here the number of modes is finite and this would make the controller like the one
above be able to stabilize all the modes, whereas the continuum model shows it is
not possible.
Then the question is: Is it possible to find a controller that has a trace of every
mode. For which
B Pi 0
or
B Yi;2 0:
Or is there a ui in the control space such that
Yi;2 ; Bui 0
for every i .
This requires that we calculate the mode shapes and take note of the dependence
on the far field speed that the control may have to be dependent on the speed, which
is an open problem at present.
Self-straining Actuators: Incompressible Flow
Because of the remarkable stabilizability properties of the self-straining

actuators, described in Chap. 2, it is natural to consider them for flutter suppression
even if the control effort may be limited. Here following [8] we examine how
effective they can be in incompressible flow where we have an explicit solution of
the Possio equation. For an account of experimental verification see [25].
We only consider the torsion controller because of its super stability capability.
As in Sect. 2.5 we have to change the definition of the state to include the end points.
Thus following [60], we have:
Z t
YP .t/ D AS Y .t/ C T Y .t/ C L.t s/YP .s/ds; (8.7)
0
where Y . : / is in the energy space

p
HE D D As H;
H D Hb Hp ;
Hb D L2 0; ` E 1 ;
Ht D L2 0; ` E 1 ;

Ab 0
AS D ;
0 At

h 0000 0 000
D.Ah / D in Hb h L2 .0; `/I h.0/ D 0 D h .0/ D h .1/ ;
h0 .`/

h EI h0000 . : /
Ah D ;
h0 .`/ EI h00 .`/

D.At / D in Ht 00 L2 .0; `/I .0/ D 0 ;
.`/

GJ 00
At D :
.`/ GJ 0 .`/

p
HE D D AS H with energy inner productAS with domain and range in HE ,
0 1
h1

x1 Bh1 0 .`/C h
D.AS / D Y D B
x1 D @ C D.AS / z1 D 2 ;
z1 1 A 2
1 .`/
and
0 1
h2
B EI h1 00 .`/ C p
B gh C
B C D As ;
@ 2 A
1 0 .`/
GJ g
h1 00 .`/ GJ 1 0 .`/

h2 0 .`/ D EI I 2 .`/ D ;
gh g
0 1
h2
B h1 00 .`/ C
B EI C
B C
B gh C
B C
B 2 C
AS Y D B C:
B 1 0 .`/ C
B GJ C
B g C
B
0000 C
@ 1 EI h 1 .`/ A
M
GJ 1 00 .`/
If gh D 0, then h001 .`/ D 0. If g D 0, then 1 0 .`/ D 0.
ReAs Y; Y D gh jh1 .`/j2 g j 1 .`/j2 so that As generates a contraction
semigroup.
Next T and L. : / are as defined in Sect. 5.4.
To determine the aeroelastic modes we can continue to use the development in
Sect. 5.4. We can continue to use (5.7) except that we must change the matrix P to
0 1
gh
B0 EI
1 0 0 0C
B C
B0 0 0 0 0 0C :
@ g A
0 0 00 1
GJ
Torsion Control
We only consider the pitch or torsion control because of their exceptional stabilizing
properties, as shown in Chap. 2. Thus we set gh D 0.
Our main interest is in the root locus and the stability curve as in Sect. 5.7 and
how the stability is enhanced by the self-straining controls. Let P denote the matrix
0 1
0010 0 0
B0 0 0 0 0 0C
B C
@ g A
0000 1
GJ
and let
d.0; g ; ; U / D det P eA./` Q (8.8)
and following [18], using the notation
faij g for eA./` ;

we have:
0 1
a33 a34 a36
B C
B a43 a44 a46 C
P eA./` Q D B C
@ g g g A
a63 C a53 a64 C a54 a66 C a56
GJ GJ GJ
and hence it follows that we can express
g
d.0; g ; ; U / D d.0; 0; ; U / C d.0; 1; ; U /; (8.9)
GJ
d.0; 1; ; U / D det P3 eA./` Q
0 1
a33 a34 a36
D det @ a43 a44 a46 A ; (8.10)
a53 a54 a56
where
0 1
001000
P3 D @ 0 0 0 1 0 0 A :
000010
If U D 0, the modes are given by (Sect. 2.3):

p
GJI cosh
` C g sinh
` D 0;
s
.2k C 1/ GJ p
k D C i !k ; !k D for g < gc D GJI ;
2` I
s
1 GJ g C gc
D log : (8.11)
2` I gg c
Hence
k .0/ D :
Now from Sect. 2.5

r
p I
d.0; g ; ; 0/ D GJI cosh
` C g sinh
`;
D
GJ
and hence we have, comparing with (8.9),

p
d.0; 0; ; 0/ D GJI cosh
`;
GJ
d.0; 1; ; 0/ D sinh
`:

Hence
@d.0; g ; k ; 0/ p
D `
GJI sinhk
` C g cosh
`k (8.12)
@
and it is important to note that the right side is not zero. If g D 0, this is immediate
because coshk
` is then zero. If g is not zero, then coshk
` cannot be zero and
because
p
GJI
Tanh k
` D ;
g
it follows that (8.12) cannot be zero.
To calculate
@d.0; g ; k ; 0/
@U
we note first that by the perturbation formula in Sect. 2.3, we have that
1
d.0; 0; ; U / D 1 C .cos ` cosh `/cosh ` C w2 w3 .; /
2
as in Sect. 5.4. And:
1
d.0; 1; ; U / D w1 .1 cos ` cosh `/ Cosh `
2
omitting cross-product terms involving wi .
Hence

@d.0; g ; k; 0/ 1 @
D .1 C cos ` cosh `/ cosh `
@U 2 @U

g
C w1 .1 cos ` cosh `/cosh ` :
GJ
Although we can continue calculating, we see that

r
k g k gc I
< D k
GJ GJ GJ
is small enough to be neglected, so that the slope of the stability curve is essentially
the same as for zero-control gain and is of course negative. Unfortunately we dont
have even an approximate formula for the flutter speed; and although the value of
the damping is now large and negative and the slope is the same as before, we cannot
still claim that the flutter speed is going to be less.
Chapter 9
Aeroelastic Gust Response
9.1 Introduction
A safety issue as well as a material fatigue issue is the effect of wind gust on
the aircraft. This is usually examined as part of the rigid-body response to wind
turbulence; see [9, 33]. A sinusoidal-gust model is considered in most aeroelastic
work starting with [6]. See [17, 84].
Here we follow [64] and use, instead, a random-gust model, the random-field
model of wind turbulence due to Kolmogorov; see [26,35]. We use a linear-structure
model because the theory for a nonlinear structure model would involve nonlinear
equations. It is not clear to us that the nonlinearity is any more important here than
in the rigid-body case, where the gust is allowed to be random, but the rigid-body
model is linear. In any event, we only consider the linear Goland beam model and
typical-section aerodynamics, as appropriate for a flexible wing of high-aspect ratio.
9.2 The Turbulence Field
We begin with a description of the Kolmogorov model of wind turbulence. Let

.x; y; z/ denote the wind field in 3D. Let k denote the unit normal to the wing
plane so that the induced normal-wing velocity is
wg .x; y; z/ D .x; y; z/:k: (9.1)
According to the Kolmogorov theory, the wind field ./ is isotropic Gaussian with
the 3D spectral density:
cn2
P3 .f1 ; f2 ; f3 / D 11=6 ; 1 < fi < 1; (9.2)
2 C f12 C f22 C f32

350 9 Aeroelastic Gust Response
where is the scale parameter [35] and cn2 is the turbulence strength whose value
may be considered arbitrary for our purposes here.
The 2D field given by (9.1) has the spectral density:
constant
P2 .f1 ; f2 / D 4=3 ; 1 < fi < 1: (9.3)
2 C f12 C f22
Temporal Wind Gust Model
For an aircraft in motion the wind-gust velocity at any given point on the wing is
a function of time, obtained by invoking the Taylor frozen field hypothesis. Thus
we assume that at the given altitude z, say z0 , and the wing velocitythe far-field
velocity in our aeroelastic modelis along the x-axis equal to U . We also take the
angle of attack to be zero. We assume that the wing displacement normal to the wing
is small compared to z0 and neglect the change in the altitude due to wing motion.
Then at any point along the chord .x; s; z0 /, where s denotes the span variable, the
vertical wind-gust component, as a function of time is given by:
Wg .t; x; s/ D wg .x U t; s; z0 /; jxj < b; 0 < s < `: (9.4)
We have thus a 2D spacetime random field which is Gaussian and stationary in

both space and time. The main thing to note is the dependence on the speed in the
temporal variable. The spacetime covariance function of the process:
EWg .t1 ; x1 ; s1 /Wg .t2 ; x2 ; s2 /;
E denoting expected value, is then given by

Z 1Z 1
1
e2 i.1 .xUt /C2 s/ 2 4=3 dv1 dv2 ; (9.5)
1 1 k C 4 1 C 22
2 2
where
t D t2 t1 I x D x2 x1 I s D s2 s1 :
In particular, we see that if we fix the point x0 on the chord, the spectral density of
the process Wg .t; x0 ; s/ is given by
1
P .x0 ; f1 ; f2 / D 2 4=3 ; 1 < f1 ; f2 < 1: (9.6)
2 2 f1 2
U k C 4 U 2 C f2
Figure 9.1 plot the spectral density in db for various values (FPS units) of speed and
scale factor. There is significant low-frequency content for small speeds and small
, but the curves flatten out resembling white noise for higher values.
9.3 The Gust Forced Aeroelastic Equations 351
25 U = 10
k = 0.7
20
db U = 40
15
U = 200
10
2 4 6 8 10
frequency
frequency
2 4 6 8 10
k = 0.1
5
db
10
U = 200
U = 40 U = 100
15
Fig. 9.1 Temporal guest spectral density
Our objective in this chapter is to consider this as the random disturbance input
to the aeroelastic system and calculate the structure response for the Goland model.
9.3 The Gust Forced Aeroelastic Equations
We assume that the structure model is the Goland model with two degrees of
freedom treated in Chap. 2. The aerodynamics equations are as in the typical-section
zero angle of attack linear case treated in Chap. 5. The field equations are as given
in (6.11):
2
@2 @2 2
2
2 @ 2 @
C 2U a1 .1 M / 2 C a1 2 D 0
@t 2 @t@x @x @z
with the now stochastic boundary condition
@.t; x; 0/
D wa .t; x; s/ C Wg .t; x; s/; (9.7)
@z
where

d
wa .t; x; s/ D .h.t; s/ C .x ab/.t; s// C U.t; s/ :
dt
We need to calculate first the lift and the moment corresponding to the input
downwash Wg . : /. For this we need to calculate first the pressure doublet, using
the Possio equation. Here we consider only M D 0 and M D 1, where we have
explicit solutions.
Thus
Z tZ b
Ag .t; x; s/ D .t ; x; /Wg .; ; s/d
0 b
with the steady-state version:

Z 1 Z b
Ag .t; x; s/ D .; x; /Wg .t ; ; s/d ;
0 b
where the system function . : / is given in Sect. 5.4.

The lift
Z b Z 1 Z b
Lg .t; s/ D
U dx .; x; /Wg .t ; ; s/d ;
b 0 b
where it is convenient to define

Z b
l.t; / D
U .t; x; /dx; (9.8)
b
so that the lift due to gust

Z 1 Z b
Lg .t; s/ D l.; /Wg .t ; ; s/d (9.9)
0 b
and the moment

Z b Z 1 Z b
Mg .t; s/ D
U .x ab/dx .; x; /Wg .t ; ; s/d ;
b 0 b
and defining
Z b
m.t; / D
U.x ab/.t; x; /dx; (9.10)
b
we have
Z 1 Z b
Mg .t; s/ D m.; /Wg .t ; ; s/d : (9.11)
0 b
These are then the random lift and moment, stationary Gaussian random processes
in time variable t and space variable s. The corresponding covariance functions are
given by
RL .t; s/ D E.Lg .t1 ; s1 /Lg .t2 ; s2 //; t D t2 t1 I s D s2 s1
Z 1 Z 1 Z 1 Z b
D d1 d 1 l.1 ; 1 /EWg .t1 1 ; 1 ; s1 /
0 0 0 b
Wg .t2 2 ; 2 ; s2 / : l.2 ; 2 /d2 d 2 ;
where
EWg .t1 1 ; 1 ; s1 /Wg .t2 2 ; 2 ; s2 /

