Weak and Strong Solutions of a Nonlinear

Subsonic Flow-Structure Interaction: Semigroup

Approach
J.T. Webster
University of Virginia
Department of Mathematics
Charlottesville, VA

January 27, 2011

1
1 Abstract
We consider a non-rotational, subsonic flow-structure interaction describing the
flow of gas above a flexible plate. A perturbed wave equation describes the
flow, and a second-order nonlinear plate equation describes the plate’s displace-
ment. It is shown that the linearization of the model generates a strongly
continuous semigroup with respect to the topology generated by “finite energy”
considerations. An interesting feature of the problem is that linear perturbed
flow-structure interaction is not monotone with respect to standard norm de-
scribing the finite energy space. The main tool used in overcoming this difficulty
is the construction of a suitable inner product on the finite energy space which
allows the application of ω−maximal monotone operator theory. The obtained
result allows us to employ suitable perturbation theory in order to discuss well-
posedness of weak and strong solutions corresponding to several classes of non-
linear dynamics including the full flow-structure interaction with Von-Karman,
Berger’s and other semilinear plate equations.

Key terms: flow-structure interaction, non-linear plate equation, von Kar-

man nonlinearity, nonlinear semigroup, well-posedness

2 Introduction and Motivation

This paper aims to provide analysis of models which arise in aeroelasticity de-
scribing the dynamics and interaction between an oscillating structure and a
surrounding inviscid flow. We aim to study solutions to a coupled system of
PDEs describing the interaction between a gas flow around a thing structure
moving through the environment with a given velocity U . One of the central
problems of aeroelasticity is the determinantion of the speed of the aircraft
corresponding to the onset of an enedemic instability termed wing “flutter”
cole,dowell,dowell1,bal1,bal2,amjad
[12, 13, 14, 2, 3, 28]. It is well known that flutter in the wing of an aircraft may
occur at high speed, and can cause the wing to break down. Thus, determining
conditions which prevent such instability are of prime concern.
At present, related studies are primarily experimental and computational:
NASTRAN and CFDO codes which produce numerical algorithms and schemes.
While these methods address various situations, they are based on finite dimen-
bolotin,b-c,bal
sional approximations of a continuum model fully described by PDEs [5, 6, 1].
The PDE nature of the physical phenomena may not be adequately reflected by
these approximations. This is particularly true in dealing a highly oscillatory
model, where large frequencies (often causing instability) are not accounted for

2
by finite dimensional analysis.

2.1 Summary
We present results on the Hadamard well-posedness of the nonlinear system
for subsonic velocities in the absence of rotational inertia. After presenting
the flow-structure model, we proceed by describing the problem abstractly in
terms of the dynamics operator T on the state space Y . We first consider
the linearization of the problem; we will find that −(T + ωI) is inadequate for
generation for all ω ≥ 0. However, we introduce an equivalent topology on the
space Y , and we show that for suitable ω, −(T + ωI) generates a C0 semigroup
on Y in the modified topology. However, through equivalence of norms, this
is sufficient for the existence and uniqueness of solutions on Y in the original
topology. Via perturbation theory, local-in-time existence and uniqueness for
the nonlinear system will reduce to the well-posedness of the linearized model.
In the final section, we specialize F and improve the abstract results obtained in
the previous sections. Specifically, we cite results on the Airy sharp regularity of
the von Karman nonlinearity, which give that it is in fact locally Lipschitz on our
state space, allowing us to employ perturbation theory. Finally, we take further
advantage of the von Karman nonlinearity by noting an a priori global-in-time
bound on the local-in-time solution, yielding global existence and uniqueness of
solutions to the von Karman nonlinear flow-structure model.

3 Description of the model

We consider the interaction of a clamped plate with a field or flow of gas above
dowellnon,dowellnon1
it [15, 16]. The oscillatory dynamics of the plate are governed by a second-order
nonlinear plate equation. To describe the behavior of the gas, we make use of
bolotin,dowell,dowell1
the theory of potential flows [5, 13, 14].

We work in R3+ = {(x, y, z) ∈ R3 z ≥ 0} and consider a thin place of negli-
gible thickness (in the z−direction), as usual in the modeling of thin structures
ciarlet
[11]. The top surface of the wing, where the flow-structure interaction takes
place, rests in the x-y-plane and will be denoted Ω ⊂ R2 . The plate’s motion
(i.e. the wing velocity) takes place in the x-direction at velocity U , where U = 1
represents the speed of sound. (U can also be thought of as the unperturbed
flow velocity.)
The function φ(x, y, z; t) : R3+ × R+ → R corresponds to the potential flow,
∂φ
with = φt giving the flow velocity and ∇φ the flow pressure. The function
∂t
∂u
u(x, y; t) : R2 × R → R represents the z-displacement of the plate, with = ut
∂t

3
 being the z-velocity of the plate. Then our coupled system is as follows (taking
x = (x, y, z) or (x, y), as dictated by context):
First, the plate equations with clamped boundary conditions are given by



 utt − α∆utt + ∆2 u + F (u) = p(x, t), x ∈ Ω, t > 0


∂u  (3.1) platesystem
u|∂Ω =  = 0,


 ∂n ∂Ω

u|t=0 = u0 (x), ut |t=0 = u1 (x).


Assuming that the flow of the gas occurs above the plate in the x-direction,
then p(x, t) is the aerodynamical pressure of the flow on the plate and is given
by
p(x, t) = p0 (x) + ν(∂t + U ∂x )φ|z=0 , x ∈ Ω, (3.2) couplingterm

where p0 (x) ∈ L2 (Ω) and ν > 0 is a parameter describing the intensity of the
interaction between the flow and the plate (typically taken to be one, without
loss of generality); α ≥ 0 is a constant corresponding to the presence of rota-
tional inertia in the model - we will restrict our attention to for a bulk of this
treatment to the more difficult case where α = 0. We take F to be an abstract
nonlinearity, requiring F : H 2 (Ω) → L2 (Ω) to be locally Lipschitz. (Here H s (D)
denotes the Sobolev space of order s defined on the domain D, and H0s (D) is
the closure of C0∞ (D) in the norm || · ||H s .) Secondly, a modified wave equation
describes the flow:



 (∂t + U ∂x )2 φ = ∆φ, x ∈ R3+
 (
ν(∂t + U ∂x )u(x, y; t), (x, y) ∈ Ω


∂z φ|z=0 = , (3.3) flowsystem


 0, (x, y) ∈
/Ω

 
φ|
t=0 = φ0 (x), φt t=0 = φ1 (x).


vonK Remark 3.1. We will begin by taking F ≡ 0, i.e. linearizing the problem. For
the resulting linear problem we shall establish the existence of a strongly contin-
uous semigroup describing the flow. This will allow us to produce results which
hold for a class of general nonlinearities, as described above. Later, we special-
ize F , and in doing so, we will strengthen our abstract results. We provide a
detailed analysis of the case when F is the von Karman nonlinearity, defined it
in terms of Airy’s stress function v(·):

F (u) ≡ −[v(u), u] (3.4) Airy

v(u) ≡ ∆−2 [u, u] in Ω (3.5) airy1

4
 u = ∇u = 0 on ∂Ω, (3.6) airy2

with
[u, v] = (∂x2 u)(∂y2 v) + (∂y2 u)(∂x2 v) − 2(∂xy
2 2
u)(∂xy v) (3.7) bracket
berger,ciarlet
denoting the von Karman bracket [4, 11].
Remark 3.2. It is possible, and physically pertinent, to consider boundary con-
ditions for the plate other than clamped. For instance, one might take the
boundary conditions to be hinged, partially clamped and hinged, or free (per-
haps the most mathematically demanding). Here, we focus on clamped bound-
ary conditions, as they represent the most immediately relevant situation and
remain mathematically challenging.
Our approach makes use of semigroup theory. This naturally leads to the
solutions
notion of strong solutions, as defined below in Section 3.3; we then obtain gen-
eralized solutions as strong limits of strong solutions. We will take our state
space for the problem to be

Y = H 1 (R3+ ) × L2 (R3+ ) × H02 (Ω) × L2 (Ω)

corresponding to the vector (φ, φt ; u, ut ).

