Beruflich Dokumente
Kultur Dokumente
O
3289 (2005)
COLE POLYTECHNIQUE FDRALE DE LAUSANNE
PRSENTE LA FACULT SCIENCES DE BASE
Institut d'analyse et de calcul scientique
SECTION DE MATHMATIQUES
POUR L'OBTENTION DU GRADE DE DOCTEUR S SCIENCES
PAR
DEA en modlisation mathmatique, Universit Libanaise, Liban
et de nationalit libanaise
accepte sur proposition du jury:
Lausanne, EPFL
2005
Hicham Georges GEBRAN
Prof. C.A. Stuart, directeur de thse
Prof. P. Fitzpatrick, rapporteur
Prof. M. Furi, rapporteur
Prof. T. Ratiu, rapporteur
Dr H.-J. Ruppen, rapporteur
QUASILINEAR SECOND ORDER ELLIPTIC SYSTEMS
AND TOPOLOGICAL DEGREE
2
Version abregee
Dans ce travail, nous considerons une grande classe de syst`emes elliptiques quasilineaires du
2`eme ordre de la forme
,=1
a
EG
EE
Abstract
We consider a large class of quasilinear second order elliptic systems of the form
,=1
a
,=1
a
(x, u, u)
2
u +b(x, u, u, ) = 0, x IR
N
lim
|x|
u(x) = 0.
Here again, they work in the Sobolev spaces X = W
2,p
(IR
N
) and Y = L
p
(IR
N
) but with
p (N, ), and this incorporates the decay to zero of the solutions and their derivatives at
innity. The problem is formulated as the search of zeros of the operator F : IR X Y
dened by
F(, u) =
N
,=1
a
(., u, u)
2
u +b(., u, u, ).
Then, the main goal is to nd ways for verifying the conditions required for the use of the
degree, and which are precisely
(a) F C
1
(IR X, Y ).
(b) D
u
F(, u)
0
(X, Y ) for all u X.
(c) F : IR X Y is proper on the closed bounded subsets of IR X.
They concentrate on the situation where the problem is asymptotically periodic as [x[ , in
the sense that there are functions a
and b
such that a
(., ) and b
(x, ) a
(x, )
= 0 and lim
|x|
[
i
b(x, , )
i
b
(x, , )[ = 0.
Setting
F
(, u) =
N
,=1
a
(., u, u)
2
u +b
(., u, u, ),
the main results of [31], show that under mild smoothness assumptions which imply (a), the
important properties (b) and (c) are equivalent to
(i) D
u
F(, 0)
0
(X, Y ).
(ii) u X, F
(, u) = 0 = 0.
1.2. DESCRIPTION OF THE PRESENT REPORT 15
Then, they explore some important situations where condition (i) can be explicitly veried,
essentially by using the spectral theory of Schrodinger operators.
To verify condition (ii) they consider three situations and accordingly give three approaches.
The rst is based on the maximum principle. The second is based on integral identities of
Pohozaev-type, and this was carried out in [33] after a study of the exponential decay of
solutions. The third approach deals with the case when F
(, u) is linear in u, so spectral
theory is used. Each situation then leads to a global bifurcation theorem.
A more detailed survey of the works of Rabier and Stuart can be found in [34].
Independently of Fitzpatrick, Pejsachowicz and Rabier, Benevieri and Furi constructed an-
other degree for C
1
Fredholm maps of index zero. We introduce these tools in 1.3.2.
1.2 Description of the present report
Many problems in science can be described mathematically as a system of several dierential
equations of several unknown functions (consider for instance Navier-Stokes equations, the
equations of 3 dimensional elasticity, reaction-diusion equations in mathematical ecology and
so on). These equations are sometimes called coupled equations. Suppose that we are given
m quasilinear partial dierential equations of second order containing m unknown functions
u
1
, u
2
, . . . , u
m
dened on a domain IR
N
, then we can write the system in the compact form
,=1
a
(x, u, u)
2
u +b(x, u, u) = 0, x = (x
1
, . . . , x
N
) , (1.1)
where
u is the vector
_
u
1
, u
2
, . . . , u
m
_
T
, u is the matrix of partial derivatives
x
j
u
i
, j =
1, . . . , N, i = 1, . . . , m.
a
(x, u, u) is an mm matrix.
b(x, u, u) is a vector of height m.
In the physical examples mentioned above, these equations correspond to the stationary case,
and x represents the independent space variables.
The rst objective of this research is to extend the approach of Rabier and Stuart [31] to
cover such systems on unbounded domains of IR
N
, and which are also elliptic in some sense.
The rst diculty is the choice of an ellipticity condition. For a single linear second order
equation, there is essentially one such condition which states that the matrix of the leading
coecients is positive denite. Indeed, dierent renements such as strong or uniform ellipticity
capture dierent behavior of the coecients when they are not constant. However, in the case
of a system, even with constant coecients, there are at least three dierent conditions of
ellipticity: that of Agmon-Douglis-Nirenberg, the strong Legendre-Hadamard, and the strong
Legendre conditions. After an exploratory investigation of these dierent conditions and some of
their functional implications, we decided to concentrate on an intermediate condition between
Agmon-Douglis-Nirenberg and the strong Legendre-Hadamard. This condition is known as
ellipticity in the sense of Petrovskii, and for which L
p
a priori estimates due to Koshelev
are available under weak assumptions about the smoothness of the coecients. Furthermore,
Petrovskii ellipticity enables us to use the same function space for all the components of the
vector u introduced above. The interest in ellipticity stems from its strong connections with the
Fredholm property of linear partial dierential operators. For problems on bounded domains,
16 CHAPTER 1. INTRODUCTION
an important connection is given by the theorems on complete collections of isomorphisms,
which establish an equivalence between (i) ellipticity (in the sense of Agmon-Douglis-Nirenberg)
together with a boundary condition (known as Lopatinskii condition), (ii) a priori estimates,
and (iii) the Fredholm property, in a scale of function spaces. See Rotberg [36], Agranovich
[2] or Wloka and al. [43] for precise statements. We mention, however, that these theorems
are stated and proved in the literature under the strong assumptions that the domain has a C
,=1
a
(., u, u)
2
u +b(., u, u) (1.2)
where b :
_
IR
m
IR
mN
_
IR
m
, a
:
_
IR
m
IR
mN
_
IR
mm
(, = 1, . . . , N) is
a family of matrix-valued maps, and is an open subset of IR
N
, whose boundary is Lipschitz
continuous and bounded.
By requiring that has a bounded boundary, we cover three situations: is bounded (in
this case, we claim to no novelty), is the exterior of a bounded set, and = IR
N
.
The goal is then the investigation of the Fredholm and properness properties of F in (1.2)
and the interplay between them, in the context of Sobolev spaces. More precisely, we study
these functional properties of the operator F acting between W
2,p
(, IR
m
) W
1,p
0
(, IR
m
) and
L
p
(, IR
m
), where p is a real number > N.
Indeed, we formulate conditions on the coecients of F, and the domain which ensure
that F is well dened i.e. maps W
2,p
(, IR
m
) into L
p
(, IR
m
), and is of class C
1
. This is
done in 2.1, after a preparatory study of some Nemytskii operators. The needed notation are
introduced and xed on pp. 25-26. The conditions on the coecients are formulated on p. 33
and they are implicitly assumed in all the relevant theorems of chapter 2.