Z 1 Z 1
D e2 i.1 . 2 1 U.t .2 1 ///C2s/
1 1
1
2 4=3 dv1 dv2
k C 4 1 C 22
2 2
and hence:
Z 1 Z 1 Z 1 Z b
RL .t; s/ D d1 d 1 l.1 ; 1 /
0 0 0 b
Z 1 Z 1
e2 i.1. 2 1 U.t .2 1 ///C2s/
4=3 dv1 dv2 l.2 ; 2 /d2 d 2
1 1 k 2 C 4 2 12 C 22
Z 1Z 1 Z Z
1 b
2 i 1 Ut C2 i 2 s
D e e2 i 1 2 2 i 1 U2
1 1 0 b
2
1

l.2 ; 2 /d2 d 2 2 4=3 dv1 dv2
k C 4 1 C 22
2 2
Z Z Z Z
1 1 1 b
2 if1 t C2 if2 s
D e e2 i.f1 =U / 2 2 if1 2
1 1 0 b
2
1

l.2 ; 2 /d2 ; d 2 4=3 df 1 df 2
f1
2
2 2 2
U k C 4 U
C f2
Z Z 2
1 1
D e 2 if1 t C2 if2 s HL f1 ; f1
U
1 1
1
4=3 df 1 df 2
2
f 2
U k 2 C 4 2 U
1
C f2
Z 1 Z b
HL .f1 ; f2 / D e2 if1 t 2 if2 l.t; /dtd (9.12)
0 b
and the spectral density

f1 2 1
D PL .f1 ; f2 / D HL f1 ; 4=3 : (9.13)
U f
2
2
U k 2 C 4 2 U
1
C f2
Similarly
Z 1 Z 1
RM .t; s/ D e2 i 1 Ut C2 i 2 s jHM .U 1 ; 1 /j2
1 1
1
2 4=3
k2 C 4 2 1 C 22
Z Z 2
1 1
D e 2 if1 t C2 if2 s HM f1 ; f1
U
1 1
1
4=3 dfl df2 ;
2
2 2 f1 2
U k C 4 U
C f2
where
Z b Z 1
HM .f1 ; f2 / D e2 if1 t 2 if2 m.t; /dtd (9.14)
b 0
and the spectral density

f1 2 1
PM .f1 ; f2 / D HM f1 ; 4=3 :
U f1
2
2 2 2
U k C 4 U C f2
To evaluate HL ; HM we need to specify M . We begin with M D 0.
The Case M D 0
Lemma 9.1. See [6, 64].

h 1 1
1
HL 1 ; D
U 2T 2 i b J0 2b
U U U
1 1 i
iJ 1 2b C 2iJ 1 2b : (9.15)
U U
This is called the Kussner Function in [6]. T ./ is the Theodorsen function; see
Sect. 5.4.
Proof. Our derivation is a variant on both [6] and [64] and follows our version of the
solution to the Possio equation for M D 0 given in Sect. 5.4. It is mostly a tedious
exercise in Fourier transforms and evaluating various integrals, as in [64].
Z 1 Z b
1
HL 1 ; D e2 i 1 t dt l.t; /e2 i. 1 =U / d
U 0 b
Z b Z b
D
U O
.2i 1 ; x; /e2 i .= 1 U / d dx
b b
and from Sect. 5.4:

Z b
O
.2i v1 ; x; /e2 i .1 =U / d
b
is the solution to the Possio equation for wO a .:; / D e2 i.1 =U / .

Hence using (5.46) let

L.k; .I C kL.0/u//
q./ D .I C kL.0// u C h ;
bekb .K0 .kb/ C K1 .kb//
where
u D 2 WO a ./;
1 b
k D 2 i
;
U
s
Z s
2 b x b b C e2 i.1 =U /
u.x/ D d :
b C x b b x
Hence we have
Z b
1
HL 1 ; D .1 C k.b x//u.x/dx
U b
Z b
L.k; .I C kL.0/u//
C .1 C k.b x//h.x/dx ;
b bekb .K0 .kb/ C K1 .kb//
where
s r
Z 1
1 bx k 2b C 1
h.x/ D e d:
bCx 0 xb
Hence
Z b
.1 C k.b x//h.x/dx
b
s r
Z b Z 1
1 b x 1 k.x b/ 2b C
D dx ek d
b bCx xb 0
Z 1 Z r r
.1 kb/dx 1 x 1 kb 2 C
Db e d
1 .x 1 / 1Cx 0
Z 1 r Z 1 r
kb 1 x kb 2C
dx e d
1 1 C x 0
Z 1 r ! Z 1 r
2C kb kb 2C
Db .1 kb/ 1 e d kb e d
0 0
"Z Z r
1 1
kb 2 C kb
Db .1 kb/e d e d
0 0
Z r # Z 1 r
1
2 C kb kb 2C
Ckb e d kb e d
0 0
Z 1
r Z 1
r
2 C kb d 2 C kb
D b e d k e d kb 2
0 dk 0
Z 1 r
kb 2C
e d
0
D bekb .K/ k.bekb .K/ C ekb .K/0 / kb 2 ekb .K/

kb kb kb K1
D be .K/ kbe .K/ C kbe K1 C C K0 kbekb .K/
kb
D bekb .K/ C ekb K1 kb.K/ekb ;
where
Z r
1
2b C
e k
d D bekb K:
0
Next:
s s
Z b Z b Z b
2 bx b C e2 i U
u.x/dx D dx d
b b bCx b b x
Z s
b
b C 2 i.=U /
D2 e d
b b
D 2bJ
Z b
b b
J D J0 2 iJ 1 2 .b x/u.x/dx
U U b
Z 1
r Z 1
r
2 1x 1sCsx 1 C s 2 i .b=U /s
Db dx e ds
1 1Cx sx 1 1s
Z 1 p
D 2b.1 C .1 s// 1 C s=1 se2 i.b=U /s ds
1

J1 2 b
U
U
D 2bJ C :

Hence
Z b
b
.1 C k.b x//u.x/dx D 2bJ C k2bJ C 2 iJ1 2 :
b U
And
Z b Z x
L.k; .I C kL.0/u// D ek.bx/ u.x/ C k u.s/ds dx
b b
Z b
D u.x/dx D 2bJ :
b
Hence
L.k; .I C kL.0/u// 2.J0 .kb/ iJ1 .kb// 2J

D D kb :
bekb .K0 .kb/ C K1 .kb// .K0 .kb/ C K1 .kb//ekb e K
And finally
Z b
L.k; .I C kL.0/u//
.1 C k.b x//h.x/dx
b bekb .K0 .kb/ C K1 .kb//
Z b
kb kb kb 2J
D be .K/ C e K1 kb.K/e C .1 C k.b x//u.x/dx
ekb K b

b
D 2bJ C k2bJ C 2iJ 1 2 C .b C T .kb/ kb/2J
U

b
D 2 T .kb/J C iJ 1 2 ;
U
which yields (9.15) upon multiplication by

U: t
u
Lemma 9.2.

1 1 C 2.1 C kb/2 .1 C kb/ kb
HM 1 ; D
U e K1 .kb/
U kb 2 k
U
..1 C kb/J1 C ikbJ2 /

1 C kb .3 C 2kb/
C 2kb C 5k 2 b 2 C kb T 2J
ke K k
h 1 1 1 i
ab
U 2T J0 2b iJ 1 2b C 2iJ 1 2b ;
U U U
where
1 1
T D T .kb/I J D J0 2b C iJ 1 2b ;
U U
2 i
K D K0 .kb/ C K1 .kb/I kD ;
b
1 1
J1 D J1 2b I J2 D J2 2b ;
U U
which is clearly more complicated than HL and there is no name attached to it yet.
Proof.
Z bZ 1
f1
HM f1 ; D e2 if1 t 2 i .f1 =U / m.t; /dtd
U b 0
Z b Z 1 Z b
2 if1 t 2 i .f1 =U /
D e
U.x ab/.t; x; /dxdtd ;
b 0 b
Z b Z b
D e2 i.f1 =U / O

U.x ab/.2if 1 ; x; /dxd
b b
Z b Z b
O f1
D
U e2 i.f1 =U /
U.x ab/.2if 1 ; x; /dxd abH f1 ; :
b b U
So we need only to calculate the first term. Now

Z b Z b Z b
e2 i.f1 =U / O
x .2if 1 ; x; /dxd D xq.x/dx;
b b b

2J
q./ D .I C kL.0// u C h kb :
e K
Now
Z b Z b Z b
x.1 C k.b x//u.x/dx D .1 C kb/xu.x/dx k x 2 u.x/dx
b b b
s s
Z b Z b Z b
2 2 2 bx b C e2 i.=U /
x u.x/dx D x dx d
b b bCx b b x
s s
Z b Z b
x2 2 bx b C 2 i .=U /
D dx e d
b x bCx b b
Z r Z s
1 1
2 2 1x 1 C 2 i.b=U /
D b 2 xC C dx e d
1 x 1Cx 1 1
Z 1 s
1 1 C 2 i.b=U /
D 2b 2 C 2 e d
1 2 1
bU
D b 2J i J2 2 :
b
Hence
Z b
x.1 C k.b x//u.x/dx
b
U kbU
D .1 C kb/J1 2 kb 2 J C i J2 2 :
b b
Next
Z b
x 2 h.x/dx
b
Z
1 r Z 1 r
2 1 .1 C /2 1x kb 2C
Db xC1C dx e d;
1 x1 1Cx 0
Z 1 r r
2 1 2 kb 2C
D b C 1 C .1 C / 1 e d
0 2 2 C
Z 1 r
2 1 kb 2C
D b C1C e d
0 2
Z 1 r !
2 2 2C
b .1 C / 1 ekb d
0
Z 1 Z 1 r
1 2C
D b 2 .1 C /2 ekb d C b 2 C 2 C 3 C 2 ekb d
0 0 2
2 C 2bk C b 2 k 2 1
D 3
C 2 ebk ..4 C 2kb/K1 .kb/ C .2kb C 5k 2 b 2 /.K0 C K1 //:
bk 2k
Hence
Z b Z b
.1 C kb/xh.x/dx k x 2 h.x/dx
b b

.1 C kb/ 1 kb 2 C 2bk C b 2 k 2
D 1C e K1 .kb/ C
k kb bk 2
1 bk
e ..4 C 2kb/K1 .kb/ C .2kb C 5k 2 b 2 /K/:
2k
Hence
Z U
b
kbU
q.x/dx .1 C kb/J1 2 kb 2 J C i J2 2
b b b

.1 C kb/ .1 C kb/ 1
C T C
k k ekb K

1
.4 C 2kb/T C .2kb C 5k 2 b 2 / 2J;
2k
which, collecting like terms
1 C 2.1 C kb/2 .1 C kb/ kb U

D 2
e K1 ..1 C kb/J1 C i kbJ2 /
kb k

1 C kb .3 C 2kb/
C 2kb C 5k 2 b 2 C kb T 2J:
ke K k
Hence

f1 1 C 2.1 C kb/2 .1 C kb/ kb U
H M f1 ; D
U e K1 ..1 C kb/J1 C ikbJ2 /
U kb 2 k

1 C kb .3 C 2kb/
C 2kb C 5k 2 b 2 C kb T .kb/ 2J
ke K k
h 1 i
ab
U 2T .kb/J C 2iJ 1 2b ;
U
which is clearly more complicated than HL and there is no name attached to

it yet. u
t
The dependence on the speed does appear to be complicated. However, as U ! 1,
we can see that both spectral densities go to zero rapidly. Turbulence is not
significant at transonic speeds, and so M D 0 is a good assumption.
Finally if we define:
Lg .t; s/
Ng .t; s/ D
Mg .t; s/
the corresponding spectral density (matrix) is defined as

PL PLM
Pg .f1 ; f2 / D ;
pML PM
where

conj HM f1 ; fU1
f1
PLM .f1 ; f2 / D HL f1 ; 4=3
U f1
2
2
U k 2 C 4 2 U C f 2

conj HL f1 ; fU1
f1
PML .f1 ; f2 / D HM f1 ; 4=3
U f1
2
2 2 2
U k C 4 U
C f2
and PL ; PM as already defined.
9.4 Structure Response to Turbulence
The structure being the Goland beam, the dynamics are now given by the stochastic
differential equation forced by the stochastic lift and moment:
Z t
R s/ C S .t;
R s/ C EI @4 h.t; s/
mh.t; 1.; /wa .t ; ; s/dd D Lg .t; s/
@s 4 0
(9.16)
2 Z t
I .t; R s/ GJ @ .t; s/
R s/ C S h.t; m.; /wa .t ; ; s/dd
@s 2 0
D Mg .t; s/ 0 < tI 0 < s < `: (9.17)
With CF or FF end conditions.