3.1 Past Considerations

past
Flow-structure models have attracted considerable attention in the past liter-
dowell
ature [13]. However, the vast majority of the work done has been devoted to
dowell,dowell1,dowellnon,dowellnon1,bolotin
numerical and experimental studies. [13, 14, 15, 16, 5]. In contrast, our interest
here concerns PDE aspects of the problem, including the fundamental issue of
well-posedness of finite energy solutions.
b-c,b-c-1
For the latter, influential papers have been written [6, 7] which provide rather
complete analysis of solutions in the rotational regime ( corresponding to α > 0).
The approach presented there relies on two main components: (i) an explicit
solver for the flow equation (Kirchhoff’s formula), (ii) estimates for the bound-
ary traces of the flow, which rely on exceedingly technical microlocal analysis
miyatake b-c,b-c-1
estimates derived in [21]. The approach developed in [6, 7] has the advantage
of being able to handle both subsonic and supersonic regimes; however, it de-
pends critically on the enhanced regularity of the velocity of the plate, namely
that ut ∈ H 1 (as is guaranteed in the rotational model). In fact, this approach
ryz,ryz1,ryz2
has been later extended to thermoelastic plate models [24, 25, 26] which also
exhibit H 1 regularity of the velocity in both the rotational and non-rotational
redbook
cases due to the analytic regularizing effects induced by thermoelasticity [20]. A
springer
recent book [10] provides an account of relevant results, including more recent

5
 applications of the Galerkin method .
Taking α > 0, which corresponds to the presence of rotational inertia in the
model, modeled by the term −α∆utt , provides a helpful regularizing term in en-
b-c,b-c-1, ryz, ryz1, ryz2
ergy calculations, and is the key factor in the analyses found in [6, 7, 24, 25, 26].
In our treatment we take α = 0, corresponding to the more difficult non-
rotational model, and we approach the problem from the semigroup point of
view - without any reliance on explicit solvers for the flow equation or Galerkin
constructions for the flow. The advantage of this approach, in addition to solv-
ing the fundamental wellposedness question, is the potential for an array of
important generalization to include more general flow equations and more gen-
eral nonlinearities appearing in the structure. It is also clear that the methods
presented here can be applied to the rotational case α > 0, but we restrict our
attention in the bulk of the treatment to the more challenging case α = 0. The
alphage0
short section 5.3 discusses how our setup also accomodates the case α > 0.

3.2 Energies
energiessection
Formal calculations with Green’s formula yield energy equalities, describing the
dynamics of the model. First, we have the energy of the plate:
Z Z
1 2 1
Epl (t) = |ut (t)| + |∆u(t)|2 + Π(u), (3.8) plateenergy
2 Ω 2 Ω

where Π0 (u) = F (u) − p0 , with Π0 (·) the Frechet derivative of Π. This quantity
is positive and corresponds to the total kinetic and potential energy of the
oscillating plate. Secondly, the energy of the flow of gas around the wing is
given by
Z Z Z
νn o
Ef l (t) = |φt (t)|2 + |∇φ(t)|2 − U 2 |φx (t)|2 . (3.9) flowenergy
2 R3+ R3+ R3+

In the subsonic case (when U < 1) this term is positive. The total physical
energy is given by E(t) = Ef l (t) + Epl (t). Lastly, the interactive energy is given
by Z


Eint (t) = νU (ux ) φ z=0
. (3.10) Intenergy
Ω

This energy corresponds to the interaction between the flow and the plate’s
displacement, and takes place on Ω, the interface of the two media. Note that
Eint (t) has indefinite sign.
The total energy (including the interaction on the interface) becomes

E (t) = Epl (t) + Ef l (t) + Eint (t); (3.11) totalenergy

6
 E (t) also has indefinite sign. The (formal) balance law for the total energy gives
E (t) = E (0) for all times t. Note: The balance law will not hold if U ≥ 1.
The structure of the energy functionals dictate the topological setup for the
problem; finite energy constraints manifest themselves in the natural require-
ments on the functions φ and u. Specifically,

u ∈ C(0, T ; H02 (Ω)) ∩ C 1 (0, T ; L2 (Ω));

φ ∈ C(0, T ; H 1 (R3+ )) ∩ C 1 (0, T ; L2 (R3+ )).

3.3 Definition of Solutions

solutions
In the discussion below, we will encounter strong (classical), generalized (mild),
and weak (variational) solutions.
Remark 3.3. In making use of semigroup theory, we will obtain generalized solu-
tions, which are the strong limit of strong solutions. These solutions will satisfy
platesystem
flowsystem
an integral formulation of (3.1)-(3.3), and are called mild by some authors. In
our treatment, we will produce a unique generalized solution, and show that
this, in turn, produces a unique weak solution.
We now define strong, generalized, and weak solutions:

Strong Solutions


A pair of functions u(x, y; t), φ(x, y, z; t) such that

u(x, t) ∈ C(0, T ; H02 (Ω)) ∩ C 1 (0, T ; L2 (Ω)), (3.12) platereq

φ(x, t) ∈ C(0, T ; H 1 (R3+ )) ∩ C 1 (0, T ; L2 (R3+ )) (3.13) flowreq

platesystem
flowsystem
is said to be a strong solution to (3.1)-(3.3) on [0, T ] if

• (φt ; ut ) ∈ L1 (a, b; H 1 (R3+ ) × H02 (Ω)) for any (a, b) ⊂ [0, T ].

• (φtt ; utt ) ∈ L1 (a, b; L2 (R3+ ) × L2 (Ω)) for any (a, b) ⊂ [0, T ].

• φ(t) ∈ H 2 (R3+ ) and ∆2 u(t) ∈ L2 (Ω) for almost all t ∈ [0, T ].

• The equation
utt + ∆2 u + F (u) = p(x, t)

holds in L2 (Ω) for almost all t > 0.

• The equation
(∂t + U ∂x )2 φ = ∆φ

holds for almost all t > 0 and almost all x ∈ R3+ .

7
 platesystem
flowsystem
• The boundary conditions in (3.1), (3.3) hold for almost all t ∈ [0, T ] and
for almost all x ∈ ∂Ω, x ∈ R2 respectively.

• The initial conditions are satisfied pointwisedly; that is

φ(0) = φ0 , φt (0) = φ1 , u(0) = u0 , ut (0) = u1 .

As stated above, generalized solutions are strong limits of strong solutions;

these solutions will correspond to semigroup solutions for an initial datum out-
side of the domain of the generator.