In 2.2 we introduce two ellipticity conditions for linear systems. The study of the Fredholm
property of F begins in 2.3. The main result there is Theorem 2.3 which states that, if the
coecients of F satisfy an ellipticity condition (in the sense of Petrovskii), and the Frechet
derivative DF(u) is semi-Fredholm (with index ,= ) for some u W
2,p
(, IR
m
)W
1,p
0
(, IR
m
),
then DF(u) is semi-Fredholm of the same index for every u. Therefore, to prove the Fredholm
property, one should only consider the linearization at a particular point. We end 2.3 by a
result on the factorization of F.
The study of properness begins in 2.4, where we relate - through some technical results -
properness on closed bounded subsets, to the Fredholm property, and to a notion of uniform
decay of sequences of functions. We end that section by proving that if F is proper on the
closed bounded subsets of W
2,p
(, IR
m
) W
1,p
0
(, IR
m
), then F is semi-Fredholm of index ,=
(Corollary 2.2).
In 2.5, we consider the situation where the coecients of F are asymptotically periodic as
[x[ , in the sense that there are functions a
and c
i
1 , N, 0 i N such that
1.2. DESCRIPTION OF THE PRESENT REPORT 17
a
(., ) and c
i
(., ) are periodic on IR
N
with the same period for all IR
m
IR
mN
and
lim
|x|
(x, ) a
(x, )
= 0 and lim
|x|
_
1
0
i
b(x, t) dt c
i
(x, )
= 0.
The periodicity condition forces the limit coecients a
(., ) and c
i
(., ) to be dened on the
whole space IR
N
even if the coecients a
and c
i
are only dened on ,= IR
N
(which is
indeed required to be unbounded).
Setting b
(x, ) =
N
i=0
c
i
(x, )
i
, we dene a limit operator
F
(u) =
N
,=1
a
(., u, u)
2
u +b
(., u, u).
Then, after a series of technical lemmas, we succeed in nding sucient conditions for the
properness (on closed bounded subsets) of the operator F in (1.2). Namely, we prove that if (i)
F is semi-Fredholm of index ,= , and (ii) the limit problem F
(X, Y ) the set of semi-Fredholm operators of index . In this work, the case = + is
not of particular interest, and we set
+
(X, Y ) =
_
Z{}
(X, Y ).
When G C
1
(X, Y ) is not necessarily linear, it is semi-Fredholm if for every u X, the Frechet
derivative DG(u) L(X, Y ) is semi-Fredholm. A semi-Fredholm operator with nite index is
called Fredholm. We recall two fundamental properties of semi-Fredholm operators. See for
instance [23] pp. 78-79.
(a) The index of semi-Fredholm operators is a locally constant function, or equivalently
(X, Y )
is open in L(X, Y ).
(b) If L
uF
1
(y)B
(DF
u
).
When F
1
(y) B is empty, the degree is naturally dened to be zero.
Now, if y / F(B) is not necessarily a regular value of F, there exists a ball U centered at y
and contained in Y F(B), and as a consequence of the Quinn-Sard theorem, this ball contains
regular values of F. But then, it is shown that deg
p
(F, B, y
1
) = deg
p
(F, B, y
2
) for any regular
values y
1
, y
2
U. Accordingly, deg
p
(F, B, y) is dened to be this common number.
A change of the base point from p to q, changes the degree by the factor (DF ) 1, 1,
where is any curve joining p to q.
This degree satises the usual properties of existence (called there normalization property),
excision, additivity on domains, while its behavior under homotopy is given by the following
reformulation of Theorem 5.1 and Corollary 5.5 of [29].
Theorem 1.1 Let h C
1
([0, 1] O, Y ) be Fredholm of index 1 (this is equivalent to saying
that D
u
h(t, .)
0
(X, Y ) for all t [0, 1]), and proper on [0, 1] B. Such a homotopy is called
Badmissible. Suppose that y / h([0, 1] B) and p
0
O is a base point of h(0, .). Then we
have the following.
(i) If p
1
O is a base point for h(1, .), then
deg
p
0
(h(0, .), B, y) = (D
u
h ) deg
p
1
(h(1, .), B, y),
where is any curve joining (0, p
0
) to (1, p
1
) in [0, 1] O.
(ii) If for some t (0, 1], h(t, .) has no base point or the equation h(t, u) = y has no solution
in B, then deg
p
0
(h(0, .), B, y) = 0.
It is sometimes useful to consider the absolute degree of (F, B, y), which is dened to be zero
if F has no base point, and to be the absolute value of deg
p
(F, B, y) whenever p is a base
point. Indeed, this is not a degree because it is not additive, however, it satises the excision
property and it is homotopy invariant. And as noticed on p. 24 in [13], these two properties
are sucient to prove the generalized homotopy invariance as in the Leray-Schauder degree.
For future reference, we state
Lemma 1.2 Let J be an open interval of the real line, J O, G : Y be C
1
Fredholm of index 1, and extendable by continuity to a proper map on the closure of . Set
for t J,
t
= u O, (t, u) . Suppose that y ,= G(t, u) for all u (
t
). Then, for any
a and b from J,
[ deg [(G(a, .),
a
, y) = [ deg [(G(b, .),
b
, y).
Finally, we mention the connection with the Leray-Schauders degree. When X = Y , B is
bounded and F is a compact perturbation of the identity, we have
deg
p
(F, B, y) = (1)
n
deg
L.S
(F, B, y),
where n is the number of negative eigenvalues of DF(p).
22 CHAPTER 1. INTRODUCTION
The degree of Benevieri-Furi
This degree is based upon an algebraic concept of orientation for linear Fredholm maps of index
zero. Let L : X Y be a Fredholm operator of index zero between two vector spaces (no
additional structure is needed at this stage). Then, there exists a linear operator A : X Y
with nite rank, such that L + A is an algebraic isomorphism
3
. An operator A having this
property is called a corrector of L and the set of correctors of L is denoted by C(L). On C(L)
is dened an equivalence relation in the following way. Given A, B C(L), the automorphism
of X:
T = (L +B)
1
(L +A) = I (L +B)
1
(B A)
is such that I T has nite rank. Hence, given any nontrivial nite dimensional subspace
X
0
containing rge (I T), the restriction of T to X
0
is an automorphism of X
0
. Therefore, its
determinant is well dened and nonzero, and furthermore it can be shown that it is independent
of the choice of X
0
. Accordingly, A is said to be Lequivalent to B if det
_
(L+B)
1
(L+A)
_
> 0.
This is indeed an equivalence relation on C(L) with just two equivalence classes. An orientation
of L is then one among these two classes, and the elements of the chosen class are called the
positive correctors of L.
When L is an isomorphism, the trivial operator 0 is indeed a corrector of L and therefore
belongs to a class in C(L), this class is called the natural orientation of L . However, if L is
already oriented, the sign of L is dened to be 1 if the trivial operator belongs to the orientation
of L and 1 otherwise.
Now let X and Y be Banach spaces. In this context, it is natural to consider only bounded
correctors of bounded Fredholm operators of index zero. Because GL(X, Y ) is open in L(X, Y ),
it follows that if A is a corrector of L
0
(X, Y ), then it is still a corrector of all operators L
) for all
in a neighborhood of .
A map is then called orientable if it admits an orientation and oriented when an orientation is
chosen.
In particular, let O be an open subset of X, and F : O Y be a C
1
Fredholm operator of
index zero. An orientation of F is an orientation (if it exists) of the continuous map DF : O
0
(X, Y ).