At this point we may use the state representation in Chap. 5 for an abstract
formulation; but here it suffices to work the Laplace Transforms to find the transfer
function of the system represented by the left-hand side of the equation. Thus we
have
2 Mx./
O C Dx./
O C Kx./
O C Ax./
O FO ./x./
O D 0; (9.18)
where

m C
b 2 S
b

3
a
MD ;
S
b 3 a I C
b 4 a2 C 18

0
b 2 U1
DD ;

b 2 U1 0

0 0
KD ;
0
b 2 U1 2
0 T .k/
1
k

bU1
FO ./ D @ T .k/ l
b 2 U1 A
k
U1 b 2 2 Ca
2
9.4 Structure Response to Turbulence 363
and x.;
O :/ is the Laplace transform of

h.t; :/
x.t; :/ D :
.t; :/
Thus the transfer function is the operator
.2 M C D C K C A FO .//1
defined for in a right half-plane. To determine this we solve:
.2 M C D C K C A FO .//x.;
O :/ D g:
Or, following Sect. 5.4:
O s/ w2 O .; s/ D 1 g1 .s/;
O s/0000 w1 h.;
h.;
EI
O
.; O s/ w4 O .; s/ D 1 g2 .s/;
s/00 w3 h.;
GJ

g
gD l
g2
with CF end conditions, to be specific. Then with y.; s/ defined as in Sect. 5.4, we

have:
y.; s/0 D A./y.; s/ C Bg.s/ 0 < s < `;
where
0 1
0 0
B0 0 C
B C
B C
B0 0 C
B DB1 C:
BEI 0 C
B C
@0 0 A
1
0 GJ
This is a two-point boundary-value problem and continuing as in Sect. 5.4, we have

Z `
x.;
O s/ D G.; s; /g./d
0
excepting the aeroelastic modes, where G.; :; :/ is the Greens function (see [64]):
G.; s; / D C eA./.s / B C eA./s QD./1 P e A./.` / B 0 < <s
C eA./s QD./1 P e A./.` / B s < < `;

where C is the matrix

!
1 0 0 0 0 0
:
0 0 0 0 1 0
With W .t; s; / denoting the inverse Laplace transform of G.; s; /, we have the
steady-state structure response to gust
Z 1 Z `
x.t; s/ D W .; s; /Ng .t ; /dd
0 0
at any point s along the span, where the gust load is

Lg .t; s/
Ng .t; s/ D 0 < s < `:
Mg .t; s/
Lemma 9.3. The spectral density of the response at any points along the span is
given by
Z 1
Pg .s; f / D O
G.2if O
; s; 2f2 /Pg .f; f2 /G.2if ; s; 2f2 / df 2 ;
1
where
Z `
O
G.2if ; s; 2f2 / D e2 if2 G.2if ; s; /d :
0
Proof. Straightforward calculation. t

u
We note that the wing displacement at any point .x; s/ is given by:
z.t; s/ D h.t; s/ C .x ab/.t; s/; jxj < b; 0 < s < `:
Hence the corresponding gust response spectral density is given by:

Pg .f; x; s/ D LP g .s; f /L ;
where
L D Row1 .x ab/:
9.5 Illustrative Example
By way of an illustrative numerical example we consider (see [12]): a model used by

Lin [25] in his road-runner ground vehicle experiments. This is typical of a forward-
swept wing [6], where the divergence speed is less than the flutter speed. This is a
torsion-mode flutter at 1.4 H flutter speed 10.7 fps and divergence Speed 9.59 fps.
The parameters are:
9.5 Illustrative Example 365
AERODYNAMIC LIFT AERODYNAMIC MOMENT

20
5
15
10 1 2 3 4 5 6
5 Freq
5 10
db db
1 3 4 6 15
2 5
5 Freq 20
10 25
Plunge Pitch
s = /4 15 s = /4
40
10
30
5
20 db
db 0.5 1 1.5 2 2.5 3
10 5 Freq
Freq 10
0.5 1 1.5 2 2.5 3 15
Plunge Pitch
64 s = 3/4 s = 3/4
62 25
60 24
db58 db 23
56 22
54 Freq 21 Freq
0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3
Fig. 9.2 Structure response spectral density at divergence speed: U D Ud Lin model
AERODYNAMIC LIFT AERODYNAMIC MOMENT

20
5 Freq
15
10 1 2 3 4 5 6
5
5 10
db
15
1 2 3 4 5 6 db
5 Freq 20
10 25
Plunge Pitch
s = /4 30 s = /4
40
30 20
20 10
db db
10 0.5 1 1.5 2 2.5 3
Freq
10 Freq
0.5 1 1.5 2 2.5 3
Plunge Pitch
64 s = 3/4
s = 3/4
62 35
60
58 30
56
25
54 db
db
0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3
Freq Freq
Fig. 9.3 Structure response spectral density at flutter speed: U D UF Lin model
FPS Units
m D0.009937
I D 0:0004403
GJ D 3:8383
EI D 1:7542
l D 10:19
a D 0I b D 0.364593
S D 0.0003623
I T 2:3H
The normalized frequency is larger than for the Goland model. The plots are given
in Figs. 9.2. and 9.3. As expected, the spectral density peaks at zero corresponding
to the divergence speed and also of course at the flutter mode. Most of the turbulence
energy is in the low frequencies, and of course low speeds.
Chapter 10
Addendum: Axial Air flow TheoryContinuum
Models
10.1 Introduction
In contrast to the previous chapters the present chapter has nothing to do with
aircraft! It is presented more as an addendum devoted to the currently emerging non-
aircraft areas of application of aeroelastic flutter such as for a biomedical application
palatal flutter [102, 105] where the air flow is axial in contrast to normal. And
piezoelectric power harvesting from wind as alternate energy source [106, 109]
where the orientation of the structure with respect to the flow whether normal or
axial is left open as part of system optimization. Both involve the dynamics of a
structure modeled as a plate or beam in an air stream and the main interest is again
in the aeroelastic flutter phenomenon. But this time our objective is not to suppress
it but rather to maintain it, as in the power harvesting application.
The corresponding continuum aeroelastic theory for axial flow turns out to be
much more complicated than that for normal flow though the starting point is of
course the same, and the basic ideas are the same, which is the reason for presenting
it here as a natural continuation of the previous theory enabling us to call attention
in particular to the differences from normal flow even as an addendum and limited
largely to problem formulation. It serves also as a harbinger of a succeeding volume
devoted to a detailed treatment including full-scale experimental results that can be
run on low speed wind tunnels and do not require hangars or expensive flight tests.
10.2 The Aeroelastic Equations
There is no longer of course the notion of the rigid-body aircraft axis. Instead we
go with the structure beam axis which can be oriented as desired with respect to the
air flow.

368 10 Addendum: Axial Air flow TheoryContinuum Models
Here we consider only axial flow. Thus with jE as before denoting the unit vector
along the structure axis, the far field flow is specified by
q1 D jEU1 : (10.1)
Or, more generally, q1 D U1 .jE cos C ksin E / so that is the angle of attack.
Here, however, we consider only D 0. The right-hand screw convention would
make Ei jE the plunge unit vector. Hence the unit z direction is given by k: E
The structure model generally depends on the application; for example, for the
application to palatal flutter [51] it is a thin plate and torsion per se is not considered.
But here we continue with the linear Goland beam model with two degrees
of freedombending and torsionas in Chap. 2 underscoring in particular the
profound difference from normal flow. With no association to an aircaft wing we
change the span variable to ` < y < ` so that the structure dimensions are:
D b; b `; `:
The structure equations continue to be

Z b
R y/ C S .t;
mh.t; R y/ C EI h0000 .t; y/ D p.t; x; y/dx; ` < y < `;
b
(10.2)
Z b
R y/ C S h.t;
I .t; R y/ GJ 00 .t; y/ D .x ab/p.t; x; y/dx;
b
` < y < `; (10.3)
where the pressure jump p.:/ is to be calculated from the aerodynamics.

We consider only CF end conditions
h; h0 ; D 0 at y D `;
00 000 0
h ;h ; D 0 at y D `:
The flow is assumed to be inviscid and isentropic and hence described again as in
Chap. 3 by a potential function .t; x; y; z/.
The axial flow makes the theory considerably more complicated. Fortunately for
the applications indicated above we have low speed flow so that we may take
M D 0, incompressible flow.
This simplifies the field equation for the potential .t; x; y; z/, and its Laplace
transform
Z 1
O
.; x; y; z/ D et .t; x; y; z/dt; Re > a 0
0
10.2 The Aeroelastic Equations 369
to the 3D Laplace equation
@2 .t; :/ @2 .t; :/ @2 .t; :/

2
C 2
C D 0; 1 < x; y; z < 1;
@x @y @z2
O
@2 .; :/ O
@2 .; :/ O
@2 .; :/
C C D 0; 1 < x; y; z < 1; (10.4)
@x 2 @y 2 @z2
omitting of course the structure boundary.

We recall the definitions of the pressure p.:/ and pI and acceleration
potential .:/ and , where by the isentropy condition
p.t; x; y/ D .t; x; y/ x; y in D 0 otherwise;
where the acceleration potential
@ 1
D C r r:
@t 2
And the Kushner doublet function:

1 @ 1 .r r/
A.t; x; y/ D D C ; x; y in ;
U U dt 2 U
so that
p.t; x; y/ D UA.t; x; y/:
Plus the Kutta condition A.t; x; `/ D 0.
And
@.t; x; y; z/

.t; x; y; z/ D :
@z
As before we use
q.:/ for r:
Fluid-Structure Boundary Conditions
Continuing with the same notation as in Chap. 3, we have for the far field:
q1 D jEU1 : (10.5)
The flow tangency condition is again

@.t; x; y; 0/ @1 Dz.t/
D C ;
@z @z Dt
where
z.t; x; y; 0C/ D 1.h.t; y/ C .x a/.t; y//
and
Dz.t/ @z.t/
D C q:rz:
D.t/ @t
We have again (3.20) and (3.21), where now @1 =@z D 0.

Thus

P y/ C .x a/P .t; y/ C q.t; x; y; 0 Ei/.t; y/
.t; x; y; 0/ D .1/ h.t;

@
C q.t; x; y; 0/ jE .h.t; y/ C .x ab/.t; y// ; x; y e :
@y
We begin with the steady state solution.
10.3 Steady-State Solution
We begin with the steady-state solution about which we need to linearize the system
equations. Following Chap. 4, this is given by
h.y/ D 0 D .y/ jyj < `.x; y; z/ D yU 1 ;
where we may note that the steady-state potential is different from that for normal
flow.
10.4 Power Series Expansion
As in Chap. 5, (5.1), and in the same notation, we invoke the power series expansion
for the potential in terms of the scalar multiplicative parameter (to be distinguished
from the Laplace transform variable).
1
X k
.; t; x; y; z/ D yU 1 C k .t; x; y; z/; (10.6)
k
kD1
where

@k .; t; x; y; z/
k .t; x; y; z/ D D 0;
@k
10.4 Power Series Expansion 371
where for each k 1 we have the field equation
@2 k .t; :/ @2 k .t; :/ @2 k .t; :/

C C D 0; 1 < x; y; z < 1:
@x 2 @y 2 @z2
And correspondingly
1
X k
r.; t; x; y; z/ D U1 jE C qk .t; x; y; z/;
k
kD1
where qk .t; x; y; z/ D rk .t; x; y; z/ with the boundary conditions

1
X k @
@.; t; x; y; 0/
D k .t; x; y; 0/
@z k @z
kD1
"
P y/ C .x a/.t;
D .1/ h.t; P y/
1
!
X k @
C k1 .t; x; y; 0/ .t; y/
.k 1/ @x
kD2
1
!
X k @
C U1 C k .t; x; y; 0/
.k 1/ @y
kD2
#
@
.h.t; y/ C .x ab/.t; y// ; x; y2:
@y
Hence

@ P y/ C .x a/P .t; y/
1 .t; x; y; 0/ D .1/ h.t;
@z

@
C U1 .h.t; y/ C .x a/.t; y// ; x; y 2
@y
and hence in particular

1 .t; x; y/ D 0.
For k 2 we have:

@ @

k .t; x; y; 0/ D k1 .t; x; y; 0/.t; y/ C k1 .t; x; y; 0/
@x @y

@
.h.t; y/ C .x a/.t; y// : (10.7)
@y
Note the difference from the corresponding formula for normal flow in
Chap. 6, where the second term is absent. Next for the acceleration potential,
we have: cf. (5.7):
1
X
@.; :/ 1 k @
.; :/ D C r.; :/ r.; :/ D k .t; x; y; z/
@t 2 k @t
kD1
1
! 1
!
1 X k X k
C E
U1 j C rk .t; x; y; z/ U1 j CE rk .t; x; y; z/ ;
2 k k
kD1 kD1
1
U 2 X k
.; t; x; y; z/ D C k .t; x; y; z/;
2 k
kD1
@ .0; t; x; y; z/
1 .t; x; y; z/ D
@
@1 @
D C U 1 .t; x; y; z/:
@t @y
And following (6.18) we have:

@k @k
k .t; x; y; z/ D CU
@t @y
X
k1
C Ck;j qj .t; x; y; z/ qkj .t; x; y; z/; k 1: (10.8)
j D1
And hence for the Kushner doublet:

1
.; t; x; y/ 1 X k
A.; t; x; y/ D D k .t; x; y/
U U k
kD1
1
X k
D Ak .t; x; y/:
k
kD1
Hence
1
Ak .t; x; y/ D k .t; x; y/
U
2
1 @k @k
D 4 CU
U @t @y
3
X
k1
C Ck;j .qj .t; x; y; z/ qkj .t; x; y; z//5 ; x; y 2 :
j D1
(10.9)
Now for each k we have from (10.2)
@2 k .t; :/ @2 k .t; :/ @2 k .t; :/

2
C 2
C D 0; 1 < x; y; z < 1:
@x @y @z2
As before we rewrite this as

2
@2 k .t; :/ @ k .t; :/ @2 k .t; :/
D C x; y in R2 :
@z2 @x 2 @y 2
Again as we did in the case of normal flow, we take Lp Lq Fourier transforms in

the spatial domain, in addition to Laplace transforms in the time domain.
Z 1 Z 1
QOk .; !1 ; !2 / D ei !1 xi !2 y O k .; x; y/dxdy
1 1
and then (10.2) becomes
@2 QOk
D .!12 C !22 /QO k ; (10.10)
@z2
which to satisfy the far field vanishing conditions leads to the unique solution
p 2 2
QOk .; !1 ; !2 ; z/ D ez .!1 C!2 / QOk .; !1 ; !2 ; 0C/ z>0
p 2 2
D ez .!1 C!2 / QOk .; !1 ; !2 ; 0/ z < 0;
which we can express also in the time domain as

p 2 2
Q k .t; !1 ; !2 ; z/ D ez .!1 C!2 / Qk .t; !1 ; !2 ; 0C/ z>0
p 2 2
Q k .t; !1 ; !2 ; 0 1/;
D ez .!1 C!2 / ./ z<0
with the tilde denoting the Fourier transform.