Generalized Solutions


A pair of functions u(x, y; t), φ(x, y, z; t) is said to be a generalized solution
platesystem
flowsystem platereq flowreq
of the problem (3.1)-(3.3) on the interval [0, T ] if (3.12) and (3.13) are satisfied
and there exists a sequence of strong solutions (φn (t); un (t)) such that
n o
lim max ||∂t φ − ∂t φn (t)||L2 (R3+ ) + ||φ(t) − φn (t)||H 1 (R3+ ) = 0
n→∞ t∈[0,T ]

and
n o
lim max ||∂t u(t) − ∂t un (t)||L2 (Ω) + ||u(t) − un (t)||H02 (Ω) = 0.
n→∞ t∈[0,T ]

We now introduce the definition of weak solutions for the above model:

Weak Solutions


A pair of functions u(x, y; t), φ(x, y, z; t) such that

u(x, t) ∈ WT ≡ {u ∈ L∞ 0, T ; H02 (Ω) , ∂t u(x, t) ∈ L∞ 0, T ; L2 (Ω) },

φ(x, t) ∈ VT ≡ {φ ∈ L∞ 0, T ; H 1 (R3+ ) , ∂t φ(x, t) ∈ L∞ 0, T ; L2 (R3+ ) },

platesystem
flowsystem
is said to be a weak solution to (3.1)-(3.3) on [0, T ] if
• u(x, 0) = u0 (x, y), ut (x, 0) = u1 (x, y)
and φ(x, 0) = φ(x, y, z), φt (x, 0) = φ1 (x, y, z)

•
Z T 


− ∂t u(t), ∂t w(t) L2 (Ω) + ∆u(t), ∆w(t) L2 (Ω)
0



+ F (u)(t) − p0 , w(t) L2 (Ω) + ν γ[φ(t)], ∂t w(t) + U ∂x w(t) L2 (Ω) dt


= − u1 − νγ[φ0 ], w(0) L2 (Ω)

8
 for all test functions w ∈ WT with w(T ) = 0.

•
Z T 

− (∂t + U ∂x )φ(t), (∂t + U ∂x )ψ(t) L2 (R3 )
+
0



+ ∇φ(t), ∇ψ(t) L2 (R3 ) − (∂t + U ∂x )u(t), γ[ψ(t)] L2 (Ω) dt
+


= − φ1 + U ∂x φ0 , ψ(0) L2 (R3 )
+

for all test functions ψ ∈ VT such that ψ(T ) = 0.


Here and below, for φ ∈ H 1 (R3+ ), γ[φ] = φz=0 is the trace of φ on the
plane {x : z = 0}.

3.4 Statement of Main Results

results
We now formally state the results to follow in this treatment. We assume that
0 ≤ U < 1 (subsonic) and p0 ∈ L2 (Ω). Here we assume α = 0 (rotational inertia
absent).
Remark 3.4. In fact, these results will hold for α > 0, making the appropriate
changes, but for the sake of exposition we restrict our discussion of this case to
alphage0
Section 5.3.
platesystem
linearresult Theorem 3.1 (Linearization). Suppose F ≡ 0 in (3.1).
platesystem
flowsystem
1. Linear Generation The dynamics operator associated to (3.1)-(3.3) gen-
erates a strongly continuous semigroup on the state space

Y = H 1 (R3+ ) × L2 (R3+ ) × H02 (Ω) × L2 (Ω).

2. Strong Solutions Assume

u1 ∈ H02 (Ω), u0 ∈ W ≡ {w ∈ H02 (Ω) : ∆2 w ∈ L2 (Ω)}.

Moroever, suppose

φ0 ∈ H 2 (R3+ ) and φ1 ∈ H 1 (R3+ )

with the following compatibility condition in place:

(
 u1 + U ∂x u0 if x ∈ Ω
∂z φ0 z=0 = . (3.14) compatibilitycondition
0 if x ∈
/Ω

9
 platesystem
flowsystem
Then (3.1)-(3.3) has a unique strong solution for any interval [0, T ]. This
solution possesses the properties

(φ; φt ; φtt ) ∈ L∞ (0, T ; H 2 (R3+ ) × H 1 (R3+ ) × L2 (R3+ )), (3.15)

(u; ut ; utt ) ∈ L∞ (0, T ; W × H02 (Ω) × L2 (Ω)) (3.16)

and satisfies the energy equality E (t) = E (s) for t > s, with E (t) defined
totalenergy
as in (3.11).

3. Generalized and Weak Solutions Assume

u0 ∈ H02 (Ω), u1 ∈ L2 (Ω), φ0 ∈ H 1 (R3+ ), and φ1 ∈ L2 (R3+ ).

platesystem
flowsystem
Then (3.1)-(3.3) has a unique generalized solution on any interval [0, T ].
This solution satisfies E (t) ≤ E (s) for t > s.

Every generalized solution is also weak.

abstractnonlinear Theorem 3.2 (Abstract Nonlinearity). Suppose F : H 2 (Ω) ∩ H01 (Ω) → L2 (Ω)
is locally Lipschitz. Then under the strong solutions assumptions of the pre-
platesystem
flowsystem
vious theorem, (3.1)-(3.3) has a unique local-in-time strong solution; under the
platesystem
flowsystem
generalized and weak solutions assumptions, (3.1)-(3.3) has a unique local-
in-time generalized (and hence weak) solution.
Moreover, if the solution (strong, weak, or generalized) is bounded in Y for
all t ∈ R+ , then it is global-in-time.

In particular, specializing F to the von Karman nonlinearity:

Airy airy2
vonkarmansolution Theorem 3.3. If F (u) = −[v(u), u] as defined in (3.4)-(3.6), then under the
linearresult platesystem
flowsystem
strong solutions assumptions Theorem 3.1, for all T , (3.1)-(3.3) has a unique
strong solution on [0, T ]; under the generalized and weak solutions assump-
platesystem
flowsystem
tions, for all T , (3.1)-(3.3) has a unique generalized (and hence weak) solution
on [0, T ].

4 Abstract Setting
We first analyze the inhomogeneous problem, and consider the coupled system
driven by a forcing term in both the plate and flow equations.
Our first task is to express the system in abstract semigroup form. Let
A : D(A) ⊂ L2 (R3+ ) → L2 (R3+ ) be the positive self-adjoint operator given by

∂f 
Af = −∆f + U ∂x2 f + µf , D(A) = f ∈ H 2 (R3+ ) :

=0 ,
∂z z=0


10
 where µ > 0. (In this context, the trace operator restricts to z = 0, i.e. ∂R3+ ).
Note that since 0 ≤ U < 1, we have that D(A1/2 ) ≡ H 1 (R3+ ), with “ ≡ ”
denoting topological equivalence. We now introduce the Neumann map

N0 : L2 (Ω) → L2 (R3+ )

given by the relation

∂Ψ  ∂Ψ 
Ψ = N0 w ⇐⇒ (−∆ + U ∂x2 + µ)Ψ = 0 in R3+ with  = w and  = 0,
∂z Ω ∂z Ω∗
(4.1) Neumannmap
with Ω∗ = ∂R3+ \Ω. When the domain of interest is R3+ , the operator N0 pos-
springer
sesses the same properties as in the case of a bounded domain [10]. Specifically,
N0 is continuous from L2 (Ω) → H 3/2 (R3+ ) ⊂ D(A3/4− ),  > 0; thus

A3/4− N0 : L2 (Ω) → L2 (R3+ ) is a continuous map. (4.2) contmap

Moreover, by Green’s formula we have that

N0∗ Af = f Ω = γ[f ], f ∈ D(A),



where N0∗ : L2 (R3+ ) → L2 (Ω) is the adjoint of N0 . We may extend this to all
functions f ∈ H 1 (R3+ ) = D(A1/2 ), since D(A) includes densely in D(A1/2 ) and
the trace operator is bounded on H 1 (R3+ ).
Introduce the differential operator D ∈ L D(A1/2 ), L2 (R3+ ) with

D(φ) ≡ 2U φx . (4.3) ops1

Now, we model the plate equation abstractly as

A u = ∆2 u, D(A ) = H 4 (Ω) ∩ H02 (Ω). (4.4) plateoperator

We take our forcing terms to be F1 and F2 which act on u and φ respectively.