With this in mind, we may proceed to the denition of the degree for regular triples. Given
an element y Y , an open subset B O, and an oriented map F : O Y , the triple (F, B, y)
is called admissible if F
1
(y) B is compact
4
. If y is a regular value of F then F
1
(y) B is
nite and the degree of (F, B, y) is
deg
B.F
(F, B, y) =
uF
1
(y)B
signDF(u),
where, as in the general case, signDF(u) = 1 if the trivial operator is a positive corrector of
the oriented isomorphism DF(u), and signDF(u) = 1 otherwise.
The assumption that y is a regular value is then removed by means of the classical Sards
theorem, after showing that given two neighborhoods U
1
and U
2
of F
1
(y), one has
deg
B.F
(F, U
1
, y
2
) = deg
B.F
(F, U
2
, y
2
), for any regular values y
1
, y
2
suciently close to y.
3
Let A
1
be an isomorphism between ker L and some complement of rge L, and let P be a projection from X
onto ker L. Then, one check that L +A
1
P is an isomorphism. If in addition X and Y are Banach space and L
is bounded, then by taking P continuous, we see that L +A
1
P is in fact a topological isomorphism.
4
It is called strongly admissible if F has a proper extension to the closure of B and y / F(B).
1.3. FUNDAMENTAL CONCEPTS AND NOTATION 23
As expected, this degree satises the usual properties of existence, additivity and invariance
under oriented homotopies. More precisely,
Denition 1.1 A homotopy of Fredholm maps of index zero from O X to Y is a continuous
map h : [0, 1] O Y , which is dierentiable with respect to the second variable and such
that, for any (t, u) [0, 1] O, the partial derivative D
u
h(t, u)
0
(X, Y ), and furthermore
the map D
u
h : [0, 1] O
0
(X, Y ) is continuous. Then, an orientation of h is an orientation
of the continuous map D
u
h : [0, 1] O
0
(X, Y ).
The invariance of the degree under oriented homotopies is stated in the following form.
Let h : [0, 1] O Y be an oriented homotopy and y : [0, 1] Y be a path. If the set
(t, u) [0, 1] O [ h(t, u) = y(t) is compact, then deg
B.F
(h(t, .), O, y(t)) is well dened and
does not depend on t [0, 1].
Note that the requirements on the homotopy are weaker than those in the Fitzpatrick-
Pejsachowicz-Rabier theory, where the homotopy is required to be C
1
in both variables, and
this is needed there in order to prove the homotopy variance of the degree at regular values via
an approximation theorem -Theorem 2.1- (see the proof of Theorem 4.1 in [29]).
Now the following question arises: how can we check that a Fredholm map or a homotopy
of Fredholm maps is orientable? Benevieri and Furi give the following answer (which is a
consequence of Theorem 3.11 in [4]). Any continuous map h :
0
(X, Y ) is orientable
provided that is simply connected and locally path connected. In particular the map DF :
O
0
(X, Y ) is orientable if O is simply connected (since O, being an open subset of a Banach
space is locally path connected).
For the orientation of a homotopy, another answer is given by Theorem 4.3 in [4] which
ensures that a homotopy h : [0, 1] O Y , is orientable if and only if for some t
0
[0, 1],
h(t
0
, .) : O Y is orientable. In this case, an orientation of h(t
0
, .) is the restriction of a unique
orientation of h.
The two degree theories that we have just sketched are strongly connected to each other,
and an instructive comparison can already be found in [4], 5. We mention, in particular, the
following (Proposition 5.6 in [4]). For an oriented path : [0, 1]
0
(X, Y ) with invertible
endpoints (0) and (1) we have
() = sign(0) sign(1).
As a consequence, when (F, B, y) is admissible and regular in both theories, and p is a base
point for F,
deg
p
(F, B, y) =
uF
1
(y)B
(DF
u
) where
u
is a path joining p to u
=
uF
1
(y)B
signDF(p) signDF(u)
= signDF(p)
uF
1
(y)B
signDF(u)
= signDF(p) deg
B.F
(F, B, y),
and from the general denition of each degree, this relation continues to hold for any triple
which is admissible in both theories.
Thus, we can draw the following picture. Each of the above degrees is dened on a class
of admissible triples. The two classes are distinct but not disjoint, and on the intersection of
these classes, the degrees coincide up to a sign. The intersection -in which we are interested-
contains the triples (F, B, y) satisfying
24 CHAPTER 1. INTRODUCTION
(a) F is a continuously dierentiable Fredholm map of index zero from an open connected and
simply connected subset O X, with values in Y .
(b) B O is open.
(c) F has a proper extension to B.
(d) y / F(B).
(e) F has a base point p O.
In the light of the above, Theorem 1.1 (i) remains true under a weaker assumption on the
homotopy. We see this as a theorem on the base point degree (but from outside the theory).
Theorem 1.2 Let h : [0, 1] O Y be a homotopy of Fredholm maps of index 0, which is
proper on [0, 1] B. Suppose that y / h([0, 1] B) and p
0
, p
1
O are respectively base points
for h(0, .) and h(1, .), then
deg
p
0
(h(0, .), B, y) = (D
u
h ) deg
p
1
(h(1, .), B, y),
where is any curve joining (0, p
0
) to (1, p
1
) in [0, 1] O.
Proof. Note rst that the map h(0, .) : O Y is orientable in the sense of Benevieri-
Furi because O is simply connected. Therefore h is orientable (Theorem 4.3 in [4])
5
. Choose
accordingly an orientation of h. Then
deg
p
0
(h(0, .), B, y) = sign (D
u
h(0, p
0
)) deg
B.F
(h(0, .), B, y).
Now the set
(t, u) [0, 1] B [ h(t, u) = y = h
1
(y)
_
[0, 1] B
_
= h
1
(y)
_
[0, 1] B
_
is compact by the properness assumption on h. Therefore, by the invariance of the Benevieri-
Furi degree under oriented homotopies
deg
B.F
(h(0, .), B, y) = deg
B.F
(h(1, .), B, y).
But deg
B.F
(h(1, .), B, y) = sign D
u
h(1, p
1
) deg
p
1
(h(1, .), B, y), and so
deg
p
0
(h(0, .), B, y) = sign D
u
h(0, p
0
) signD
u
h(1, p
1
) deg
p
1
(h(1, .), B, y)
= (D
u
h ) deg
p
1
(h(1, .), B, y),
where is any curve joining (0, p
0
) to (1, p
1
) in [0, 1] O.
We end by some remarks. First, we mention that the degree of Fitzpatrick-Pejsachowicz-
Rabier is available for a class of admissible triples larger than we presented. As in the Benevieri-
Furi theory, for the admissibility of a triple (F, B, y), the properness assumption, and the con-
dition y / F(B) can be replaced by the requirement that F
1
(y) B is compact. This is
discussed in section 7 of [29].
Another extension is discussed in section 8, where it is shown that the assumption on the
simple connectedness of O can be removed provided that one considers oriented maps (in the
sense of Fitzpatrick-Pejsachowicz-Rabier). See [4] 5 for a comparison.
5
This also follows from the fact that [0, 1] O is simply connected and path connected.
Chapter 2
Fredholm and properness properties
of quasilinear second order elliptic
systems
Before beginning the study of Fredholm and properness properties, we need rst to introduce
some notation which will be used in the rest of this work.
Notation
Let N and m be two integers 1. N will always denote the dimension of the space of the
independent variable i.e. IR
N
, and m the dimension of the system (m equations with m unknown
functions). The real number p (which appears in the Sobolev spaces L
p
and W
2,p
) will always
satisfy N < p < . The elements of IR
m
are viewed as columns.
Let be an open subset of IR
N
, with bounded and Lipschitz continuous boundary .