Hence

Q k .t; !1 ; !2 ; 0C/
p D Qk .t; !1 ; !2 ; 0C/;
.!1 2 C !2 2 /

Q k .t; !1 ; !2 ; 0/
p D Qk .t; !1 ; !2 ; 0/: (10.11)
.!1 2 C !2 2 /
The flow tangency boundary condition is

P y/ C .x a/.t;

.; t; x; y; 0/ D .1/ h.t; P y/

@ @
C .; t; x; y; 0/ .t; y/ C .; t; x; y; 0/
@x @y

@
.h.t; y/ C .x a/.t; y// ; x; y 2 :
@y
Hence
1
X k

k .t; x; y; 0/
k
kD1

P y/ C .x a/P .t; y/
D .1/ h.t;
1
!
Xk @k1
C .t; x; y; 0/ .t; y/
k 1 @x
kD1
1
!
X k @k1 @
C .t; x; y; 0/ .h.t; y/ C .x a/.t; y// :
k 1 @y @y
kD1
For k D 1,

P y/ C .x ab/.t;

1 .t; x; y; 0/ D .1/ h.t; P y/

@
CU .h.t; y/ C .x ab/.t; y//
@y
!
1 .t; x; y; 0C/ D
1 .t; x; y; 0/:
Hence by (10.11),

Q 1 .t; !1 ; !2 ; 0/
p D Q 1 .t; !1 ; !2 ; 0C/;
.!1 2 C !2 2 /

Q 1 .t; !1 ; !2 ; 0/
q D Q 1 .t; !1 ; !2 ; 0/
.!12 C !2 2 /
and hence
1 .t; !1 ; !2 ; 0/ D 1 .t; !1 ; !2 ; 0C/;

!
. r
Q 1 .t; !1 ; !2 / D 2
Q 1 .t; !1 ; !2 ; 0/ !1 C !2
2 2
: (10.12)
For k 2:

@k1
k .t; x; y; 0/ D k.1/ .t; x; y; 0/.t; y/

@x

@k1 @
C .t; x; y; 0/ .h.t; y/ C .x ab/.t; y// :
@y @y
(10.13)
For k D 2:

@1
2 .t; x; y; 0C/ D .2/ .t; x; y; 0C/.t; y/

@x

@1 @
C .t; x; y; 0/ .h.t; y/ C .x ab.t; y/// ;
@y @y

@1

2 .t; x; y; 0/ D .2/ .t; x; y; 0/.t; y/
@x

@1 @
C .t; x; y; 0/ .h.t; y/ C .x ab.t; y///
@y @y
D v2 .t; x; y; 0C/
! by (10.11)
2 .t; x; y; 0/ D 2 .t; x; y; 0C/:
More generally then:

k .t; x; y; 0/ D k .t; x; y; 0C/ for k even;
Qk .t; !1 ; !2 ; 0/ D Qk .t; !1 ; !2 ; 0C/ for k odd;

Q k .t; !1 ; !2 ; 0C/ D
Q k .t; !1 ; !2 ; 0/ for k odd;
vQ k .t; !1 ; !2 ; 0C/ D Qvk .t; !1 ; !2 ; 0/ for k even; (10.14)
1
Ak .t; x; y/ D k .t; x; y/
U
2
X
k1
1 4 @k @k
D CU C Ck;j qj .t; x; y; 0C/ qkj .t; x; y; 0C/
U @t @y j D1
3

qj .t; x; y; 0/ qkj .t; x; y; 0/ 5 ; x; y : (10.15)
For normal flow we showed in Chap. 6 that
X
k1
ck;j .rj .t; x; y; 0C/ rkj .t; x; y; 0C/
j D1
rj .t; x; y; 0/ rkj .t; x; y; 0// D 0:
However, as we show below, this wont hold in the present case because obviously
if it did, the whole problem would reduce to linear because the field equation is now
linear. As before we start with the linear theory.
10.5 Linear Theory
We study first the linear case: k D 1.

Define the Laplace transforms
Z 1
AO1 .; x; y/ D et A.t; x; y/dt Re > a 0;
0
Z 1
O1 .; x; y/ D et 1 .t; x; y/dt Re > a 0;
0
Z 1
O 1 .; x; y/ D et 1 .t; x; y/dt Re > a 0
0
and others similarly below.

Then we have:
O1 @ O1
AO1 .; x; y/ D D k O1 C ; D
U @y U
and with the tilde again denoting the Fourier transform, we have:
QO ! ; ! / D 2 p C i !2
.;
A.; QO !1 ; !2 /:
1 2
!1 2 C !2 2
Or
p
QO 1 !1 2 C !2 2 QO

.; !1 ; !2 / D A.; !1 ; !2 /: (10.16)
2 C i !2
This is similar to what we had for the finite plane case in normal flow (cf. (5.18))
with M D 0 and D 0 where now of course the x- and y-axes are switched. But
there we specialized to the typical section case. Here we need to work with
10.5 Linear Theory 377

wO a;1 .; x; y/ D .1/ U hO 0 .; y/ C .x ab/O 0 .; y/

O y/ C .x ab/.;
C h.; O y/ ; x; y in :
As we have noted, what is different from (5.5) is the appearance of derivatives with
respect to y; in (5.5) there are no derivatives with respect to x.
The analogue of the Possio equation is:
Z 1Z 1
1
O a;1 .; x; y/ D
w eix!1 Ciy!2
4 2 1 1
p
1 .!1 2 C !2 2 / QO
A1 .; !1 ; !2 /d!1 d!2 ; for x; y in :
2 . C i !2 /
(10.17)
However we cannot quite follow the Possio procedure as in Chap.

p 5 because
.!1 2 C !2 2 /=. C i !2 / is not a Mikhlin multiplier on Lp R2 ; it is not even
bounded in !1 . Moreover the Fourier transform is not defined. At this point for
normal flow we simplified the problem by specializing to the typical section case.
We need to determine whether there is a parallel here. We take advantage of the
decomposition
wO a;1 .; x; y/ D xf 1 .; y/ C f2 .; y/ x; y in ;
where

f1 .; y/ D .1/ U O 0 .; y/ C U .;
O y/ ;

O 0 O 0
f2 .; y/ D .1/ U h .; y/ C .ab/ .; y/

O O
C h.; y/ C .ab/.; y/ :
Hence for x, y in
Z 1Z 1
1
xf 1 .; y/ C f2 .; y/ D eix!1 Ciy!2
4 2 1 1
p
1 .!1 2 C !2 2 / QO
A1 .; !1 ; !2 /d!1 d!2 ; (10.18)
2 . C i !2 /
where, in our usual tilde notation for the Fourier transform:

Z bZ `
AQO1 .; !1 ; !2 / D O x1 ; y1 /dx 1 dy 1
eix1 !1 iy1 !2 A.;
b `
and we require:
AO1 .; x; y/ ! 0 as y ! ` :
This is then the version of the Possio equation valid for axial flow. We can obtain
the solution as follows.
We determine AQO11 .; :/ from
Z 1Z 1
1
xf 1 .; y/ D 0 .!1 /ei !1 x eCiy!2
4 2 1 1
p
1 .!1 2 C !2 2 / QO
A11 .; !2 /d!1 d!2 :
2 . C i !2 /
Or,
Z 1
1 i j!2 j
x AQO11 .; !2 /d!2 ;
1
xf 1 .; y/ D eCiy!2 jyj < `;
4 21 2 . C i !2 /
Z 1
1 Ciy!2 1 j!2 j QO
f1 .; y/ D e i A 11 .; !2 / d!2 ; jyj < `;
4 2 1 2 . C i !2 /
(10.19)
and similarly AQO12 .; :/ from

Z 1 Z 1
p
1 1 .!1 2 C !2 2 / QO
f2 .; y/ D .!1 /ei !1 x eCiy!2 A12 .; !2 /d!1 d!2 :
4 2 1 1 2 . C i !2 /
Or
Z 1
1 j!2 j
AQO12 .; !2 /d!2 ;
1
f2 .; y/ D eCiy!2 jyj < `: (10.20)
4 2 1 2 . C i !2 /
And
AO1 .; x; y/ D ix AO11 .; y/ C AO12 .; y/ x; y in : (10.21)
We are fortunate that we can use the results from Chap. 5 to solve (10.19) and
(10.20).
Thus we have, noting that the functions f1 .:/; f2 .:/ are in C1 `; `:

i AO11 .; :/ D .I C L.0//
2 3
6 L ; .I C L.0// O

1 .; :/ 7
6
4
O .; :/ C h. ; :/ 7; (10.22)
` .K0 .` / C K1 .` //e` 5
1
10.6 Stability: Aeroelastic Modes 379
where

O 1 .; :/ D 2T f1 .:/
2 3
L ; .I C L.0//
O 2 .; :/
6 7
AO12 .; :/ D .I C L.0// 6
4
O 2 .; :/ C h. ; :/ ` .K0 .` / C K1 .` //e` 5 ;
7
(10.23)
where

O 2 .; :/ D 2T f2 .:/; (10.24)
where T is the linear bounded operator on C1 `; ` into Lp `; ` defined by (as
before except for change of axis)
s Z s
1 ` y ` ` C f ./
T f D gI g.y/ D d; jyj < `:
` C y ` ` y
Z y
L.0/f D gI g.y/ D f .s/ds; jyj < `;
`
Z `
L. ; f / D e .`s/ f .s/ds
`
s Z r
1
1 `y 2` C 1
h. ; y/ D e d; jyj < `:
`Cy 0 y `
10.6 Stability: Aeroelastic Modes
Our primary concern is of course the stability of the structure under aerodynamic
loading. We begin as before with the aeroelastic dynamic equationlinearized about
the steady-state solutionin terms of the Laplace transforms with all structure
initial conditions set to zero.
We have:
0000
m2 h.; O
O y/ C 2 S .; y/ C EI hO .; y/
Z b
D U O x; y/dx;
A.; (10.25)
b
O
I 2 .; O y/ GJ O 00 .; y/
y/ C S 2 h.;
Z b
D 1U O x; y/dx:
.x ab/A.; (10.26)
b
And we have the two-point boundary value conditions:
h; h0 ; D 0 at y D `;
h00 h000 ; 0 D 0 at y D ` (10.27)
that the solution must satisfy.

An important feature of axial flow is that if we set
a D 0I S D 0; (10.28)
which certainly holds in the applications of interest, the aeroelastic equations above
become uncoupled. In fact we have:
Z b
O y/ C EI hO 0000 .; y/ D U
m2 h.; O x; y/dx;
A.; (10.29)
b
Z b
I 2 O .; y/ GJ O 00 .; y/ D U O x; y/dx;
x A.; (10.30)
b
where the main point is that the right side of (10.26) does not contain the pitching
mode and the right side of (10.28) does not contain the bending mode. This does not
happen for normal flow.
From now on we assume (10.26). As in Chap. 5, we first look at the D 0.
Divergence Speed
We have defined divergence speed for normal flow in Chap. 5. We follow the same
definition here for axial flow. Thus we consider the case where the state variables
do not depend on time; all time derivatives are set to zero. And look for values of U
for which these time invariant equations have a nonzero solution as in an eigenvalue
problem.
Equivalently, we set D 0 in (10.25) and (10.26).
Then we have:
f1 .y/ D .1/U 0 .y/;

f2 .y/ D .1/U h0 .y/:
And the Possio equation: for x, y in (cf. (10.21))

Z 1Z 1
1
xf 1 .y/ C f2 .y/ D eix!1 Ciy!2
4 2 1 1
p
1 .!1 2 C !2 2 / Q
A1 .!1 ; !2 /d!1 d!2 ;
2 .i !2 /
where
Z b Z `
AQ1 .!1 ; !2 / D eix1 !1 iy1 !2 A.x1 ; y1 /dx 1 dy 1
b `
A1 .x; y/ ! 0 as y ! `
The solution is obtained by specializing (10.22) and (10.23) to D 0,

yielding
A1 .x; :/ D x2T f1 C 2T f2 :
And the aeroelastic equations are:

Z b
0000
EI h .:/ D U A1 .x; :/dx D 4bU 2 T h0
b
with the CF boundary conditions. Note that
h0 D I3 h0000 ;
where
Z y Z ` Z `
I3 f D gI g.y/ D f ./ddtds:
` s t
Thus we have the eigenvalue problem in Lp `; `; 1 < p < 2:
4bU 2
f D T I3 f; D :
EI
Similarly
4bU 2
00 D T 0:
3GJ
Or we have the eigenvalue problem in Lp `; `; 1 < p < 2:
4bU 2
f D T I1 f; D ;
3GJ
Z `
I1 f D gI g.y/ D f .s/ds
y
and we note that the operators T I3 and T I1 are compact.