To maintain as much generality as possible, we have introduced the forcing term
F1 corresponding to the flow equations (not initially present in the exposition
of the flow-structure model).
specialize Remark 4.1. When we specialize to the von Karman linearity, we will take

F1 (φ) ≡ −µφ, F2 (u) ≡ F (u) − p0 = −[v(u), u] − p0 , (4.5) vonknonlin

11
noting that F2 (u) = Π0 (u) when

1
Π(u) = ||∆v(u)||2L2 (Ω) − (p0 , u)L2 (Ω) . (4.6) potential
4
The superlinearity of F2 will play an indispensible role in global well-posedness
considerations of the full von Karman system.
With the above notation, taking ν = 1 without loss of generality, the abstract
model for the nonlinear flow-structure interaction is given by

φtt + A(φ + N0 (ut + U ux )) + D(φt ) + F1 (φ) =0


utt + A u − N0∗ A(φt
+ U φx ) + F2 (u) =0 (4.7) abstractsystem


φ(0) = φ , φ (0) = φ , u(0) = u , u (0) = u .
0 t 1 0 t 1

The state space for the above problem is

Y := Y1 × Y2 = D(A1/2 ) × L2 (R3+ ) × D(A 1/2 ) × L2 (Ω),

where the flow component (φ, φt ) ∈ Y1 , and the plate component (u, ut ) ∈ Y2 :

Y1 := D(A1/2 ) × L2 (R3+ ) ≡ H 1 (R3+ ) × L2 (R3+ ),

Y2 := D(A 1/2 ) × L2 (Ω) ≡ H02 (Ω) × L2 (Ω),
results
reconciling our state space with its earlier definition in Section 3.4.
The natural inner product on the state space Y is given by:
for y = (φ1 , φ2 ; u1 , u2 ) and y 0 = (φ01 , φ02 ; u01 , u02 )

(y, y 0 )Y =(φ1 , φ01 )D(A1/2 ) + (φ2 , φ02 )

+ < u1 , u01 >D(A 1/2 ) + < u2 , u02 >

=(A1/2 φ1 , A1/2 φ01 ) + (φ2 , φ02 )

+ < ∆u1 , ∆u01 > + < u2 , u02 >,

where the subscripts indicate what space the inner product is taken with respect
to; we have adopted the convention that (·, ·) refers to an inner product in
L2 (R3+ ) and < ·, · > in L2 (Ω). We also introduce the overall dynamics operator

12
 abstractsystem
T : D(T ) ⊂ Y → Y corresponding to the linearization of (4.7), given by:
  
0 −I 0 0 φ1
 A D U AN0 ∂x AN0  φ2 
 
Ty =  (4.8)
 
T
−I  u1 
 
 0 0 0
−U N0∗ A∂x ∗
−N0 A A 0 u2
 
−φ2
Aφ + D(φ ) + AN (u + U ∂ u )
 1 2 0 2 x 1 
= , (4.9)
 −u2 
A u1 − N0∗ A(φ2 + U ∂x φ1 )

with

D(T ) = (φ1 , φ2 ; u1 , u2 ) ∈ D(A1/2 ) × D(A1/2 ) × D(A 1/2 ) × D(A 1/2 ) :



Aφ1 + AN0 (u2 + U ∂x u1 ) ∈ L2 (R3+ ), A u1 ∈ L2 (Ω) .

Note that D(T ) densely embeds into Y by standard trace theory.

Analyzing D(T ) and the action of T , we see first that A u1 ∈ L2 (Ω), which
implies that u1 ∈ H 4 (Ω) (since A is the biharmonic operator). Hence, since
u1 ∈ H 2 (Ω) and u2 ∈ H 2 (Ω), it is certainly true that u2 + U ∂x u1 ∈ H 1 (Ω)
flowsystem
and AN0 (u2 + U ∂x u1 ) ∈ L2 (Ω); our coupling condition in (3.3) then gives that

∂z φ1 z=0 ∈ H 1 (Ω). Thus, since

Aφ1 + AN0 (u2 + U ∂x u1 ) ∈ L2 (R3+ ),

∂φ1 
we have that Aφ1 ∈ L2 (R3+ ) and ∈ H 1 (Ω). We can then conclude from
∂n z=0

elliptic theory that φ1 ∈ H 2 (R3+ ), so

D(T ) ,→ H 2 (R3+ ) × H 1 (R3+ ) × H 4 (Ω) × H 2 (Ω).

For the second-order flow-structure system, we derive the first order formu-
lation of the nonlinear model:

y 0 + T y = F (y); y(0) = y0 , (4.10) firstorder

with F given by F (φ1 , φ2 ; u1 , u2 ) = (0, −F1 (φ1 ); 0, −F2 (u1 )). Naively, we hope
that T is accretive with respect to the natural inner product on Y ; this is not
generation
the case. The computation to follow in Theorem 5.4 shows that

(T y, y)Y = U < ∂x u1 , N0∗ Aφ2 > +U < ∂x u2 , N0∗ Aφ1 >,

13
which has indeterminate sign, unless U = 0.

5 Semiflow Generation and Generalized Solu-

linearresult
tions - Proof of Theorem 3.1
In this section, our aim is to show that T generates a C0 semigroup on Y .

5.1 Inner Product Structure

When U = 0, one can verify that T generates a C0 semigroup on Y . However,
owing to the the loss of maximality of the operator T when U 6= 0, we adjust
the inner product structure on Y in order to obtain generation:
We introduce the following modified inner product on Y :
For y = (φ1 , φ2 ; u1 , u2 ) and y 0 = (φ01 , φ02 ; u01 , u02 )

((y, y 0 ))Y =(φ1 , φ01 )D(A1/2 ) + (φ2 , φ02 )

+ < u1 , u01 >D(A 1/2 ) + < u2 , u02 >
+ U < ∂x u1 , γ[φ01 ] > +U < ∂x u01 , γ[φ1 ] >
+ λ < u1 , u01 >H01 (Ω)
=(A1/2 φ1 , A1/2 φ01 ) + (φ2 , φ02 )
+ < ∆u1 , ∆u01 > + < u2 , u02 >
+ U < ∂x u1 , γ[φ01 ] > +U < ∂x u01 , γ[φ1 ] >
+ λ < ∇u1 , ∇u01 >

for a particular choice of λ (to be determined below). (When U = 0, this inner

product agrees with the natural inner product on Y .) To verify that ((·, ·))Y is
indeed an inner product on the Y , we need to check for positivity.
We have immediately that

||φ1 ||2D(A1/2 ) = ||A1/2 φ1 ||2L2 (R3 ) (5.1) Anorm

= ||∇φ1 ||L2 (R3 ) − U ||∂x φ1 ||2L2 (R3 ) + µ||φ1 ||2L2 (R3 ) . (5.2)

Hence for µ > 0 and U < 1,

||φ1 ||2D(A1/2 ) ≥ (1 − U )||∇φ1 ||2L2 (R3 ) . (5.3) ellipticbound

Also, we have the following:

14
 Hardy Proposition 5.1.
||γ[φ]||L2 (Ω) ≤ CΩ ||∇φ||L2 (R3+ )

for all φ ∈ H 1 (R3+ ).

evans
Proof. Beginning with the Hardy inequality [17], we see that
Z Z
|φ(x)|2 dx ≤ CR |∇φ(x)|2 dx,
QR R3+

where φ ∈ H 1 (R3+ ) and QR = {x ∈ R3+ : |x| ≤ R}. Hence, applying the trace
theorem we have
||rΩ γ[φ1 ]||L2 (Ω) ≤ CΩ ||∇φ||L2 (R3+ ) ,

where rΩ : L2 (R2 ) → L2 (Ω) is the restriction operator.