For a vector-valued function u =
_
u
1
, u
2
, . . . , u
m
_
T
: IR
m
, we denote by
i
u the vector
(
i
u
1
, . . . ,
i
u
m
) and by u the mN matrix with columns
i
u, i = 1 . . . , N. For convenience,
we set
0
u = u.
If z
1
, z
2
IR
m
, and A is an m m matrix, z
1
z
2
denotes the scalar product of z
1
and z
2
,
and Az
1
denotes the usual matrix-vector multiplication. For any integer d 1, the Euclidean
norm on IR
d
is denoted by [ [.
In the sequel we deal with functions f :
_
IR
m
IR
mN
_
IR. The arguments of f will
often be denoted by x and IR
m
IR
mN
. Whenever we need to display the components
of IR
m
IR
mN
, we shall write = (
0
,
), where
= [
1
2
N
] , with
k
=
_
_
_
1
k
.
.
.
m
k
_
_
_
for k = 0, . . . , N.
When we deal with Nemytskii operators, we form the expression f(x, u(x), u(x)), in which
u(x) takes the place of
0
, and
1
u(x), . . .
N
u(x) take the place of
1
, . . . ,
N
respectively.
Remark 2.1 The space IR
mN
of m N matrices is indeed isomorphic to the space IR
mN
and therefore it is possible to consider u as an element of IR
mN
. This notation was used in
[17]. What is important in both cases is the block decomposition of u which simplies the
subsequent calculations and statements, by avoiding having to handle a lot of indices.
The Nemytskii operator generated by f: u f(., u, u) will be denoted by f i.e.
f(u)(x) = f(x, u(x), u(x)).
25
26 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
We denote by
k
f(x, ) =
_
1
k
, . . . ,
m
k
_
f(x, ),
the partial gradient with respect to the
k
block variable.
B
r
. Since is
unbounded, there is y
B
r
. Now recall that
B
r
is path connected, so we can join x to
y by a path in
B
r
. This path joining an exterior point to an interior point of , should meet
the boundary, but it does not since the boundary lies inside the ball B
r
. Therefore K =
is
bounded (compact).
Remark 2.3 Let be unbounded. For every ball B
r
containing we have
( B
r
) = B
r
.
Proof. We clearly have B
r
B
r
. Let us prove the reverse inclusion. Let x B
r
.
Then either (i) x B
r
or (ii) x B
r
. Let rst x B
r
and V be an open neighborhood of x.
If V (B
r
) = , then x V B
r
K :=
K
.
Therefore x B
r
. Next if x B
r
B
r
, then for all > 0 suciently small,
B(x, ) . Now clearly B(x, ) B
r
,= and so B(x, ) ( B
r
) ,= , and once again
x B
r
. Finally B
r
= B
r
.
2.1. SMOOTHNESS OF SOME NEMYTSKII OPERATORS 27
On the other hand,
B
r
=
B
r
.
So
( B
r
) =
B
r
_
B
r
B
r
= B
r
_
B
r
=
_
B
r
,
since B
r
B
r
and B
r
B
r
.
Remark 2.4 The connection between the Dirichlet problem and the space
D
p
() =
_
W
2,p
() W
1,p
0
()
_
m
introduced above is given by the following theorem (see Brezis [6] Theor`eme IX. 17 for example).
Let have a C
1
boundary and u W
1,q
() C() with 1 q < . Then the following
conditions are equivalent.
(i) u = 0 on .
(ii) u W
1,q
0
().
2.1 Smoothness of some Nemytskii operators
Our rst task is to make sure that the operator in (1.2) maps W
2,p
(, IR
m
) to L
p
(, IR
m
) and
has enough smoothness for the subsequent discussion. Therefore it is necessary to study the
smoothness of the Nemytskii operators u b(., u, u) and u a
f,
[[ k, exist and are equicontinuous C
0
bundle maps.
We recall that equicontinuity of (f(x, .))
x
at a point
0
IR
M
means that for all > 0
there is = (
0
, ) > 0 such that [
0
[ [f(x, ) f(x,
0
)[ < for all x . When
can be chosen independently of
0
for
0
in some set B, we have uniform equicontinuity on B.
Note that f = (f
1
, . . . , f
d
) is an equicontinuous C
k
bundle map.
Now we give some important properties and examples of equicontinuous C
k
bundle maps.
Lemma 2.1 Let f : IR
M
IR
d
be an equicontinuous C
0
bundle map. Then we have the
following.
(i) The collection (f(x, .))
x
is uniformly equicontinuous on the compact subsets of IR
M
.
28 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
(ii) If A is a measurable subset of and f(., 0) L
(A)
+
_
[[
_
+ 1.
But [[ is bounded by the diameter of B, so the proof is complete.
Remark 2.5 Let g :
_
IR
m
IR
mN
_
IR
m
be an equicontinuous C
0
bundle map, such
that g(., 0) L
(, IR
m
). Then, for i = 0, . . . , N, (x, ) g(x, )
i
is a scalar-valued
equicontinuous C
0
bundle map.
Proof. Fix IR
m
IR
mN
. Then
g(x, )
i
g(x, )
i
= g(x, ) (
i
i
) + (g(x, ) g(x, ))
i
.
The result follows from the equicontinuity of (g(x, .))
x
at , and its equiboundedness on
bounded subsets of IR
m
IR
mN
(Lemma 2.1 (ii)).
Remark 2.6 Let f :
_
IR
m
IR
mN
_
IR be an equicontinuous C
1
i
f(x, t) dt.
Then g is an equicontinuous C
0
bundle map.
Proof. Fix IR
m
IR
mN
. Then
g(x, ) g(x, ) =
_
1
0
(
i
f(x, t)
i
f(x, t)) dt.
If [ [ 1 then t and t belong to the closed ball with center 0 and radius [[ + 1. Thus
the conclusion follows from Lemma 2.1 (i) applied to
i
f.
2.1. SMOOTHNESS OF SOME NEMYTSKII OPERATORS 29
Remark 2.7 If f is of class C
k
and f(., ) is periodic in x with period T = (T
1
, . . . , T
N
) for
every IR
m
IR
mN
, then f is an equicontinuous C
k
f on [0, T
1
] . . . [0, T
N
] K for every compact K IR
m
IR
mN
and [[ k (see 2.5).
Lemma 2.2 Let f : IR
m
IR
mN
IR be an equicontinuous C
0
bundle map. Suppose
that f(., 0) L
().
(ii) It is well dened and continuous from
_
W
2,p
()
_
m
to L
() (hence also to
L
q
(), 1 q ).
(iv) The multiplication (u, v)
_
W
2,p
()
_
m
L
p
() f(., u, u)v L
p
() is weakly sequen-
tially continuous.
Proof. (i) If u C
1
d
(, IR
m
), the function x f(x, u(x), u(x)) is continuous and hence
measurable. From the boundedness of u and u on , there is a bounded subset K IR
m
IR
mN
containing (u(x), u(x)) for all x . Therefore by Lemma 2.1 (ii) there is a constant
M
K
> 0 such that [f(x, u(x), u(x))[ M
K
x . This means that f(., u, u) L
().
To prove the continuity, let u
n
, u C
1
d
(, IR
m
) and u
n
u in C
1
d
(, IR
m
). Then since
u u
n
, n IN is compact and hence bounded in C
1
d
(, IR
m
), there is a compact K
IR
m
IR
mN
containing (u(x), u(x)) and (u
n
(x), u
n
(x)) for all x and n IN. But
[(u
n
(x), u
n
(x)) (u(x), u(x))[ can be made arbitrary small uniformly in x , for n large
enough, so by Lemma 2.1 (i), given > 0, we have
[f(x, u
n
(x), u
n
(x)) f(x, u(x), u(x))[ x .