We have thus at most a sequence of values of the speed and if any sequence of
speeds is nonempty, the smallest speed defines the divergence speed, if any. Note
that in contrast to the normal flow case we have set the eigenvalue problem in the
Lp space, 1 < p < 2 rather than p D 2.
Flutter Speed/Stability Curve
Let us now go on to consider the general case. The first question concerns the space
of functions .:/ and h.:/ in which to seek solutions of (10.28) and (10.29). This
depends on the space in which we can locate the right-hand side functions. Here we
consider only the bending modes. Note that they are indeed pure bending modes in
contrast to the case of normal wind Thus for (10.28) we have:
Z b
O y/ C EI hO 0000 .; y/ D U
m2 h.; O x; y/dx
A.;
b
D 2bU AO12 .; y/; jyj < `;
where

AO12 .; :/ D .I C L.0//
O 2 .; :/ C h. ; :/ L ; .I C L.0//VO2 .; :/

` .K0 .` / C K1 .` //e` : (10.31)
Note that

O 2 .; :/ D 2T f2 .; :/;

f2 .; y/ D .1/ U hO 0 .; y/ C h.;
O y/
and the range of T is only in Lp `; ` for 1 < p < 2, even for f2 .:/ in C1 `; `.
Hence we can no longer use a Hilbert space formulation as we did for the case of
normal flow. In addition we have a more profound difference from the normal flow
case. Thus, if we go to the formulation as in Chap. 5, but now consider x.:/ as an
element in the space Lp `; `4 W x.:/ D col x1 .:/; : : : x4 .:/.
Then we can write
Dx.:/ D A.; U /x.:/; (10.32)
where
0 1
0 I 0 0
B 0 0 I 0C
A.; U / D B
@ 0 0 0 I A;
C
W1 W2 0 0
Dx.:/ D z.:/I z.s/ D x 0 .s/; jsj < `
defining a closed linear operator on a dense domain in Lp `; `4 , I is the identity
operator on Lp `; ` into itself, and w1 ; w2 are no longer scalar multipliers (as in
the case of normal flow) but are linear bounded operators on Lp `; ` into itself.
1 h h
W1 h D m2 h C 2bU.I C L.0// 2T h C h. ; :/
EI
. ii
.L. ; .I C L.0//2T h// .` .K0 .` / C K1 .` //e` / ;
1 h h
W2 h D 2bU.I C L.0// 2U T h C h. ; :/
EI
. ii
.L. ; .I C L.0//2U T h// ` .K0 .` / C K1 .` //e` :
We go on to formulate the two-point boundary value problem for (10.30).

We can integrate (10.30) as
x.s/ D x.`/ C g.s/; jsj < `;

g D I4 A.; U /x.:/;
where I4 is the integral operator

Z s
I4 x D zI z.s/ D x./d; jsj < `:
`
For x D colx1; x2; x3; x4, let

x3
L24 x D :
x4
We choose
x.`/ D col0; 0; x3 ; x4 :
Then we can write

x3
L24 x.`/ D D.; U / ;
x4
where D.; U / is 2 by 2 and the two-point boundary value problem can be stated:

x3
D.; U / D 0:
x4
Or, we have as before, the aeroelastic modes are the zeros:
d.; U / D det D.; U / D 0:
We note that the region of analyticity of d.; U / just as in the normal flow case in
Chap. 5 is determined by the Bessel K functions appearing in the denominator in
(10.23). In other words the aeroelastic modes are in the same region as before.
We note also one technique for solving (10.31). Define the function
x0 .y/ D x.`/; jyj < `:
Then we can express (10.31) as an equation in Lp `; `4 : x I4 A.; U /x D x0

and for small enough U (as in the applications we are considering) we may use a
Neumann expansion
1
X
xD .I4 A.; U //n x0 :
nD0
We stop here because this is only intended to be an addendum limited to problem

formulation as already noted.
Next we go on to the linear time domain formulation.
10.7 Linear Time Domain Theory: The Convolution/

Evolution Equation
As before, to obtain the time domain equations we take inverse Laplace transforms
in (10.22) and (10.23). Again we only consider the case where a D 0 and S D 0.
This yields
Z b
R y/ C EI h0000 .t; y/ D U
mh.t; A.t; x; y/dx;
b
Z b
R y/ GJ 00 .t; y/ D U
I .t; .x ab/A.t; x; y/dx:
b
With A1 .t; y/ denoting the inverse Laplace transform of i AO11 .; y/ and A2 .t; y/
O y/;
denoting the inverse Laplace transform of A.;
we have
A.t; x; y/ D xA1 .t; y/ C A2 .t; y/:
Next the inverse Laplace transform of vO 1 .; :/ is given by

v1 .t; :/ D T 0 P
4U bh .t; :/ 4b h.t; :/ :
And similarly that of vO 2 .; :/

4b 3 U 0 4b 3 P
v2 .t; :/ D T .t; :/ .t; :/ :
3 3
And we face the same problem of the choice of spaces as we saw for the Laplace
domain. Note that there is no analogue for this in the case of normal flow. There is
no longer the distinction of noncircular and circular terms either.
10.7 Linear Time Domain Theory: The Convolution/Evolution Equation 385
Proceeding further we invert

U.I C L.0//
O i .; :/ C h. ; :/ L ; .I C L.0//
O i .; :/

` .K0 .` / C K1 .` //e `
i D 1; 2
to the time domain. Following Sect. 5.5, this yields

Z
1 y
U i.t; y/ C Pi .t; s/ds ; jyj < `; i D 1; 2;
U `
where
Z t Z y
i .t; y/ D vi .t; y/ q.t ; y/ vP i .; s/ dsd;
0 `
Z Ut
q.t; y/ D U r.t ; y/c.U t / d;
0
s r
1 `y 1 2` C t
r.t; y/ D ;
`Cy `y Ct t
Z 1
1
et c.t/dt D R q :
1
0
0 et 2`Ct
t
dt
Thus we have for the time domain equations

Z y
R y/ C EI h0000 .t; y/ D U 1 .t; y/ C 1
mh.t; P1 .t; s/ds ; jyj < `;
U `
Z
R 00 1 y
I .t; y/ GJ .t; y/ D U 2 .t; y/ C P2 .t; s/ds ; jyj < `:
U `
We can again formulate this as a convolution/evolution equation similar to (5.1)

for the incompressible normal flow case, except that it will be an equation in a
Banach space.
Finally we go on to formulate the stability problem for the nonlinear system
continuing with the power series expansion (10.4) following the development
in Chap. 6.
10.8 Nonlinear Stability Theory: Hopf Bifurcation/Flutter

LCO
It should be noted that we did not consider the incompressible case M D 0 for
normal flow in Chap. 6 on nonlinear stability. Here we do it for axial flow while still
following pretty much the material therein in principle.
Without apology, we specialize to the case: a D 0 and S D 0. As we have seen,
this uncouples the structure equations for the linear case.
For k > 1, we have the field equation
@2 k .t; :/ @2 k .t; :/ @2 k .t; :/

C C D 0; 1 < x; y; z < 10000
@x 2 @y 2 @z2
with the boundary condition:

@k .x; y; 0/ @ @ @
D .1/ k1 .t; x; y; 0/ h.t; y/ C .x ab/ .t; y/
@z @y @y @y

@
C k1 .t; x; y; 0/ .t; y/ x; y in :
@x
The aeroelastic equation is

Z b
R y/ C EI h0000 .t; y/ D U
mh.t; A.t; x; y/dx; ` < y < `;
b
Z b
I R .t; y/ GJ 00 .t; y/ D U xA.t; x; y/dx ` < y < `;
b
1
X An .t; x; y/
A.t; x; y/ D ;
nD1
n
where An is given by (10.15) and again is the solution of a Possio equation for each
n, and the expansion leads to a Volterra series.
We end the addendum here, deferring more details to Volume. 2.
References
List of Books
1. Mikhlin, S.G.: Mathematical Physics, an Advanced Course. North-Holland, Amsterdam

(1970)
2. Thompson, P.A.: Compressible Fluid Dynamics. McGraw-Hill, New York (1972)
3. Van Dyke, M.: Perturbation Methods in Fluid Mechanics. The Parabolic Press, California
(1975)
4. Chorin, A.I., Marsden, J.E.: A Mathematical Introduction to Fluid Mechanics. Springer,
New York (1993)
5. Hodges, D.H., Pierce, G.A.: Introduction to Structural Dynamics and Aeroelasticity.
Cambridge University Press, Cambridge (2007)
6. Bisplinghoff, R.L., Ashley, H., Halfman, R.L.: Aeroelasticity. Addison-Wesley, Cambridge
(1955)
7. Marsden, J.E., M. McCracken: The Hopf Bifurcation and its Applications. Springer,
New York (1976)
8. Miller, R.K.: NonLinear Volterra Integral Equations. Benjamin, W.A. (1971)
9. Etkin, B., Reid, L.D.: Dynamics of Flight. John Wiley, New york (1996)
10. Hille, E., Phillips, R.S.: Functional Analysis and Semigroups. American Mathematical
Society, Providence (1957)
11. Tricomi, F.G.: Integral Equations. Dover Publications, New York (1955)
12. Landau, L.D., Lifshitz, E.M.: Fluid Mechanics. Elsevier, Amsterdam (2004)
13. Oleinik, O.A., Samokhin, V.N.: Mathematical Models in Boundary Layer Theory.
Chapman & Hall, London (1999)
14. Meyer, R.E.: Introduction to Mathematical Fluid Dynamics. Wiley Interscience, New York
(1971)
15. Schlichting, H., Gersten, K.: Boundary Layer Theory. Springer, Berlin (2003)
16. Balakrishnan, A.V.: Applied Functional Analysis. Springer, New York (1980)
17. Dowell, E. et al: A Modern Course in Aeroelasticity. Kluwer, New York (2004)
18. Kuethe, A.M., C-Y. Chow: Foundations of Aerodynamics. John Wiley, New York (1976)
19. Temam, R.: Navier Stokes Equations. RMS Chelsey, Providence (1984)
20. Mikhlin, S.G.: Multidimensional singular integrals and integral equations. Pergammon,
New York (1965)
21. Cole, J.D., Cook, L.P.: Transonic Aerodynamics. North Holland, Amsterdam (1986)
22. Butzer, P.L., Nessel, R.J.: Fourier Analysis and Approximation, vol. 1. Academic Press,
New York (1971)
23. Knopp, K.: Theory of Functions, Part II. Dover, New York (1947)