Now,

((y, y))Y = ||∇φ1 ||L2 (R3 ) − U ||∂x φ1 ||2L2 (R3 ) + µ||φ1 ||2L2 (R3 )

+ ||φ2 ||2L2 (R3 ) + ||∆u1 ||2L2 (Ω) + ||u2 ||2L2 (Ω)

Z
+ 2U (∂x u1 )(γ[φ1 ]) + λ||u1 ||2H 1 (Ω)
0
Ω

must be greater than or equal to 0.

Since Z

Eint (u1 , φ1 ) = U (∂x u1 )(φ1 z=0 )
Ω
Z
has indeterminate sign, we must check that |2U (∂x u1 )(γ[φ1 ])| is dominated
Ω
by other positive terms in the potential inner product. In particular, we have
the following:

interenergybound Proposition 5.2. There exists λ > 0 which depends on U such that
Z 
2U  (∂x u1 )(γ[φ1 ]) ≤ λ||u1 ||2H 1 (Ω) + ||φ1 ||2D(A1/2 ) . (5.4)
 
0
Ω
Hardy
Proof. We apply Cauchy-Schwarz, Young’s inequality, and Proposition 5.1 to
obtain the inequality
 Z 
2U (∂x u1 )(γ[φ1 ]) ≤ C U ||u1 ||2H 1 (Ω) + U CΩ ||∇φ1 ||2L2 (R3 )
 
0 +
Ω

1
Choosing  such that U CΩ < (1−U ), and then choosing λ such that λ > U C
2

15
 we have that
 Z  1
2U (∂x u1 )(γ[φ1 ]) ≤ λ||u1 ||2H01 (Ω) + (1 − U )||∇φ1 ||2L2 (R3+ )
 
Ω 2
ellipticbound
using (5.3) ≤ λ||u1 ||2H 1 (Ω) + ||φ1 ||2D(A1/2 )
0

Hence for an appropriate choice of λ we have verified positivity, and ((·, ·))Y
is indeed an inner product on the state space Y .
interp Remark 5.1. By Sobolev interpolation, we have for u ∈ H02 (Ω) and for all δ > 0

||u||H01 (Ω) = ||∇u||L2 (Ω) ≤ δ||∆u||L2 (Ω) +Cδ ||u||L2 (Ω) = δ||u||D(A 1/2 ) +Cδ ||u||L2 (Ω) ,
(5.5)
where Cδ is a positive constant depending on δ.

equivalenceofnorms Corollary 5.3. The topology induced by ((·, ·))Y is equivalent to the natural
topology on Y induced by (·, ·)Y .

Proof.

((y, y)) = (y, y) + 2U < ∂x u1 , γ[φ1 ] > +λ||u1 ||2H 1 (Ω)

0
interenergybound
interp
by Prop. 5.2 and 5.1 ≤ 2||φ1 ||2D(A1/2 ) + ||φ2 ||2L2 (R3 ) + (1 + Cδ )||u2 ||2L2 (Ω) + (1 + 2λδ)||u1 ||2D(A 1/2 )
≤ C(y, y)

Similarly,

((y, y)) ≥ ||φ1 ||2D(A1/2 ) + ||φ2 ||2L2 (R3 ) + ||u1 ||2D(A 1/2 ) + ||u2 ||2L2 (Ω)
− 2U | < ∂x u1 , γ[φ1 ] > | + λ||u1 ||2H 1 (Ω)
0

1
≥ (y, y)
2

Henceforth, Ye := Y((·,·)) will denote Y under the topology induced by

((·, ·))Y .

5.2 Generation of Linear Semigroup

We now verify the assumptions of the Lumer-Philips theorem for a suitable
T
translation of the overall dynamics operator T , as introduced in (4.8) on the
space Ye . This yields the generation of C0 semigroup on Ye .

16
 Remark 5.2. We will have that −(T + ωI) (ω sufficiently large) generates a C0
semigroup of contractions on Ye . Viewing −(T + ωI) as a bounded perturbation
of −T , we know that −T also generates a strongly continuous semigroup with
respect to this topology, however we may not retain the contraction property.
Then, by the equivalence of norms, we have global existence and uniqueness of
the generalized solution corresponding to T when reverting to the the natural
finite energy topology on Y . Having lost the contraction property, any stability
considerations for solutions (with respect to the finite energy topology on Y )
will require a different methodology.

generation Theorem 5.4. T + ωI : D(T ) ⊂ Y → Y is m − monotone with respect to the

inner product ((·, ·))Y .

Proof. Accretivity
 
φ1
φ 
 2
Let y =  . Consider
u1 
u2
 
−φ2 φ1
Aφ + D(φ ) + AN (u + U ∂ u ) φ 


 1 2 0 2 x 1 2 
T y + ωy, y Y =   + ω((y, y))Y
 −u2 u1 
A u1 − N0∗ A(φ2 + U ∂x φ1 ), u2 Y

Using integration by parts, zero boundary conditions, the self-adjointness of Aα ,

and making the appropriate cancellations:

=(D(φ2 ), φ2 ) + (AN0 u2 , φ2 )− < N0∗ Aφ2 , u2 >

+ U (AN0 (∂x u1 ), φ2 ) − U < ∂x u1 , γ[φ2 ] >
− U < N0∗ A(∂x φ1 ), u2 > −U < ∂x u2 , γ[φ1 ] >
− λ < ∇u2 , ∇u1 >
= − λ < ∇u2 , ∇u1 > +ω((y, y))Y ,

In the final step we have made use of the relations

(AN0 f, g) =< f, N0∗ Ag >, (D(φ2 ), φ2 ) = 2U (∂x φ2 , φ2 ) = 0;

these are also due to integration by parts, and the fact that the x-boundary
terms vanish, since φ has L2 (R3+ ) membership. Now, analyzing the final term,

17
we have

−λ < ∇u1 , ∇u2 > +ω((y, y))Y = λ < ∆u1 , u2 > +ω||∆u1 ||2L2 (Ω) + ω||u2 ||2L2 (Ω)
 Z 
+ ω ||φ1 ||2D(A1/2 ) + ||φ2 ||2L2 (R3 ) + 2U (∂x u1 )(γ[φ1 ]) + λ||u1 ||2H 1 (Ω)
0
Ω

Now, by the Holder and Young inequalities, we may choose ω > λ/2 such that

λ   
λ| < ∆u1 , u2 > | ≤ ||∆u1 ||2L2 (Ω) +||u2 ||2L2 (Ω) < ω ||∆u1 ||2L2 (Ω) +||u2 ||2L2 (Ω) .
2
Lastly, we note that the quantity
 Z 
ω ||φ1 ||2D(A1/2 ) + ||φ2 ||2L2 (R3 ) + 2U (∂x u1 )(γ[φ1 ]) + λ||u1 ||2H 1 (Ω) ≥ 0
0
Ω
interenergybound
by Proposition 5.2.
Hence, we conclude that for ω sufficiently large



T y + ωy, y Y
≥ 0,

which is to say −(T + ωI) is dissipative on Ye .