Lastly if B C
1
d
(, IR
m
) is bounded, there is a bounded subset K IR
m
IR
mN
containing
(u(x), u(x)) for all x and u B. The boundedness of f(B) follows from Lemma 2.1 (ii).
(ii) Follows from the imbedding W
2,p
(, IR
m
) C
1
d
(, IR
m
).
(iii) The above imbedding is compact when is bounded.
(iv) Let u
n
u in W
2,p
(, IR
m
) and v
n
v in L
p
(). From part (ii), the sequence (f(u
n
))
is bounded in L
f(u)[
in L
f(u)v[
in L
p
(
).
Now if a subsequence of (f(u
n
)v
n
) converges weakly to w in L
p
() and hence in L
p
(
), we
have w[
= f(u)v[
, and therefore w = f(u)v since the ball is arbitrary. This means that
(f(u
n
)v
n
) has a unique weak cluster point, which yields f(u
n
)v
n
f(u)v in L
p
(), by Note
A1 of the appendix.
Lemma 2.3 Let f : IR
m
IR
mN
IR have the form
f(x, ) = f
0
(x) +
N
i=0
g
i
(x, )
i
(2.1)
where g
i
is a C
0
bundle map, with g
i
(., 0) L
(, IR
m
), 0 i N. Suppose that f
0
L
p
().
In particular, the above conditions hold if f is a C
1
f(., 0) bounded in . Then the Nemytskii operator has the following properties.
30 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
(i) It is well dened and continuous from W
2,p
(, IR
m
) to L
p
() and maps bounded subsets
onto bounded subsets.
(ii) It is weakly sequentially continuous from W
2,p
(, IR
m
) to L
p
().
Proof. To see the in particular, note that for an equicontinuous C
1
t
f(x, t) dt
=
_
1
0
N
i=0
i
f(x, t)
i
dt
=
N
i=0
__
1
0
i
f(x, t) dt
_
i
. (2.2)
Take
g
i
(x, ) =
_
1
0
i
f(x, t) dt,
then by Remark 2.6, g
i
is an equicontinuous C
0
bundle map. Furthermore g
i
(., 0) =
i
f(., 0)
L
(, IR
m
).
(i) Applying Lemma 2.2 (ii) to each component of g
i
, we have that g
i
: W
2,p
(, IR
m
)
L
(, IR
m
) is continuous and maps bounded subsets onto bounded ones. As a result, the
operator (recall that
0
u = u)
u
N
i=0
g
i
(u)
i
u L
p
() (2.3)
is continuous and maps bounded subsets onto bounded ones. By (2.1), this is f f
0
, and the
conclusion follows from the assumption f
0
L
p
().
(ii) Let u
n
u in W
2,p
(, IR
m
). By part (i), (f(u
n
)) is bounded in L
p
(). Let
be an open ball. Since f f
0
is an equicontinuous C
0
bundle map (see Remark 2.5) and
vanishes when = 0, Lemma 2.2 (iii) applies and yields f(u
n
)[
f(u)[
in L
p
(
). Now if
a subsequence of (f(u
n
)) converges weakly to some w in L
p
() and hence in L
p
(
), we have
w[
= f(u)[
, and therefore w = f(u) since the ball is arbitrary. This means that (f(u
n
))
has a unique weak cluster point, and thus f(u
n
) f(u) in L
p
().
Theorem 2.1 Let f : IR
m
IR
mN
IR be an equicontinuous C
1
() (resp. f(., 0) L
p
()) and that
(), (resp L
p
()) with derivative
Df(u)v =
N
i=0
i
f(., u, u)
i
v. (2.4)
Furthermore Df is bounded on the bounded subsets of W
2,p
(, IR
m
), and hence f is uniformly
continuous on these subsets.
2.1. SMOOTHNESS OF SOME NEMYTSKII OPERATORS 31
Proof. Dene for u W
2,p
(, IR
m
),
T
u
v =
N
i=0
i
f(., u, u)
i
v.
By Lemma 2.2 (ii) applied to each component of
i
f, we have
i
f(., u, u) is bounded on
. Thus
|T
u
v|
0,p,
N
i=0
|
i
f(., u, u)|
0,,
|
i
v|
0,p,
const. |v|
2,p,
and
|T
u
v|
0,,
const. |v|
1,,
const. |v|
2,p,
.
Therefore T
u
is linear and bounded from W
2,p
(, IR
m
) to L
p
() and to L
().
Note that
f(., u +v, (u +v)) f(., u, u) =
_
1
0
t
f(., u +tv, u +tv) dt
=
N
i=0
__
1
0
i
f(., u +tv, u +tv) dt
_
i
v.
So
f(., u +v, (u +v)) f(., u, u) T
u
v
=
N
i=0
__
1
0
i
f(., u +tv, u +tv)
i
f(., u, u) dt
_
i
v.
Thus if we dene
k
u,i
(x, ) :=
_
1
0
(
i
f( x, u(x) +t
0
, u(x) +t
i
f(x, u(x), u(x))) dt
where
= [
1
N
], we get
f(., u +v, (u +v)) f(., u, u) T
u
v =
N
i=0
k
u,i
(., v, v)
i
v. (2.5)
Now, one can check as in Remark 2.6, that k
u,i
is an equicontinuous C
0
bundle map satisfying
k
u,i
(., 0) = 0 L
(, IR
m
). Therefore by Lemma 2.2 (ii) applied to each component of k
u,i
we
have that k
u,i
is continuous from W
2,p
(, IR
m
) to L
(, IR
m
). So, given > 0, we have that
|k
u,i
(v)|
0,,
provided |v|
2,p,
is small enough.
Now, if f(., 0) L
(), f maps W
2,p
(, IR
m
) to L
f(., u, u)
f(., u
0
, u
0
)|
0,,
where |Df(u) Df(u
0
)| denotes either the norm in L(X
p
(), L
()) if f(., 0) L
(), or
the norm in L(X
p
(), L
p
()) if f(., 0) L
p
(). The result then follows from Lemma 2.2 (ii)
applied to each component of
i=0
i
f
k
(., u, u)
i
v
_
_
_
(2.6)
Similarly if f =
_
f
k,j
_
k,j=1,...,m
is an mm matrix, the derivative of the Nemytskii operator f
is the matrix
Df(u)v =
_
Df
k,j
(u)v
_
k,j=1,...,m
with
Df
k,j
(u)v =
N
i=0
i
f
k,j
(., u, u)
i
v
_
_
(2.7)
Lemma 2.4 (Lemma 2.9 of [31]) Let X, Y and Z be normed spaces with X Y and let
f : X Z be uniformly continuous on the bounded subsets of X. Suppose that there is a
dense subset D X such that whenever u D and (u
n
) X is a bounded sequence with
u
n
u in Y, we have f(u
n
) f(u) in Z. Then the restriction of f to the bounded subsets of
X remains continuous for the topology induced by Y .
Lemma 2.5 Let f : IR
m
IR
mN
IR be an equicontinuous C
1
i
f(., 0) (L
p
() L
())
m
, 0 i N. Then the restriction
of the Nemytskii operator to any bounded subset of W
2,p
(, IR
m
) is continuous into L
p
() for
the topology of C
1
d
(, IR
m
).