DOI 10.1007/978-1-4614-3609-6, Springer Science+Business Media, LLC 2012
388 References
24. Dowell, E.H., Tang, D.: Dynamics of Very High Dimensional Systems. World Scientific,
Singapore (2003)
25. Lin, J.: Suppression of Bending Torsion Wing Flutter Using Self Straining Controllers.
Dissertation, Engineering UCLA (2003)
26. Balakrishnan, A.V.: Introduction to Random Processes in Engineering. Wiley, New York
(2005)
27. Liepmann, H.W., Roshko, A.: Elements of Gas Dynamics. Dover, New York (1985)
28. Carracedo, C.M., Alix, M.S.: The Theory of Fractional Powers of Operators. Elsevier,
Amsterdam (2001)
29. Watson, G.N.: A Treatise on Bessel Functions. Cambridge University Press, Cambridge
(1955)
30. Anderson, J.D.: Modern Compressible Flow, 3rd edn. McGrawHill, New York (2003)
31. Gohberg, I., Krupnik, N.: One Dimensional Singular Integral Equations, vol. 1, Introduction.
Birkhauser Verlag, Boston (1992)
32. Timoshenko, S.: Vibration Problems in Engineering, 2nd edn. Van Nostrand, New York
(1937)
33. Schmidt, L.V.: Introduction to Aircraft Flight Dynamics. AIAA Education Series, CA (1998)
34. Timoshenko, S., Young, D.H.: Vibration Problems in Engineering. Van Nostrand Reinhold,
Dordrecht (1955)
35. Tatarski, V.I.: The Effects of the Turbulent Atmosphere on Wave Propgation. US Dept. of
Commerce, Springfield (1971)
36. Meyer, R.E. (ed.): Transonic Shock and Multidimensional Flows. Academic Press, New York
(1982)
37. Kielhofer, H.: Bifurcation Theory An Intrudction with Application to pde. Springer, New
York (2004)
38. Nixon, D. (ed.): Unsteady Transonic Aerodynamics, Progress in Astronautics and Aeronau-
tics. AIAA, Washington (1989)
39. Landhal, M.T.: Unsteady Transonic Flow. Pergammon, New York (1961)
40. Liepmann, H.W., Puckett, A.E.: Introduction to Aerodynamics of a Compressible Fluid.
Wiley, New York (1947)
41. Arendt, W., Batty, C.J.K., Hieber, M., Neubrander, F.: Vector valued Laplace Transforms and
Cauchy Problems. Birkhauser Verlag, Basel (2000)
42. Pontryagin, L.S.: Ordinary Differential Equations. Addison Wesley, London (1962)
43. Mikhlin, S.G.: Multidimensional Singular Integrals and Integral Equations. Pergammon,
New York (1965)
44. Courant, R., Hilbert, D.: Methods of Mathematical Physics, vols. 12. Interscience, New York
(1966)
45. Barbu, V.: Nonlinear Semigroups and Differential Equations in Banach Spaces. Nordhoff,
Leyden (1976)
46. Hochstadt, H.: Integral Equations. Wiley, New York (1973)
47. Fung, Y.C.: An Introduction to the Theory of Aeroelasticty. Dover Publications, New York
(1983)
48. Goldberg, I.C., Krein, M.G.: Introduction to the Theory of Linear Nonselfadjoint Operators,
vol. 18. AMS Translations, Providence (1969)
49. Anderson, J.D.: Modern Compressible Flow. McGrawHill, New York (2003)
50. Paidoussis, M.P.: Fluid Structure Interactions, Slender Structures and Axial Flow,
vol. 12. Elsevier Academic Press, London (2004)
51. Rudin, W.: Real Complex Analysis. McGrawHill, New York (1966)
52. Lovitt, W.V.: Linear Integral Equations. Dover, New York (1950)
References 389
List of Papers
53. Balakrishnan, A.V.: Vibrating Systems with singular mass-inertia matrices. In: Sivasundaram,
S. (ed.) Proceedings ICNPAA, Embry-Riddle Aeronautical University Press (1996)
54. Balakrishnan, A.V.: Dynamics and Control of Articulated Anisotropic Timoshenko beams In:
Tzou, H.S., Bergman, L.A. (ed.) Dynamics and Control of Distributed Systems. Cambridge
University Press, Cambridge (1998)
55. Balakrishnan, A.V.: On the nonnumeric mathematical foundations of linear aeroelasticity. In:
Sivasundaram, S. (ed.) Proceedings ICNPAA, August 2002
56. Balakrishnan, A.V.: Possio Integral Equation of Aeroelasticity. J. Aero. Eng. 16(4), 138154
October (2003)
57. Balakrishnan, A.V.: An Integral Equation in Aeroelasticity. In: Aman, H. et al. (ed.)
Functional Analysis and Evolution Equations. Birkhauser, New York (2007)
58. Balakrishnan, A.V.: On superstability of Semigroups. In: Polis, M.P. et al. (ed.) Proceedings,
IFIP TC7 System Modelling and optimization 1219 (1999)
59. Balakrishnan, A.V.: NonLinear Aeroelasticity Theory: Continuum Models, Contemporary
Mathematics, vol. 420. American Mathematical Society, Providence (2007)
60. Balakrishnan, A.V.: Subsonic Flutter Suppression using self-straining actuators. J. Franklin
Inst. 338, 149170 (2001)
61. Balakrishnan, A.V.: Control of flexible flight structures. In: Analyse Mathematique et
Applications, Gauthier-Villars (1988)
62. Balakrishnan, A.V.: Theoretical Limits of Damping Attainable by Smart beams with rate
feedback. In: Proceedings of the SPIE, vol. 3039 (1997)
63. Balakrishnan, A.V.: Damping operators in continuum models of flexible flight structures.
Explicit models for proportional damping in Beam Torsion. J. Differ. Integr. Equat. 13 (1990)
64. Balakrishnan, A.V.: Modelling Response of flexible high aspect ratio wings to wind turbu-
lence. J. Aero. Eng. 19(2), 121132 (2006)
65. Balakishnan, A.V.: Transonic small disturbance potential equation. AIAA J. 16(4), 139154
(2004)
66. Balakrishnan, A.V.: NonLinear Possio Integral Equation and aeroelastic flutter limit cycle
oscillation, NonLinear Studies, vol. 16. I & S. Publishers (2009)
67. Balakrishnan, A.V.: NonLinear aeroelasticity, the steady state theory. AIAA J. 44, 10061012
(2007)
68. Balakrishnan, A.V.: State space representation of flexible sturcture dynamics in air flow.
J. Franklin Inst. 347(4), 1729 (2010)
69. Balakrishnan, A.V.: Damping Performance of strain actuated beams. Comput. Appl. Math. 18
(1999)
70. Balakrishnan, A.V.: Control of structures with selfstraining actuators: coupled Eu-
ler/Timoshenko model: I. In: Sivasundaram, S. (ed.) NonLinear Problems in Aviation and
Aerospace, Gordon and Breach, Amsterdam (2000)
71. Balakrishnan, A.V.: Analytical solution of the Possio Integral Equation for the sonic case.
J. Aerosp. Eng. 434 (2009)
72. Balakrishnan, A.V.: In: Ceragioloi, F. et al. (eds.) The Possio Integral Equation of Aeroelas-
ticity: A Modern View, System Modelling and optimization. Springer, Berlin (2005)
73. Balakrishnan, A.V., Iliff, K.W.: Continuum aeroelastic model for inviscid subsonic bending
torsion Wing flutter. J. Aerosp. Eng. 20(3), 152164 (2007)
74. Balakrishnan, A.V., Shubov, M.A.: Asymptotic behaviour of the aeroelastic modes for an
aircraft wing model in a subsonic air flow. Proc. R. Soc. 27(4), 329362 (2004)
75. Beran, P.S., Straganac, T.W., Kim, K.: Studies of store induced Limit Cycle Oscillations Using
a model with full system nonlinearities. AIAA 1730, 20031730.
76. Dowell, E.H., Traybar, J., Hodges, H.: An experimental theoretical study of nonlinear bending
and torsion deformations of a cantilever bar. J. Sound Vib. 533544 (1977).
77. Goland, M., Luke, M.L.: The flutter of a uniform wing with tip weights. J. Appl. Math. 15(1)
(1948).
390 References
78. Goland, M.: The flutter of a uniform cantilever wing. J. Appl. Mech. 12, 197208 (1945)
79. AlvarezSalazar, O.S., Iliff, K.W.: Destabilizing effects of rate feedback on strain actuated
beams. J. Sound Vib. 221(2), 289307 (1999)
80. Majda, A.: Disappearing solutions for the dissipative wave equation. Indiana J. Math. 24(12),
132164 (1975)
81. Patil, M.J., Hodges, D.H.: Flight dynamics of highly flexible flying wings. AIAA J. Aircr.
43(6), 17901799 (2006)
82. Su, W., Cesnick, C.E.S.: Dynamic response of highly flexible flying wings. AIAA J. 49,
324339 (2011)
83. Tang, D., Dowell, E.: Experimental and Theoretical study on Aeroelastic response of High
Aspect Ratio wings. AIAA J. 39(8), 14301441 (2001)
84. Tang, D., Dowell, E.: Experimental and theoretical study of gust response for highaspect
ratio wings. AIAA J. 40(3), 419429 (2002)
85. Edwards, J.W., Ashley, H., Breakwell, J.V.: Unsteady Aerodynamic modelling for arbitrary
motions. AIAA J. 15(4), 593595 (1977)
86. Williams, M.H.: Linearisation of unsteady transonic flows containing shocks. AIAA J. 17(4),
394397 (1978)
87. Jameson, A.: Steady state solution of Euler equations in transonic flow. In: Meyer, R.E. (ed.)
Transonic Shock and Multidimensional Flows: Advances in Scientific Computing. Academic,
New York (1981)
88. Albano, S., Rodden, R.P.: A doublelattice method of calculating lift distributions on
oscillating surfaces in subsonic flows. AIAA J. 7, 279285 (1969)
89. Possio, C.: Lazionne aerodynamica sul profilo oscillante in un fluido compressibile a velocite
iposonara. LAerotecnica 18(4) (1938)
90. Garrick, I.E.: Bending Torsion Flutter calculations modified by subsonic compressibility
corrections. NACA TN 1034, May (1946)
91. Balakrishnan, A.V., Triggiani, R.: Lack of generation of Strongly continuous semigroups by
the damped wave operator on HH. Appl. Math. Lett. 6, 3337 (1993)
92. Moretti, G.: Toward a closer cooperation between theoretical and numerical analysis in gas
dynamics. In: Meyer, R.E. (ed.) Transonic, Shock, and Multidimensional Flows: Advances in
Scientific Computing, pp. 241257. Academic, New York (1981)
93. Holmes, P.J.: Bifurcations to divergence and flutter in flow-induced oscillationsfinite
dimensional analysis. J. Sound Vib. 36(4), 17301754 (1977)
94. Grenier, E.: Non derivation des equations de Prandtl, Seminaire Seminaire Equations aux
derivees partielles (Polytechnique) (19971998), Exp. No. 18, p. 12.
95. Serre, D.: Von Neumanns comments about existence and uniqueness for the InitialBoundary
value problems. AMS Bull. 47(1), (2010)
96. Kemp, N.H., Homicz, G.: Approximate unsteady thin airfoil theory for subsonic flow.
AIAA J. 14, 139144 (1998)
97. Balakrishnan, A.V., Edwards, J.W.: Calculation of the transient motion of elastic aorfoils
forced by control surface motion and gust. NASA TM 81351, August (1980)
98. Glimm, J.: Reflections and prospecives. AMS Bull. 47(1), 127136 (2010)
99. Bendiksen, O.O.: Transonic flutter. AIAA paper 20021488. (2002)
100. Wang, D.M., Amomoto, H.Y., Dowell, E.H.: Flutter and limit cycle oscillation of two
dimensional panels in three dimensional axial flow. J. Fluids Struct. (2003)
101. Samko, S.: Singular integral equations: solutions with variable integrability. Unpublished
Manuscript
102. Huang, L.: Flutter of cantilevered plates in axial flow. J. Fluids Struct 9, 127147 (1995).
103. Balakrishan, A.V.: Explicit LQG optimised control laws for flexible structures/collocated rate
sensors. AIAA 941657. Adaptive Structures Forum, La Jolla (1993)
104. Friedman, P.P.: Renaissance of Aeroelasticity and its future. J. Aircr. 36, 105121 (1999)
105. Ellis, P.D.M., Williams, J.E.F., Shneerson, J.W.: Surgical Relief of Snoring due to Palatal
Flutter: a preliminary report. Ann. R. Coll. Sugeons Engl. 286290 (1995).
References 391
106. Dunnmon, J.A., Stanton, S.C., Mann, B.P., Dowell, E.H.: Power Extraction from aeroelastic
limit cycle oscillations. J. Fluids Struct. 27, 11821194 (2011)
107. Tuffaha, A.: Flutter stability analysis of a wedge shaped airfoil with nonzero thickness in
nonviscous air flow. NonLinear Stud. I S publishers 16(3), 235257 (2009)
108. O.S. Alvarez Salazar, Balakrishnan, A.V., Iliff, K.W.: Aeroelastic Flight Test NASAUCLA,
Flight Systems Research Center, UCLA, May (2000)
109. Wang, D.A., Kuo, H.H.: Piezoelectric energy harvesing from flow induced vibration. J.
Micromech. Microeng. 20, 19 (2010)
110. Sears, W.R.: Operational methods in the theory of air foils in non-uniform motion. J. Franklin
Inst. 230, 95111 (1940)
Index
A Aspect ratio, 9, 75
Abel integral equation, 152 The attached flow condition, 5657
Abstract functional analysis, 9, 11 Axial airflow theory, 367386
Abstract version of Possio equation, 123127 Axial flow, 7, 266, 367, 368, 378, 380, 386
Acceleration potential and pressure, 5354 Axial in contrast to normal, 367
Adjoint operator, 217
Adjoint theory, 217220
Aerodynamic lift and moment, 3, 57, 155 B
Aeroelastic Balakrishnan formula, 5, 113115
equations, 4, 6, 65102, 223, 230, 241, 249, Banach space, 7, 184, 185, 188, 299, 385
270296, 326, 351362, 367369, 380, Beams of infinite length, 4346
381, 386 Bending and torsion, 26, 235, 261, 368
gust response, 349366 Bending flutter, 7
modes, 5, 146, 169172, 174176, 178, Bending mode, 19, 26, 37, 40, 42, 43, 235,
190, 191, 199201, 204, 206, 208, 211, 239, 282, 380
226228, 232, 234, 235, 240242, 244, Bending/torsion modes, 19, 26, 235
315, 339, 344, 350, 363, 379384 BeranStraganac model, 40, 41, 96, 249
modes, M D 0, 5, 169, 171172 Bessel K functions, 120, 157, 174, 181, 206,
The aeroelastic equations, linear structure 383
model, 270296 Bifurcation/flutter LCO, 386
Aeroelasticity, 13, 5, 8, 9, 11, 47, 49, 54, 57, Biorthogonal, 34, 220, 222
63, 65, 7280, 101, 103267, 269326, Biorthogonalize, 220
337 Biorthogonal sequence, 222
The aeroelastic problem, 2, 4, 4763, 328 The Blasius solution, 332335
Airflow, 6, 10, 11, 35, 42, 107, 269 Bounded inverse, 15, 137, 139, 144, 152, 170,
Airflow decomposition theory, 323326 171, 192
The airfoil equation, 7680 Bounded vertical strip, 5, 138
Air turbulence, 7 Branch cut, 174, 206, 212, 213, 226, 234
Air/wing interaction dynamics, 56 Bread and butter part, 5, 103
Angle of attack, 4, 5, 10, 55, 5859, 66, 77,
82, 84, 101, 113, 115119, 147148,
151153, 246, 269, 281, 298, 315, 323, C
325, 326, 330331, 334, 350, 351, 368 Canonical example, 94
Applied mathematics, 46 Cantilever-beam, 2
Approximation, 16, 8, 54, 63, 91, 121, 147, CC clampedclamped condition, 11, 12
183, 232234, 247, 248, 259, 266, CF. See Clampedfree (CF)
321323, 337 Chorin, A., 326