Maximality
In order to show that T is maximal, we need only show that R(T + ηI) = Y
for some η > 0. Given x = (ψ1 , ψ2 ; w1 , w2 ) ∈ Y we must solve ηy + T y = x for
y ∈ D(T ), i.e.



 ηφ1 − φ2 = ψ1 ∈ D(A1/2 )

ηφ + Aφ + D(φ ) + AN (u + U ∂ u ) = ψ ∈ L2 (R3 )

2 1 2 0 2 x 1 2 +
(5.6) rangecondition
ηu1 − u2

 = w1 ∈ D(A 1/2
)

ηu + A u − N ∗ A(φ + U ∂ φ )
 2
= w2 ∈ L (Ω).
2 1 0 2 x 1

Eliminating φ1 and u1 , we arrive at the equation

 
  1 U
φ2 ψ
 2 η − Aψ1 − AN ∂ w
0 x 1 
η
L + G)   =  , (5.7) rangeconditionreduced
1 U ∗
w2 − A w1 + N0 A∂x ψ1
 
u2
η η

18
where  
1 U
A+D AN0 (I + ∂x )
 η η
L=  (5.8) L
U 1
−N0∗ A(I + ∂x ) A + ηI
 
η η
and    
φ2 ηφ2
G  =  . (5.9) G
u2 0
We proceed by first showing that

L : D(A1/2 ) × D(A 1/2 ) → [D(A1/2 )]0 × [D(A 1/2 )]0

is coercive; the duality is taken with respect to the pivot space L2 (R3+ ) × L2 (Ω).
! !
 φ φ  1 1
L , = ||A1/2 φ||2L2 (R3 ) + ||A 1/2 u||2L2 (Ω) + η||u||2L2 (Ω)
u u 2 3 2
L (R+ )×L (Ω) η + η
2U
+ < N0∗ Aφ, ∂x u >
η
1 1
= ||A1/2 φ||2L2 (R3 ) + ||A 1/2 u||2L2 (Ω) + η||u||2L2 (Ω)
η + η
Z
2U


+ ∂x u γ[φ]
η Ω
interenergybound interp
using Proposition 5.2 and Remark 5.1
1 1
≥ ||A1/2 φ||2L2 (R3 ) + (1 − λδ)||A 1/2 u||2L2 (Ω)
2η + η
+ (η − Cδ )||u||2L2 (Ω) .

The above inequality holds for all δ > 0; hence we may choose δ such that
1 − λδ > 0. Following that selection, we may take η suitably large, such that
η − Cδ > 0. Thus we have for appropriate δ and η
! !
 φ φ 
≥ C ||A1/2 φ||2L2 (R3 ) + ||A 1/2 u||2L2 (Ω) ,


L ,
u u 2 3 2
L (R+ )×L (Ω) +

giving that L is coercive from

D(A1/2 ) × D(A 1/2 ) → [D(A1/2 )]0 × [D(A 1/2 )]0 .

Continuity of L in this setting is straightforward. Now, G is trivially monotone

and hemicontinuous, and hence m-monotone. Moreover, this gives immediately

19
that L + G is coercive.
Summarizing, we have that G is m-monotone, L is continuous and coercive,
springer
and L + G is coercive, so by a corollary to Minty’s theorem [10]

L + G : D(A1/2 ) × D(A 1/2 ) → [D(A1/2 )]0 × [D(A 1/2 )]0

has full range.

To finish the range condition, we must check that, in fact
 
1 U
ψ
 2 η− Aψ 1 − AN ∂ w
0 x 1 
 η  ∈ [D(A1/2 )]0 × [D(A 1/2 )]0 .
1 U ∗
w2 − A w1 + N0 A∂x ψ1
 
η η

Firstly, ψ2 ∈ L2 (R3+ ) ⊂ [D(A1/2 )]0 , and since ψ1 ∈ D(A1/2 ), we have that Aψ1 ∈
[D(A1/2 )]0 . Since w1 ∈ D(A 1/2 ) = H02 (Ω), we have that ∂x w1 ∈ H 1 (Ω) ⊂
contmap
L2 (Ω). Hence for ξ ∈ D(A1/2 ), recalling (4.2), we have that

(AN0 ∂x w1 , ξ)L2 (R3+ ) = (A3/4− N0 ∂x w1 , A1/4+ ξ)

1 U
is well-defined. Hence, ψ2 − Aψ1 − AN0 ∂x w1 ∈ [D(A1/2 )]0 .
η η
Now, arguing as above, we have w2 , A 1/2 w1 ∈ [D(A 1/2 )]0 . Moreover, for
v ∈ D(A1/2 ) = H 2 (Ω) ∩ H01 (Ω), we have

< N0∗ A∂x ψ1 , v >L2 (Ω) = − < N0∗ Aψ1 , ∂x v >= − < γ[ψ1 ], ∂x v >,

by commuting the ∂x operator and making use of integration by parts with

vanishing boundary conditions. Since ψ1 ∈ D(A1/2 ) = H 1 (R3+ ), the trace
γ[ψ1 ] ∈ H 1/2 (R3+ ) is well-defined; since v ∈ H02 (Ω) this inner-product is defined,
1 U
and thus w2 − A w1 + ∂x γ[ψ1 ] ∈ [D(A 1/2 )]0 .
η η rangeconditionreduced
Hence we have found a solver φ2 ∈ D(A1/2 ) and u2 ∈ D(A 1/2 ) of (5.7).
rangecondition
Recovering φ1 and u1 from (5.6), we see that

(φ1 , φ2 ; u1 , u2 ) ∈ D(A1/2 ) × D(A1/2 ) × D(A 1/2 ) × D(A 1/2 ).

rangecondition
Moreover, since y = (φ1 , φ2 ; u1 , u2 ) is indeed a solver to (5.6), it is immediate
that y ∈ D(L + G) = D(T ).
Hence we have shown R(T + ηI) = Y , and the operator T + κI for κ =
max{ω, η} is m-accretive.

Now, clearly −(T + ωI) is a bounded perturbation of (−T ) on Ye . Hence by

20
 Pazy
[23] we know that (−T ) generates a C0 semigroup (not necessarily contraction)
on Ye . By the equivalence of norms induced by ((·, ·))Y and (·, ·)Y , we know that
(−T ) still generates a C0 semigroup on Y under the topology induced by (·, ·)Y .
platesystem
flowsystem
Thus, for y0 ∈ D(T ) we have a strong solution to (3.1)-(3.3). For y0 ∈ Y , we
obtain through a strong limit of strong solutions a solution to the corresponding
integral equations. Hence we have the following as a corollary:

linearresult1 Theorem 5.5. The operator (−T ) generates a C0 semigroup on Y taken with
compatibilitycondition
the natural, finite energy topology. Hence, if the compatibility condition (3.14)
holds, then

1. If (φ0 , φ1 ; u0 , u1 ) ∈ D(T ), we have the existence of a unique strong solution

platesystem
flowsystem
to (3.1)-(3.3) in the case where F ≡ 0;

2. If (φ0 , φ1 ; u0 , u1 ) ∈ Y , then we have the existence of a unique generalized

platesystem
flowsystem
solution to (3.1)-(3.3) in the case where F ≡ 0.