Proof. Recall that f is uniformly continuous on the bounded subsets of W
2,p
(, IR
m
) by
Theorem 2.1. Note also that if
D =
_
u C
(, IR
m
) [ v C
0
(IR
N
, IR
m
) such that v[
= u
_
,
then D is dense in W
2,p
(, IR
m
) (Adams [1] Theorem 3.18). We show that if u D and (u
n
) is
a bounded sequence from X
p
() converging to u in C
1
d
(, IR
m
), then f(u
n
) f(u) in L
p
().
The result will follow from Lemma 2.4 with X = W
2,p
(, IR
m
), Y = C
1
d
(, IR
m
), Z = L
p
()
and D dened above.
In Lemma 2.3 we have already established that
f(u) = f(., 0) +
N
i=0
g
i
(u)
i
u,
where
g
i
(x, ) =
_
1
0
i
f(x, t) dt,
2.1. SMOOTHNESS OF SOME NEMYTSKII OPERATORS 33
and by Remark 2.6, the g
i
are equicontinuous C
0
bundle maps. Hence by Lemma 2.2 (i),
applied to each component of g
i
, g
i
: C
1
d
(, IR
m
) L
(, IR
m
) is continuous.
Clearly the problem reduces to showing that g
i
(u
n
)
i
u
n
g
i
(u)
i
u in L
p
(), 0 i N.
To see this we write
g
i
(u
n
)
i
u
n
g
i
(u)
i
u = (g
i
(u
n
) g
i
(u))
i
u
n
+ g
i
(u)
i
(u
n
u). (2.8)
The rst term tends to zero in L
p
() because g
i
(u
n
) g
i
(u) in L
(, IR
m
), and (
i
u
n
) is
bounded in L
p
(, IR
m
). On the other hand (u
n
u) 0 in C
1
d
(, IR
m
) which is continuously
imbedded in W
1,
(, IR
m
). The last term of (2.8) tends to zero if we show that g
i
(u)
L
p
(, IR
m
). And this is true for the following reason: let
(, IR
m
) L
, IR
m
) L
p
(
, IR
m
), and secondly, when x
, we have
g
i
(u)(x) = g
i
(x, u(x), u(x)) = g
i
(x, 0) =
i
f(x, 0) . But
i
f(., 0) L
p
(, IR
m
), therefore
g
i
(u) L
p
(
, IR
m
). And thus g
i
(u) L
p
(, IR
m
) as claimed.
Lemma 2.6 Let f : IR
m
IR
mN
IR be an equicontinuous C
1
)
N
i=0
i
f(x, u(x), u(x))
i
where
i
f(., u, u) is continuous and bounded (Lemma 2.2 (ii)). So one can check using
Lemma 2.1 (i), that g is an equicontinuous C
1
i
g(x, ) =
i
f(x, u(x) +
0
, u(x) +
i
f(x, u(x), u(x)).
Furthermore g(0) = f(u) L
p
() (Lemma 2.3 (i)), and
i
g(., 0) = 0 L
p
(, IR
m
)
L
(, IR
m
). Thus g veries the conditions of Lemma 2.5, and therefore g(v
n
) g(0) in
L
p
(), which completes the proof.
Smoothness of F
Let the coecients of F in (1.2) satisfy the following assumptions:
a
are equicontinuous C
1
(., 0) and
are
continuous from W
2,p
(, IR
m
) to L
(, IR
mm
) and they map bounded subsets onto bounded
subsets. By Lemma 2.3 (i) (applied to each component of b), b is continuous from W
2,p
(, IR
m
)
to L
p
(, IR
m
) and maps bounded subsets onto bounded ones. This proves the continuity and
the boundedness properties.
Now if (u
n
) W
2,p
(, IR
m
) converges weakly to u, we have
2
u
n
2
u in L
p
(, IR
m
)
1
.
By Lemma 2.2 (iv), a
(u
n
)
2
u
n
a
(u)
2
u in L
p
(, IR
m
). Next by Lemma 2.3 (ii),
b(u
n
) b(u) in L
p
(, IR
m
). This proves the weak continuity of F.
Remark 2.9 Note that the proof of the above lemma requires only the following weaker as-
sumptions: a
are equicontinuous C
0
bundle maps, with a
i=0
c
i
(x, )
i
where b
0
L
p
(, IR
m
), and c
i
are equicontinuous C
0
bundle maps with c
i
(., 0) bounded. This
will be used in 2.5.
Theorem 2.2 The operator F in (1.2) is of class C
1
from X
p
() = W
2,p
(, IR
m
) to Y
p
() =
L
p
(, IR
m
), with derivative
DF(u)v =
N
,=1
a
(u)
2
v + Db(u)v
N
,=1
(Da
(u)v)
2
u (2.14)
where Db(u) and Da
,=1
a
(u)
2
u + b(u).
By Theorem 2.1, b C
1
(X
p
(), Y
p
()), and a
C
1
_
X
p
(), L
(, IR
mm
)
_
. Now let
G(u) := a
(u)
2
u, and B : L
(, IR
mm
)Y
p
() Y
p
() be the bounded bilinear operator
dened by B(/, A) = /A. Then G = B (a
,
2
: X
p
() Y
p
() is linear and bounded and therefore weakly continuous.
2.2. ELLIPTICITY AND EXAMPLES 35
Let A
, B
,=1
A
(x)
2
v +
N
=1
B
(x)
v +C(x)v, (2.15)
where v : IR
m
.
For the majority of our results in this chapter, the following condition of ellipticity is su-
cient.
Denition 2.2 (Petrovskii) An operator L of the form (2.15) is said to be elliptic at x in
the sense of Petrovskii, if there exists a positive constant (= (x)) such that
det
_
_
N
,=1
(x)
_
_
[[
2m
IR
N
. (2.16)
We say that L is strictly elliptic on a subset K , if in the above denition one can choose
the same for all x K.
We mention right away that the (linear) Stokes system is not elliptic in the sense of Petrovskii.
But a study of the Fredholm and properness properties of the Navier-Stokes operator on un-
bounded domains, was already carried out by Galdi and Rabier [15], [16]. The two physical
examples we have in mind are steady reaction diusion systems and elasticity, and these satisfy
a stronger condition of ellipticity known as the strong Legendre-Hadamard (see [8], [25]).
Denition 2.3 (The Strong Legendre-Hadamard ellipticity) An operator of the form
(2.15) is called strongly elliptic at x in the sense of Legendre-Hadamard, if there is = (x) > 0,
such that
T
_
_
N
,=1
(x)
_
_
[[
2
[[
2
for all IR
N
, IR
m
. (2.17)
We say that L is strongly elliptic on a subset K IR
N
in the sense of Legendre-Hadamard, if
(2.17) holds with the same > 0 for all x K.
Indeed this is stronger than Petrovskii ellipticity, since (2.17) means that the matrix
N
,=1
(x)
is positive denite for ,= 0 (and so all its real eigenvalues are positive), whereas Petrovskii
condition means that it has a positive determinant. Note that the strong Legendre-Hadamard
condition is convex in the sense that all operators on the segment joining two elliptic operators
(in the sense of Legendre-Hadamard) are elliptic. However, this is not true for Petrovskii-
ellipticity. On the other hand, both conditions concern only the higher order coecients of
the system, and furthermore they are stable under small enough perturbation of the leading
coecients. We mention nally that the Petrovskii condition is available for higher order
systems (see [22]).