DOI 10.1007/978-1-4614-3609-6, Springer Science+Business Media, LLC 2012
394 Index
Circular terms, 384 Damping term, 235

Clampedclamped case, 16, 20 2-D continuum model, 260265
Clampedfree (CF) 2D/3D flow, 4, 7, 106, 323
boundary conditions, 67, 381 Decomposition, 6, 8284, 206, 272276,
end conditions, 11, 19, 22, 42, 60, 75, 96, 278281, 323326, 377
99, 175, 242, 251, 255, 363, 368 Definition of flutter speed, 63, 239, 240
Classical theory, 115, 266 Degradation, 34
Collaborator, 267 Delta function, 41, 162, 166, 289, 298, 301
Colocated, 3 Delta-function derivative, 289
Compact, 11, 13, 18, 27, 28, 32, 137, 142, 144, Dependence on far field speed, 243245
169, 170, 192, 194, 204, 243, 253, 254, Dependence on Mach number, 245
316, 381 3D Euler full potential equation, 53
Compensator, 337 Differential operator, 12, 14, 15, 31, 41, 203
Complete, 4, 7, 10, 13, 18, 34, 47, 57, 58, 91, Digitization, 1
271, 303 Discontinuity in the flow, 56, 62
Computer program, 241, 243, 260267, 318 Dissipation function, 328
Conservation laws, 6, 47, 4950, 63, 327336 Disturbance potential, 83, 84, 107
Conservation of energy, first law of Divergence speed, 5, 75, 7880, 91, 98, 103,
thermodynamics in Eulerian form, 49 175, 208, 230234, 266, 364366,
Conservation of mass, continuity equation, 49 380381
Conservation of momentum, Euler momentum Domain, 5, 9, 1119, 31, 32, 3638, 46, 106,
equation, 49 107, 111, 112, 121122, 127, 146184,
Constant entropy, 3 186, 187, 189, 195199, 202, 224, 225,
Constructive solution, 92, 201, 309, 310, 333 252, 253, 259, 266, 269, 281285, 289,
Continuum aeroelastic theory for axial flow, 292294, 296, 297, 315, 325, 326, 343,
367 373
Continuum model theory, 367386 Doublet function, 73, 105, 111, 147, 278, 296,
Continuum theory, 1, 3, 4, 6, 8, 183 369
Contour integral, 214, 215 DowellHodges model, 4143, 58, 96,
Contour integration, 195 249256
Contraction, 27, 28, 32, 168, 169, 316, 344 Downwash, 104, 107, 116, 123, 271, 352
Control design, 3, 7, 337 1D Possio equation, 112
Controllability, 2730, 338, 339 D self-adjointand nonnegative definite, 27
Controllabilitystabilizability, 339342 3D spectral density, 349
Controllable system, 185 Duhamel principle, 271
Control law, 3, 30, 342 Dynamics of a structure in air flow, 1
Control model, 337
Control operator, 26
Control theory, 3, 7, 2635, 230, 337347 E
Convolution/evolution Eigenfunction expansion, 13
equation, 5, 6, 101, 155169, 183, 188, Eigenfunctions, 5, 13, 21, 256, 340, 341
203, 298, 303309, 384385 Eigenvalue, 3, 5, 13, 15, 18, 19, 23, 28, 29, 32,
form, 338 41, 90, 171, 189, 210, 220, 255, 256,
semigroup, 5 259, 339, 341, 380
Covariance functions, 350, 353 Eigenvalue problem, 75, 78, 102, 175, 241,
C0 semigroup, 17, 28, 29, 32, 168, 184186, 242, 381
189, 196, 204, 224, 228, 316 Eigenvector, 3, 13, 18, 19, 21, 28, 208, 339,
Curl-free, 51 341
Curl of a gradient, 51 Elastic energy, 3, 31, 37, 38, 46, 315
Curl-/vortex-free, 51 Elastic structure, 9, 289
End conditions, 2, 1012, 19, 22, 37, 42, 44,
D 56, 60, 75, 80, 90, 96, 98, 99, 172, 173,
Damping, 3, 11, 26, 27, 29, 30, 3335, 202, 175, 241, 242, 251, 252, 255, 270, 362,
235, 341, 347 363, 368
Index 395
Energy-flux equation, 328330 Flutter

Energy inner product, 14, 168, 217, 343 analysis, 1, 5, 43, 46, 103, 169, 171,
Energy space, 1418, 31, 168, 204, 208, 298, 230248
300, 309, 343 LCO, 269, 323
Engineering problems, 8, 19, 46 speed, 1, 47, 63, 234, 239243, 245, 246,
Enhancement, 2635, 337 248, 260266, 269, 318, 326, 342, 347,
Entire function, 24, 33, 91, 124, 129, 175, 197, 364, 365, 382384
205, 255 speed/stability curve, 240, 382384
Entire function of finite order, 24 suppression, 7, 337347
Entropy, 3, 47, 48, 50, 326, 328, 330, 331 Flying wings, 11, 3540
Equilibrium state, 4 The force/momentum balance equations,
erf function, 196198 56
Euler full potential equation, 4, 4958, 66, Forward-swept wing, 364
256259 Freefree articulated beam model, 3538
EulerGamma, 162 Functional analysis, 9, 11, 46
Existence and uniqueness of solution, 40, Fung assertion, 266
111112, 230 Fung, Y.C., 266
Existence theorem, 2, 138146, 269 Fusilage, 55
Exponential stability, 2830, 34, 35
G
Garrick formula, 245
F Garrick, I.E., 103, 245
FAA. See Federal Aviation Administration Generalized resolvent, 170, 171, 178180, 200,
(FAA) 204, 208211, 228
Far-field condition, 55, 334 equation, 178
Far-field velocity, 47, 66, 146, 249 General nonlinear case, 8095
Fatous lemma, 308 German pioneersKussner, Schwarz, Sohngen,
Federal Aviation Administration (FAA), 245 and Wagner, 8
Feedback control, 3, 2635, 46, 337, 338, 340, Girsanovs theorem, 8
342 Goland beam model, 7, 911, 40, 368
Feedback control laws, 3, 342 GolandLuke model, 35, 46
FEM, 2, 8 Goland model, 2, 3, 35, 41, 46, 58, 60, 65, 97,
FEM/CFD, 1, 8 100, 243, 251, 260267
FF freefree condition, 11 DowellHodges model, 41
Field equation/conservation laws, 327336 Greens function
Finite Hilbert transform, 4, 76, 101 representation, 2126, 210211, 224
Finite plane model, 4, 147 time domain, 224
First few modes, 235 Growth bound, 169, 229
Flexibility, 2, 20, 63 Gust forced aeroelastic equations,
Flight dynamics, 7 351362
Flight test, 1, 323, 367
Flow decomposition, 8284, 272276,
278281, 323326 H
Flow divergence, 328, 329 Half-chord, 55, 155
Flow-structure interaction, 55 Hankel functions, 266
Flow tangency, 4, 56, 61, 62, 67, 82, 85, 104, High aspect ratio, 2, 4, 55, 61, 63, 82, 147
106, 146, 250, 269, 270, 276, 281, 329, HilbertSchmidt, 13
336, 369, 374 Hilbert space, 3, 5, 7, 9, 11, 15, 18, 30, 31, 38,
Flow-tangency condition, 56, 61, 62, 67, 82, 165, 185, 251, 298, 338, 339, 382
85, 250, 369 Hilbert space formulation, 11, 165, 251
Fluid-structure boundary conditions, 2, 47, 67, Hilbert transform, 4, 76, 77, 94, 101, 117, 123,
96, 329, 331, 369370 124, 126, 205
Fluid-structure interaction, 4763, 271 Hodges and Pierce, 11, 63
396 Index
Hopf bifurcation point, 4, 269 Laplace domain Possio (LDP), 112, 127, 138,
Hopf bifurcation theory, 4, 6, 315, 322 146, 205, 285, 289
Laplace inversion formula, 211217
Laplace transform theory, 169
I LCO. See Limit cycle oscillation (LCO)
Idempotent, 180, 216, 217 The LCO amplitude, 322323, 326
Illustrative numerical example, 364366 LDP. See Laplace domain Possio (LDP)
Implicit function theorem, 234 L1 Fourier transform, 73
Incompressible, 116, 122, 215, 266, 328330, Liapunov, 4
385, 386 Lift, 6, 10, 83, 91, 102, 156, 160, 302, 326, 352
Incompressible flow, 6, 7, 115116, 146150, due to gust, 352
155, 241, 243, 260265, 329330, and moment, 3, 57, 101, 155, 156, 271,
342344, 368 303, 305, 352, 353, 362
Infinitesimal generator, 28, 184186 and moment calculation, 299303
Initial-value problem, 179, 184, 309, 310 Limit cycle oscillation (LCO), 6, 269326, 386
Inner product, 14, 15, 18, 31, 38, 40, 168, 217, Linear aeroelasticity theory, the finite plane
253, 343 case, 7280
Inputoutput point of view, 269 Linear Goland beam model, 349, 368
Inputoutput relation, 5, 107, 184, 296 Linear Goland model, 41, 65, 97, 249, 270
Inverse Fourier transform, 94, 110, 325 Linear homogeneous equation, 69
Inverse Laplace transform, 162, 164, 166, 179, Linearization
194, 195, 203, 228, 282, 295, 364, 384 BeranStraganac model, 249
Isentropic flow, 5051, 54, 296, 326, 332, 337 DowellHodges model, 249256
Isolated simple poles, 179 Linearizing the Euler full potential equation,
Isolated singularities, 234 256259
Isometric, 17, 45 Linear model, 5, 40, 322, 337
Linear Possio equation, compressible flow,
122155
J Linear Possio integral equation, 106122
Jump, 5, 57, 62, 73, 78, 83, 105, 107, 112, 155, The linear TDP kernel, 258
180, 181, 212, 213, 266, 271, 281, 298, Linear theory, 5, 72, 103, 336, 376379
368 Linear time domain airfoil dynamics, 155183
Logarithmic singularity, 5, 138, 174, 197, 205
Long-slender wing, 2
K Low-speed wind tunnels, 367
Kinetic energy, 16 Lp Lq Fourier transform, 72, 112, 373
Kolmogorov model, 7, 349 Lp , 1<p<2, 7, 116, 266, 381
Kolmogorov theory, 349
Kussner doublet function, 105, 111, 147, 278
Kussner function, 355 M
Kutta condition, 4, 5, 57, 62, 74, 93, 101, 110, Mach number, 54, 69, 122, 245248
112, 116, 127138, 142, 271, 284, 289, Mass/inertia operator, 14
329, 332, 369 Material fatigue, 349
KuttaJoukowsky boundary conditions, 269 Matrix Volterra equation, 25
KuttaJoukowsky conditions, 67, 71, 92, Method of variation of parameters solution, 25,
96100, 105, 277278, 281, 329, 334, 187, 309
336 Meyer, R.E., 8, 63, 102, 328, 329
Mikhlin multiplier, 5, 115, 116, 123, 147, 240,
244, 294, 377
L Modal expansion, Greens function, 2021
LagrangianEuler fluid structure boundary, 2 Modal expansion of solution, 179183
Laplace domain, 5, 121122, 151154, 197, Modal representation, 13, 20, 21, 200
199 Modal stability, 29
Index 397
Modes and mode shapes, 2, 3, 9, 1820, Orthogonal, 13, 21, 34, 49, 183, 206, 215, 217
206208 Oscillating wing, 266
Mode shape, 5, 19, 20, 22, 26, 171, 175177, Outer layer, 333
183, 200, 235, 242, 319, 339, 340, 342
Moment, 5, 10, 37, 89, 156, 163, 212, 303,
307, 309, 339, 352 P
Moment-balance, 37 Pade approximation, 5
Multiplier, 45, 46, 77, 116, 117, 123126, 133, Palatal flutter, 7, 367, 368
151, 244, 292, 382 PaleyWiener theorem, 45
Perfect gas law, 3, 48, 49
Perturbation formula, 232, 235, 246, 346
N Perturbed eigenvalues, 256
NavierStokes equation, 6, 327, 329, 330, 332 Physical constants, 4748
Neumann expansion, 142, 143, 194, 228, 247, Physically realizable nonlinear convolution,
384 304
Neumann type boundary value problem, 106 Piezoelectric power generation, 7
Neutrally stable, 20, 26 Piezoelectric power harvesting from wind, 367
Newton step, 256 Piezzoelectric strip, 30
Non-aircraft and non-flight applications, 7 Pitch, 2, 4, 7, 9, 55, 78, 80, 90, 91, 98, 344
Non-aircraft areas of application of aeroelastic Plancheral theorem, 45
flutter, 367 Plunge axis, 55
Non-circular terms, 384 Point, 3, 4, 6, 8, 30, 35, 3941, 44, 68, 71,
Non-circulatory terms, 5, 168, 198, 201, 202, 77, 91, 92, 101, 113, 118, 153, 172,
289, 298 188192, 194, 211, 226, 227, 232, 234,
Nonhomogeneous Cauchy problem, 186187 239, 255, 269, 318, 322, 337, 350,
Nonhomogeneous linear equation, 81 362364, 367, 377, 380, 383
Nonlinear aeroelastic dynamics, 298 Point spectrum, 44, 153, 189192, 194, 211,
Nonlinear convolution evolution equation, 298, 226, 227
303309 Positive definite, 166
Nonlinear convolution evolution equation in a Positive square root, 31, 38
Hilbert space, 298 Possio equation, 5, 6, 87, 101, 111, 112, 115,
Nonlinear equations, 269, 349 121123, 127, 136, 142, 144146,
Nonlinear filtering, 8 148150, 152, 156, 159, 200, 204206,
Nonlinear Possio integral equation, 296298 258, 266, 269, 281285, 295, 297, 298,
Nonlinear stability theory, Hopf, 386 342, 352, 355, 377, 378, 380, 386
Nonlinear static Euler equation, 67 Possio equation time domain solution, 148150
Nonlinear structure models, 3, 4, 4043, 47, Possio integral equation (1938), 4, 76, 101,
96101, 248256, 349 103267, 281
Nonlinear Volterra equation, 91 Potential energy, 1216, 38
Nonnegative definite, 13, 14, 27, 31, 33, 3638, Potential flow, 4, 54, 112, 147, 325, 333
46, 252 Power series expansion, 71, 88, 98, 103105,
Non-rational component, 183, 223 147, 269, 272277, 370376, 385
Nonviscous flow, 3, 47, 4958, 328 Prandtl boundary layer hypothesis, 7
Nonzero coupling, 22 Prandtl boundary layer theory, 332
Nonzero static solution, 4, 6771 Prandtl equations, 332
Nonzero viscosity, 327 Prandtl limit equations, 332335
Normalization, 156, 200, 218224, 243 Predominant 3rd harmonic, 322
No-slip boundary condition, 332 Pressure doublet, 305, 352
Numerical example, 364 Pressure jump, 5, 57, 62, 73, 78, 83, 105, 107,
112, 155, 266, 271, 281, 298, 368
O Program, 65, 241, 243, 260266, 318
Operational ampliflier, 34, 46 Projection operator, 115, 124, 184, 188
Optimal control theory, flutter suppression, Proportional damping, 27
337347 Pseudo resolvents, 216
398 Index
R Spectral density, 7, 349351, 354, 355, 361,