5.3 Case: α > 0

alphage0
The purpose of this short section is to show how our setup accomodates the
case α > 0. In this case, the plate equation becomes

Mα utt + ∆2 u + F (u) = p(x, t),

with
Mα := (I − α∆Dirichlet ), D(Mα ) ≡ H 2 (Ω) ∩ H01 (Ω).
abstractsystem
Abstractly, we replace the plate equation in (4.7) with

Mα utt + A u − νN0∗ A(φt + U φx ) + F2 (u) = 0.

The state space in this case changes only in the plate component, which becomes

Y2 = D(A 1/2 ) × D(Mα ) ≡ (H 2 (Ω) ∩ H01 (Ω)) × H01 (Ω).

Having made these adjustments, one may easily check that the linear ar-
gument above for α = 0 goes through. Previous well-posedness results (as
past
mentioned in Section 3.1) did not produce a semigroup, and were unable to
obtain in the α = 0 case. In addition, the nonlinear results to follow do apply
results
to the α > 0 case; that is, the results in Section 3.4 can be taken, mutatis
mutandis, with α ≥ 0.

21
 5.4 Nonlinear Perturbation
We now cite two theorems that generalize our result on linear generation to
well-posedness in the nonlinear case, when F is locally Lipschitz on Y .

liptheorem Theorem 5.6 (Locally Lipschitz Perturbation of an M-Monotone Operator). Let

A : D(A) ⊂ H → H be a maximally monotone operator a Hilbert space H. Let
J : H → H be locally Lipschitz on H, i.e. ||Ju − Jv||H ≤ L(K)||u − v||H pro-
vided ||u||H , ||v||H ≤ K. Then for u0 ∈ D(A) and f ∈ W 1,1 (0, t; H) for all t,
there there exists a tmax ≤ ∞ such that

ut + Au + Ju = g, u(0) = u0 ∈ H (5.10) lipperturb

has a unique strong solution u on [0, tmax ). If we only assume u0 ∈ D(A) = H

lipperturb
and g ∈ L1 (0, t ; H), then there exists a tmax such that (5.10) has a unique
generalized solution u ∈ C(0, tmax ; H).
Moreover, in both cases if tmax < ∞, then ||u(t)||H → ∞ as t → tmax .
attractor
Remark 5.3. A nice, concise proof of this theorem can be found in [8].
Applying this theorem to our setup, we have

Corollary 5.7. The equation

y 0 + T y = F (y), y(0) = y0 ∈ Y
T
with F (y) = (0, −F1 (φ1 ); 0, −F2 (u1 )) and T as in (4.8) has a unique local-in-
time strong solution y(t) if F1 is locally Lipschitz on H 1 (R3+ ) and F2 locally
Lipschitz on H 2 (Ω) ∩ H01 (Ω).
When y(0) ∈ Y , we have a unique local-in-time generalized solution ye(t) to
abstractode
(5.7).
In both cases, when tmax (y0 ) < ∞, we have that ||y(t)||Y , ||e
y (t)||Y → ∞ as
t % tmax (y0 ).

Lastly, we show that the generalized solutions provided by the above abstract
solutions
arguments are in fact weak solutions, as defined in Section 3.3.

gentoweak Lemma 5.8. For F locally Lipschitz on D(A 1/2 ) = H 2 (Ω) ∩ H01 (Ω), any gen-
platesystem
flowsystem
eralized solution to (3.1)-(3.3) is also a weak solution, as defined in Section
solutions
3.3

Proof. The variational form of solution given as the definition of a weak solu-
platesystem
flowsystem
tion to (3.1)-(3.3) is verified for strong solutions by straightforward integration.
Using the (strong) limit definition of generalized solutions, passing the limit in
the linear terms is tautological. We need only consider passing the limit in the

22
nonlinear term. Since F is locally Lipschitz continuous, it is weakly continuous
and thus we can easily pass the limit; hence, the limiting procedure shows that
platesystem
flowsystem
the generalized solution satisfies the weak formulation of (3.1)-(3.3). This weak
solution is in fact unique - for the proof, we defer to the method presented in
springer
[10].

6 Nonlinear Problems - Examples

In modeling the flow-structure system, we initially assumed no forcing term in
the flow equation (i.e. F1 ≡ 0). However, describing the system abstractly,
we generalized by adding a forcing term in the flow equation F1 (φ) and took
F (y) = (0, −F1 (φ1 ); 0, −F2 (u1 )); in full generality, we may take take

F : Y → (L2 (R3+ ))2 × (L2 (Ω))2

to be any locally Lipschitz operator on Y , and we will retain local-in-time solu-

liptheorem
tions, according to Theorem 5.6.
We now included specific examples of pertinent choices of the nonlinearity
platesystem
F in (3.1). For each of the following, we take F1 (φ) = −µφ.

1. Berger Plate. In this case we take

Z
F (u) = ∆u |∇u|2 , (6.1) bergernon
Ω

F2 (u) = F (u) − p0 (6.2)

and Z
1
Π(u) = |∇u|4 . (6.3) bergerpot
2 Ω

We have that F2 is in fact locally Lipschitz on the plate component of Y ,

and we also have an abstract a priori bound of the form

Π(u) ≤ C(||u||H 2 (Ω) ),

springer
[10].
vonK
2. Von Karman Plate. This case was introduced in Remark 3.1. Here F
is cubic-nonlocal. This case will be discussed in detail below.

3. Semilinear Polynomial. We note that any nonlinearity of the form

F (u) = g(∇u) (6.4) semipoly

23
 where g is a polynomial will be locally Lipschitz on the plate component,
and hence falls under the umbrella of our results up to this point. Note
that we may not necessarily have an a priori bound (of any kind) on the
solution when the polynomials are superlinear.

6.1 Von Karman Nonlinearity

We now specialize the nonlinearity to the von Karman nonlinearity, i.e. for the
remainder of this section take

F1 (φ) = −µφ and F2 (u) = F (u) − p0 = Π0 (u) = −[v(u), u] − p0 ,

with
1
Π(u) = ||∆v(u)||2L2 (Ω) − (p0 , u)L2 (Ω) ,
4
Airy
v denoting Airy’s stress function as in (3.4), and [·, ·] von Karman bracket as
bracket
in (3.7). We assert that the term F (y) = (0, −F1 (φ1 ); 0, −F2 (u1 )) is locally
Lipschitz on Y ; this statement is equivalent to the fact that F (u) = −[v(u), u]
is locally Lipschitz on H 2 (Ω)∩H01 (Ω), which follows from the “sharp” regularity
springer
of the Airy function. We cite the result from [10], p.44, Corollary 1.4.5:

sharpreg Theorem 6.1 (Sharp Regularity of the Airy Function). Let f ∈ H 2 (Ω). Then
for any u ∈ H 2 (Ω), the problem

∂v 
∆2 u + [u + 2f, u] = 0, v ∂Ω =

 = 0, (6.5) airy3
∂n ∂Ω
2+2/p
has a unique solution v = v(u) in H02 (Ω) ∩ Wp (Ω), for all 1 ≤ p ≤ ∞.
This solution (often referred to as the Airy stress function with respect to the
displacement u) has the following properties:

||v(u1 )−v(u2 )||W 2+2/p (Ω) ≤ Cp ||u1 +u2 +2f ||H 2 (Ω) ||u1 −u2 ||H 2 (Ω) for all 1 ≤ p ≤ ∞;
p
(6.6) airyest1
and

||v(u1 )−v(u2 )||H 3−δ (Ω) ≤ Cδ ||u1 +u2 +2f ||H 2 (Ω) ||u1 −u2 ||H 2−δ (Ω) for all 0 ≤ δ ≤ 1.
(6.7) airyest2
Lastly, we also have the estimate

||[v(u1 ), u1 ]−[v(u2 ), u2 ]||H −δ (Ω) ≤ Cδ ||u||2H 2 (Ω) +||u2 ||2H 2 (Ω) +||f ||2H 2 (Ω) ||u1 −u2 ||H 2−δ

(6.8) airyest3
for all ≤ δ ≤ 1.