36 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
Steady reaction-diusion of particles in a uid ow
For N 3, let v : IR
N
IR
N
denote the velocity eld of a stationary ow containing m types
of particle in suspension. Let u
k
(x, t) denote the density of particles of type k at time t and
position x IR
N
. The particles move with the uid, diuse with diusion coecients D
k
> 0
and take part in a chemical reaction. According to [14], the evolution of the particle densities
in the uid is governed by the system
t
u
k
=
D
k
2
u
k
+v(x).u
k
+f
k
(x, u
1
, ..., u
m
) for k = 1, 2..., m.
In a matrix-vector form, steady states of this system satisfy
,=1
_
_
_
_
_
D
1
2
0 0
0
D
2
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0
D
m
2
_
_
_
_
_
_
_
_
_
_
_
u
1
.
.
.
.
.
.
u
m
_
_
_
_
_
_
_
_
_
_
_
_
v u
1
+f
1
(x, u)
.
.
.
.
.
.
v u
m
+f
m
(x, u)
_
_
_
_
_
_
= 0,
which indeed is a system of the form (1.1) with
a
ij
(x,
0
, [
1
N
]) =
D
i
2
ij
for , = 1, . . . , N and i, j = 1, . . . , m,
b
j
(x,
0
, [
1
N
]) =
_
N
=1
v
(x)
j
+f
j
(x,
0
)
_
.
Now the linearization of the above system is of the form (2.15) with A
= a
. Indeed
N
,=1
=
N
=1
=
[[
2
2
diag (D
1
, . . . , D
m
),
and therefore
T
_
_
N
,=1
_
_
minD
k
2
[[
2
[[
2
.
Thus the system is strongly elliptic in the sense of Legendre-Hadamard.
Our conditions (2.10)(2.13) are satised provided that
(i) v
C(IR
N
) L
(IR
N
), and
(ii) f
j
: IR
N
IR
m
IR is an equicontinuous C
1
f
j
(, 0) is bounded on IR
N
.
2.3 Fredholmness
We now begin the investigation of the Fredholmness of the second order dierential operator
(1.2). To the hypotheses (2.10)(2.13), we add an ellipticity condition, which implies that the
linearization DF(u) is Petrovskii-elliptic.
For all IR
N
, (x, )
_
IR
m
IR
mN
_
det
_
N
,=1
(x, )
_
(x, )[[
2m
_
_
(2.18)
2.3. FREDHOLMNESS 37
where :
_
IR
m
IR
mN
_
(0, ) is bounded from below by a positive constant on every
compact subset of
_
IR
m
IR
mN
_
.
Note that in the case of a single equation (m = 1), this condition reduces to
N
,=1
a
(x, )
(x, )[[
2
IR
N
, (x, ) IR
N+1
,
and this is the ellipticity condition used in [31] with = IR
N
. In the remainder of this chapter,
the coecients of the operator F in (1.2) will satisfy the hypotheses (2.10)(2.13) and (2.18).
Note that
DF(u)v = L(u)v
N
,=1
(Da
(u)v)
2
u,
where
L(u)v :=
N
,=1
a
(u)
2
v + Db(u)v (2.19)
and clearly L(u) L(X
p
(), Y
p
()).
Lemma 2.8 Let u X
p
(). Then the dierence DF(u) L(u) is compact between X
p
() and
Y
p
(). Therefore, given Z , and any closed subspace E of X
p
(), we have
DF(u)
(E, Y
p
()) L(u)
(E, Y
p
()).
Proof. If we show that the dierence is compact, the second statement will follow from the
stability of
(E, Y
p
()) under compact perturbations.
By (2.19) we have
DF(u)v L(u)v =
N
,=1
(Da
(u)v)
2
u.
Clearly it suces to show that each (Da
(u)v)
2
j=1
(Da
(u)v)
k,j
2
u
j
(k = 1, . . . , m), is compact. Now each
term of this sum is by (2.7)
_
N
i=0
i
a
k,j
(., u, u)
i
v
_
u
j
.
Once again writing the components, we deal with the terms
_
l
i
a
k,j
(., u, u)
2
u
j
_
i
v
l
, k, j, l =
1, . . . , m. Now indeed
l
i
a
k,j
(., u, u)
2
u
j
L
p
(). But since N < p < , the multiplication
by a xed function of L
p
() is a compact operator from W
1,p
() to L
p
() (see the appendix).
Thus w Tw :=
_
l
i
a
k,j
(., u, u)
2
u
j
_
i
w is compact from W
2,p
() to
2
L
p
(). Now going
back through the steps, we see that DF(u) L(u) is a compact operator from X
p
() to Y
p
(),
and therefore also from D
p
() to Y
p
().
2
w
n
w in W
2,p
i
w
n
i
w in W
1,p
Tw
n
Tw in L
p
.
38 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
Fix u and let A
(x) = a
(u)(x), B
b
k
(x, u(x), u(x)),
k = 1, . . . , m, and C(x) be the matrix with lines
0
b
k
(x, u(x), u(x)), k = 1, . . . , m. Then
L(u)v =
N
,=1
A
(x)
2
v +
N
=1
B
(x)
v +C(x)v
is a linear second order dierential operator, with continuous and bounded coecients. Now
condition (2.18) implies
det
_
_
N
,=1
(x)
_
_
(x, u(x), u(x))[[
2m
.
As x varies over a compact set, the continuity of u and u ensures that (x, u(x), u(x)) remain
in a compact set K. Therefore by (2.18) there exists
K
> 0 such that
det
_
_
N
,=1
(x)
_
_
K
[[
2m
.
Thus, for each xed u X
p
() the dierential operator L(u) is strictly Petrovskii-elliptic on
the compact subsets of .
Lemma 2.9 (Koshelev [22], Theorem 17 pp. 150-151) Let
IR
N
be a bounded do-
main with C
2
boundary, and 1 < q < . Let the linear operator L in (2.15) be strictly
Petrovskii-elliptic in
. If v W
2,q
(
, IR
m
)W
1,q
0
(
, IR
m
),
then v satises the a priori estimate
|v|
2,q,
c
_
|Lv|
0,q,
+|v|
0,1,
_
, (2.20)
where c is a positive constant.
Lemma 2.10 Assume that is of class C
2
. Let L be a second order linear dierential
operator, strictly Petrovskii-elliptic on the compact subsets of , with continuous bounded
coecients
3
. If (u
n
) D
p
(), is a sequence converging weakly to zero in D
p
(), and Lu
n
0
in Y
p
(), then u
n
0 in X
p
(
.
Proof. We distinguish between two cases.
Case 1 ( is bounded). Since u
n
0 in W
2,p
(, IR
m
)
comp
L
1
(; IR
m
), we have u
n
0
in L
1
(, IR
m
). On the other hand Lu
n
0 in L
p
(, IR
m
). Note that (u
n
), L and satisfy
the conditions of Lemma 2.9. So by letting q = p and v = u
n
in (2.20), we get: u
n
0 in
W
2,p
(, IR
m
).
Case 2 ( is unbounded). For every r > 0, set B
r
= x IR
N
; [x[ < r, and
r
= B
r
.
Clearly it is equivalent to show that the result hold when
=
r
for r > 0 large enough
4
.
So let B
r
be a ball containing , and R > r. It follows from the remarks made about at
the beginning of this chapter, that
R
= B
R
so that
R
is C
2
since B
R
= .
3
The boundedness of the coecients ensures that L maps continuously W
2,p
(, IR
m
) into L
p
(, IR
m
).
4
Because
r
is open and bounded, and every open and bounded subset of is contained in a subset of the
form
r
.
2.3. FREDHOLMNESS 39
Now dene : IR
N
IR to be a C
,
_
_
u
n
+
,
(
u
n
+
,
(
u
n
+
) u
n
.