Radius of gyration, 37 364366
Random-gust model, 349 Spectral theory, aeroelastic modes, 169183
Rate feedback, 30 Spectrum, 13, 18, 27, 31, 32, 40, 41, 44, 153,
Resolvent, 13, 15, 18, 21, 22, 27, 28, 32, 124, 189192, 194, 211, 226227
169171, 178180, 187, 190195, 200, Spectrum of A, 13, 18, 31, 32, 44, 189192,
204, 208211, 216, 226228, 230, 253, 226227
254, 316 Speed parameter, 4, 65, 272, 315, 337
Resolvent of A, 18, 28, 227, 253, 781 Square root, 13, 15, 31, 38, 44, 45, 108, 109
Resolvent set, 13, 194, 254 Square root operator, 1314
Reynolds number, 48, 332 Stability, 37, 9, 11, 18, 2635, 40, 4244,
Riemann problem, 326 63, 65, 91, 101, 103, 111, 146, 169,
Riemann sheet, 181 189191, 234242, 315318, 323, 326,
Riesz property, 222 329, 330, 337, 339344, 347, 379, 385,
Rigid-body response, 349 386
Road-runner ground vehicle, 364 aeroelastic modes, 379384
Robust control, 3 curve, 6, 234242, 344, 347, 382384
Rodden computer algorithm, 5 curve for M D 0, 239
Rodden, R.P., 266 index, 29, 34
Root-locus, 234239, 244, 344 of nonlinear system, 317318
Root locus and stability curve, 234239 operator, 5, 169, 189191, 341
of the structure in air flow, 11, 323, 330,
379
S Stabilizability, 7, 339342
Sears formula, 120121 Stable mode, 235
Sears function, 162, 281 State space formulation, 16
Second-degree polynomial, 302 Statespace model, 7, 337
Second normalization, 220224 Statespace representation, 5, 184194, 196,
Self-adjoint and nonnegative definite, 27, 31, 230
3638, 252 State space theory, 5, 183230, 309
Self-straining actuators, incompressible flow, State space theory, M0, 224228
342344 Static NavierStokes equation, 330
Semigroup generated by As , 28, 168, 194196, Static solution, 4, 65, 6771, 97, 99, 100, 230,
228230, 309 249, 256, 330, 331
Semigroup on Riesz space, 315316 Static/steady state equation, 330
Semigroup theory, 9, 123, 125, 170, 315 Stationary Gaussian random processes, 353
Separable Hilbert space, 185, 338 Steady oscillatory motion, 266
Series solution, 251, 309315 Steady-state response, 65, 309, 321
Several-point masses, 40 Steady state solution, 65, 315, 335, 337, 370,
Shocks, 6, 7, 95, 326 379
Shocks in isentropic flow, 326 Steady-state structure response, 8991
The simplest wing structure model, 55 Steady-state structure response to gust, 364
Singular case, 326 Stiffness operator, 14, 46
Sinusoidal-gustmodel, 349 Stochastic lift and moment, 362
Slant, 215 Strip theory, 3, 47, 61, 75
Small viscosity, 6, 332 Strongly stable, 27, 28, 310, 315317
Sohngen, 4, 8, 101, 103 Structural dynamics and fluid dynamics, 1
Solution of Possio equation Laplace domain, Structure dynamics nonlinear
M D 0, 121122 Beran Straganac, 60
Some open problems, 332 DowellHodges, 60
Sonic case, 6, 151155, 196206 Structure equations, 62, 68, 7475, 167, 249,
Space-time covariance function, 350 289, 368, 386
Span axis, 3, 55 Structure modes, 39, 239, 240, 242, 243, 339
Spectral analysis, 9 Structure modes and mode shapes, 1826
Index 399
Structure response to turbulence, 362364 Tricomi operator, 93, 118, 127, 152, 168
Superstability, 3, 34, 35, 343 Tricomi theorem, 129, 130
Symbolic software, 1 Tricomi transform, 87
Symmetric case, 7, 3940 Turbulence field, 349351
System, 1, 2, 4, 6, 7, 26, 29, 30, 34, 40, 42, Turin, Italy, 267
43, 55, 65, 96, 101, 111, 115, 184, 185, Twist angle, 55
189, 190, 232, 239, 245, 259263, 296, Two-point boundary value problem, 172176,
310, 314317, 339, 341, 342, 351, 362, 363, 383
367, 370, 385 Typical section, 3, 4, 47, 6163, 112, 115, 290,
function, 352 298, 325, 334
inputoutput relation, 184 airfoil theory, 7580
approximation, 4, 63
case, 63, 101, 112, 325, 376, 377
T (strip) theory, 6163
Taylor frozen field hypothesis, 350 theory, 3, 47, 6163, 82, 335
Taylor series, 4, 69 Typical section theory, nonzero , 8284
Taylor series expansion, 4 Typical section/zero angle of attack, 113122,
TDP. See Time domain Possio (TDP) 330332
Tear drop, 2, 43
Tear drop shape, 43
Temporal wind gust model, 350351
Theodorsen, 6, 103, 159, 161, 164, 197, 266, U
355 Uniform stability, 29
approximation, 6 Unsteady motion, 266
function, 6, 159, 161, 164, 197, 355
Thermal diffusivity, 328
Thermodynamic relation, 48, 328 V
Thermodynamics, 3, 49, 63, 328 Velocity potential, 6, 51, 52, 66, 9194
3rd harmonic, 6, 321, 322 Vertical wind-gust component, 350
Time domain Volterra equation, 25, 91, 259, 310315
airfoil dynamics, the sonic case, 155169, Volterra expansion, 119, 304, 305
196209 Volterra integral equation, 46, 255
description, 9 Volterra operator, 118, 129, 152
operator, 293 Volterra series, 309315, 386
potential, 150151, 325 Vortex-free, 4, 47, 51
potential flow, 325
solution, 5, 14, 46, 112, 122, 146, 154, 183,
184, 224, 315
W
solution incompressible flow, 146151
Wagner, 8, 103, 161162
solution M D 1, 5, 154155
Wagner function, 161162
Time domain Possio (TDP), 258, 281297
Wind
Time-invariant version, 65
gust, 7, 349351
Time marching, 102, 326
turbulence, 349
Timoshenko, S., 10, 11, 19, 27
Wing camber model, 2, 43, 45, 46
Torsion control, 344347
Wing model, 2, 10, 3540, 45, 55
Torsion flutter, 7, 241, 243, 364
World War II, 267
Torsion modes, 19, 20, 26, 40, 235, 237, 241,
243, 261262, 364
Total displacement, 56, 61
Transform of the lift, 156 Z
Transform of the moment, 156 Zero angle of attack, 4, 8489, 113119,
Transonic dip, 4, 5, 79, 80, 234 147155, 246, 281, 298, 315, 323, 325,
Transonic peak, 78 326, 330331, 334, 351
Tricomi, F.G., 4, 77, 87, 93, 101, 118, 127, Zero coupling, 26
129, 130, 152, 153, 168 Zero slip, 329

A V Balakrishnan Aeroelasticity The Continuum Theory

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A V Balakrishnan Aeroelasticity The Continuum Theory

Uploaded by

Copyright:

Available Formats

Aeroelasticity

ISBN 978-1-4614-3608-9 ISBN 978-1-4614-3609-6 (eBook)

Springer Science+Business Media, LLC 2012

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Durham, USA Earl Dowell

from the mathematics alone. What we do is indeed applied mathematics in the

It is a pleasure to acknowledge the generous help of my colleagues and my peers:

Los Angeles, California, USA A V Balakrishnan

5.7 Flutter Analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 230

The subject matter of aeroelasticity is the dynamics of a structure in air flow,

A V Balakrishnan, Aeroelasticity: The Continuum Theory, 1

question here concerns the presence of shocks. There is no mathematical proof

Notes and Comments

It is interesting to note that in the beginning in aeroelasticity theory only

2.2 The Goland Beam Model

This is the earliest continuum model introduced in aeroelasticity by Goland [78] in

and it is assumed that the aspect ratio

A V Balakrishnan, Aeroelasticity: The Continuum Theory, 9

Fig. 2.1 Wing model x

and the torsion by the (Saint Venant, string, or Timoshenko) equation:

where the forcing terms:

h.t; 0/ D 0I h0 .t; 0/ D 0I h00 .t; `/ D 0I h000 .t; `/ D 0I

h00 .t; 0/ D 0I h000 .t; 0/ D 0I h00 .t; `/ D 0I h000 .t; `/ D 0;

h.t; 0/ D 0I h0 .t; 0/ D 0I .t; 0/ D 0;

2.3 Time Domain Analysis

We begin with the abstract differential operator denoted A.

Although strictly speaking we need to distinguish between these operators in

Ax; y D Ay; x for y in D.A/:

Or, A is self-adjoint. Furthermore we can calculate that

Or, A is nonnegative definite.

Or, zero is not in the spectrum of A. Any such that

is called an eigenvalue of A and x is the eigenvector corresponding to .

and we have the eigenfunction expansion:

The square root operator

The Energy Space

endowed with the energy inner product:

Denote this inner-product space by H.

be a Cauchy sequence in H. Then

thus it is seen that yn is a Cauchy sequence in H , by a similar argument. Hence it

For the clampedclamped case, for example, it is given by

0<s<` .0/ D .`/ D 0:

See [63] for more on this.

YP .t/ D AY .t/ C V.t/; (2.11)

Thus defined it is closed with domain dense in H.

jjY .t/jj D jjS.t/Y jj D jjY jj: (2.12)

The energy remains a constant, in other words.

m.t/ D S.t/Y; S.t/Y E D jjY .t/jj2 :

Then for Y in D.A/

and hence for Y in D.A/

It should be noted that the differential equation:

YP .t/ D AY .t/ C V.t/

has the solution

2.4 Structure Modes and Mode Shapes

Hence it follows that 2  0I and hence 2 D ! 2 ; ! real. But cannot be zero,

!k2 mhk .y/ D EI h0000

plus the CF end conditions:

The !k then are pure bending modes.

1 C Cosh k Cos k D 0 (2.17)

C1 .sinh !k y sin !k y/ C C2 .sinh !k y C sin !k y/; 0 < y < `:

For more see [32, 34].

yielding an interesting variation of (2.4).

Modal Expansion: Greens Function

Hence it follows that 2 0I and hence 2 D ! 2 ; ! real. But cannot be zero,

u./ L2 Rm ; 0; T for every 0 < T < 1:

Hence Ad now generates a contraction semigroup Sd .t/t 0:

denote the infinitesimal generator of a C0 semigroup S.t/; t 0, over the Hilbert

jjS.t/jj Me.! C/t

f1 .s/ D c sinh ..` s//; 0 < s < `;

GJ I cosh ` C g sinh ` D 0:

D.A / D jf =f 00 ./ L2 .0; `/j; A f D gI