24
 abstractsystem
Our final task is to provide an a priori bound on the solution to (4.7); we
liptheorem
then simply apply the second part of Theorem 5.6 to arrive at global existence
and uniqueness for the fully nonlinear fluid-structure system with von Karman
nonlinearity. We begin with a lemma:

energybounds Lemma 6.2 (Energy Bounds). There exists constants c, C and M (p0 ) such that

cE(t) − M ≤ E (t) ≤ CE(t) + M, (6.9) energyrelations

energiessection
with E (t) and E(t) as in Section 3.2.

This lemma is then used in conjunction with the energy inequality E (t) ≤
springer
E (s), ∀ t ≥ s to obtain the desired result [10]:

aprioribounds Theorem 6.3 (A priori Bounds). Any strong, weak, or generalized solution (u(t); φ(t))
platesystem
flowsystem
to (3.1)-(3.3) for t ≥ 0 must sastisfy
n o
sup ||ut (t)||2L2 (Ω) + ||∆u(t)||2L2 (Ω) + ||φt (t)||2L2 (R3 + ||∇φ(t)||2L2 (R3 < +∞
+ +
t≥0
(6.10) aprioribound

Hence, we conclude with a statement of the final result for the flow-structure
interaction with von Karman nonlinearity:

Theorem 6.4 (von Karman Well-Posedness). If F (u) = −[v(u), u] as defined

Airy airy2 linearresult
in (3.4)-(3.6), then under the strong solutions assumptions of Theorem 3.1,
platesystem
flowsystem
for all T , (3.1)-(3.3) has a unique strong solution on [0, T ]; under the gener-
platesystem
flowsystem
alized and weak solutions assumptions, for all T , (3.1)-(3.3) has a unique
generalized (and hence weak) solution on [0, T ].

25
 References
bal [1] A. V. Balakrishnan, A continuum aeroelastic model for inviscid subsonic
wing flutter, J. of Aerospace Engineering, June 2007.

bal1 [2] A.V. Balakrishnan, A Mathematical theory of flutter instability phenomena

in aeroelasticity, Inst. for Systems Dynamics and Control Theory, SB RAS.
Irkutsk, Russia, 2007.

bal2 [3] A. V. Balakrishnan, Nonlinear Aeroelasticity: continuum theory, flutter,

divergence speed and plate wing model, J. of Aerospace Engineering, vol
19, pp. 194-202, 2006.

berger [4] M.S. Berger, On von Karman’s equations and the buckling of a thin elastic
plate, I. the clamped plate, Commun. Pure Appl. Math., 20, pp. 687-719,
1967.

bolotin [5] V.V. Bolotin, Nonconservative problems of elastic stability. Pergamon

Press, Oxford, 1963.

b-c [6] A. Boutet de Monvel and I. Chueshov. The problem of interaction of von
Karman plate with subsonic flow gas, Math. met. Appl. Sc. 22, pp. 801-810,
1999.

b-c-1 [7] L. Boutet de Monvel and I. Chueshov, Nonlinear oscillations of a plate in

a flow of gas, C. R. Acad. Sci, Paris, 322, 1001-1006, 1996.

attractor [8] I. Chueshov, M. Eller, and I. Lasiecka, On the attractor for a semilinear
wave equation with critical exponent and nonlinear boundary dissipation,
Communications in Partial Differential Equations, Vol. 27, pp. 1945-1948,
2002.

ch-l [9] I. Chueshov and I. Lasiecka , Long time behavior of second order evolutions
with nonlinear damping, Memoires of AMS, vol 195, 2008.

springer [10] I. Chueshov and I. Lasiecka, Von Karman Evolution Equations, Springer
Verlag, pp. 22, 44, 259, 2010.

ciarlet [11] P. Ciarlet and P. Rabier, Les Equations de Von Karman, Springer Verlag,
1980.

cole [12] J. Cole and L. Cook. Transonic Aerodynamics, North Holland, 1986.

dowell [13] E. Dowell, Aeroelasticity of Paltes and Shells , Nordhoff, Leyden, 1975.

26
 dowell1 [14] E. Dowell, A Modern Course in Aeroelasticity, Kluwer Academic Publish-
ers, 2004.

dowellnon [15] E. Dowell, Nonlinear Oscillations of a Fluttering Plate, AIAA J., vol. 4,
no. 7, 1966.

dowellnon1 [16] E. Dowell, Nonlinear Oscillations of a Fluttering Plate. II, AIAA J., vol.
5, no. 10, 1967.

evans [17] L. Evans, Partial Differential Equations, 2nd Ed., Graduate Studies in
Mathematics, vol. 19, AMS, 2010.

cbms [18] I. Lasiecka, Mathematical Control Theory of Coupled PDEs, CMBS-NSF

Lecture Notes, SIAM, 2002.

lions [19] I. Lasiecka, J.L. Lions, and R. Triggiani, Nonhomogenuous boundary value
problems for second order hyperbolic operators, J. Math. Pure et Appliques,
vol. 65, pp. 149-192, 1986.

redbook [20] I. Lasiecka and R. Triggiani, Control Theory for Partial Differential Equa-
tions,vol. I, II, Cambridge University Press, 2000.

miyatake [21] S. Miyake, Mixed problems for hyperbolic equations, J. Math. Kyoto Univ.,
13 (1973), pp. 435-487.

patil [22] M. Patil and D. Hodges, Flight dynamics of highly flexible flying wings, J.
of Aircraft , vol. 43, pp. 1790-1799, 2006.

Pazy [23] A. Pazy, Semigroups of linear operators and applications to PDE, Springer,
New York, p 76, 1986.

ryz [24] I. Ryzhkova , Stabilization of von Karman plate in the presence of thermal
effects in a subsonic potential flow of gas, J. Math. Anal. Apppl, v. 294, pp.
462-481, 2004.

ryz1 [25] I. Ryzhkova, On a retarder PDE system for a von Karman plate with
thermal effects in the flow of gas, matem. Fizika, Analiz, Geometry vol. 12
(2) pp. 173-186, 2005.

ryz2 [26] I. Ryzhkova , Dynamics of a thermoelastic von Karman plate in a subsonic

gas flow, Z. Angew. math. Phys, vol. 58, pp. 236-261, 2007.

temam [27] R. Temam, Infinite Dimensional Dynamical Systems in Mechanics and

Physics, Springer Verlag, 1988.

27
amjad [28] A. Tuffaha, Flutter stability analysis of a wedge shaped airfoilwith nonzero
thickness in non-viscuous airflow. Nonliear Studies, vol. 16, pp. 255-150,
2009.

28