Thus due to the boundedness of , of its derivatives and of the coecients of L, we get Lv
n
0
in L
p
(
R
, IR
m
).
Estimate (2.20) now gives that v
n
0 in X
p
(
R
), and therefore also in X
p
(
r
). This nally
implies that u
n
X
p
(
r
).
For the next results, we need the following concept introduced in [31].
Denition 2.4 Let X and Y be real Banach spaces with X reexive and let T, L L(X, Y )
be given. We say that T is compact modulo L if, for every sequence (u
n
) X, we have
u
n
0 in X, Lu
n
0 in Y Tu
n
0 in Y .
Lemma 2.11 (Lemma 3.7 of [31]) Let X and Y be real Banach spaces with X reexive
and let L
0
, L
1
L(X, Y ) be given. Suppose L
0
L
1
is compact modulo both L
0
and L
1
. Then
we have the following.
(i) If (u
n
) X is a sequence converging weakly to zero, we have L
0
u
n
0 in Y if and only
if L
1
u
n
0.
(ii) L
0
+
(X, Y ) if and only if L
1
+
(X, Y ).
For t [0, 1] dene L
t
:= tL
1
+ (1 t)L
0
. If L
0
L
1
is compact modulo L
t
t [0, 1], then
(iii) L
t
+
(X, Y ) for all t [0, 1] if and only if this holds for some t
0
[0, 1], and in this
case, the index of L
t
is independent of t.
Lemma 2.12 Assume that is C
2
. For L(u) dened by (2.19), the relation
L(u)
+
(D
p
(), Y
p
())
holds for every u D
p
(), if and only if it holds for some u
0
D
p
().
Proof. We shall prove that L(u) L(u
0
) is compact modulo L(u). By exchanging the roles of
u and u
0
, this shows that L(u) L(u
0
) is compact modulo both L(u) and L(u
0
), the conclusion
follows from Lemma 2.11 (ii).
Let then (v
n
) D
p
() be such that v
n
0 in D
p
() and L(u)v
n
0 in Y
p
(). From the
equicontinuity of a
and D
(x, ) a
b(x, )
(x, u
0
(x), u
0
(x))[ <
and
[
b(x, u
0
(x), u
0
(x))[ <
whenever [x[ r. Now let
r
= x [ [x[ < r,
r
= x [ [x[ > r, and recall that
_
L(u) L(u
0
)
_
v =
N
,=1
_
a
(u) a
(u
0
)
_
v+
N
i=0
_
i
b(., u, u)
i
b(., u
0
, u
0
)
_
i
v.
Therefore
|
_
L(u) L(u
0
)
_
v|
0,p,
r
m
2
(N
2
+N + 1)|v|
2,p,
r
for v D
p
().
Hence
|
_
L(u) L(u
0
)
_
v
n
|
0,p,
r
m
2
M(N
2
+N + 1) (2.21)
where M is a bound for |v
n
|
2,p,
.
As already observed L(u) veries the conditions required in Lemma 2.10, thus v
n
0 in
X
p
(
r
), so L(u
0
)v
n
and L(u)v
n
converge to zero in Y
p
(
r
)
5
, which means that for any > 0
and n large enough,
|
_
L(u) L(u
0
)
_
v
n
|
0,p,
r
. (2.22)
Together (2.21) and (2.22), yield that |
_
L(u) L(u
0
)
_
v
n
|
0,p,
can be made arbitrary small for
n large enough. This completes the proof.
Theorem 2.3 Let be of class C
2
. The operator F in (1.2) is semi-Fredholm of index
Z (i.e. DF(u)
(D
p
(), Y
p
()) for every u D
p
()), if and only if there is
some u
0
D
p
() such that DF(u
0
)
(D
p
(), Y
p
()).
Proof. By Lemma 2.8 and 2.12,
DF(u
0
)
+
(D
p
(), Y
p
()) L(u
0
)
+
(D
p
(), Y
p
())
L(u)
+
(D
p
(), Y
p
()) u D
p
()
DF(u)
+
(D
p
(), Y
p
()) u D
p
().
Now by Theorem 2.2, DF is continuous as a map from D
p
() into L(D
p
(), Y
p
()). Recall
also that the index of a semi-Fredholm operator is locally constant, whence u index DF(u)
is locally constant and therefore constant since D
p
() is connected.
5
L(u) L(X
p
(
r
), Y
p
(
r
)) by Theorem 2.2 with replaced by
r
.
2.3. FREDHOLMNESS 41
A useful factorization of F
Theorem 2.4 There exists an operator G : X
p
() L(X
p
(), Y
p
()), having the following
properties.
(a) F(u) F(0) = G(u)u, for every u X
p
().
(b) For every Z and u D
p
(), we have DF(u)
(D
p
(), Y
p
()) if and only
if G(u)
(D
p
(), Y
p
()).
(c) Gis continuous and maps bounded subsets of X
p
() into bounded subsets of L(X
p
(), Y
p
()).
Proof. (a) Recall that
F(u) =
N
,=1
a
(u)
2
u + b(u), (2.23)
and b(u) can be written as
b(u) = b(0) +
N
i=0
c
i
(u)
i
u (2.24)
where c
i
is the Nemytskii operator generated by the (matrix-valued) equicontinuous C
0
bundle
map
c
i
(x, ) =
_
1
0
i
b(x, t) dt.
By Theorem 2.2,
DF(u)v =
N
,=1
a
(u)
2
v + Db(u)v
N
,=1
(Da
(u)v)
2
u
=
N
,=1
a
(u)
2
v +
N
i=0
b
i
(u)
i
v +K(u)v (2.25)
where indeed K(u) is compact as already observed in Lemma 2.8, and b
i
is the Nemytskii
operator associated with
i
b(x, ). In particular when v = u we obtain
DF(u)u =
N
,=1
a
(u)
2
u +
N
i=0
b
i
(u)
i
u +K(u)u. (2.26)
Combining (2.23), (2.24) and (2.26) we get since F(0) = b(0),
F(u) F(0) = DF(u)u K(u)u
N
i=0
(b
i
(u) c
i
(u))
i
u. (2.27)
Letting
T(u)v =
N
i=0
(b
i
(u) c
i
(u))
i
v, (2.28)
and
G(u) = DF(u) K(u) T(u) = L(u) T(u), (2.29)
where L(u) is dened in (2.19), we get F(u) F(0) = G(u)u.
42 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
Note that since T(0) = K(0) = 0, we have G(0) = L(0) = DF(0).
(b) We show that T(u) is compact. From the equicontinuity of (
i
b(x, .))
x
at = 0, given
any > 0, there is > 0 such that x
[
i
b(x, )
i
b(x, 0)[ <
2
whenever [[ < . Therefore x , t [0, 1], we have [
i
b(x, )
i
b(x, t)[ < if [[ < .
But since u C
1
d
(, IR
m
) there is an r > 0 such that [ (u(x), u(x)) [ < whenever [x[ > r. All
this means that t [0, 1], [x[ > r, we have
[
i
b(x, u(x), u(x))
i
b(x, tu(x), tu(x))[ < ,
and so by integrating with respect to t we get
lim
|x|
_
i
b(x, u(x), u(x))
_
1
0
i
b(x, tu(x), tu(x)) dt
_
= 0.
But multiplication by a bounded function vanishing at innity is a compact operator from W
1,q
to L
q
for all 1 < q < (see Note E2 in the appendix). Therefore T(u) is compact from D
p
()
to Y
p
() for all u D
p
().
(c)
G(u)v = L(u)v T(u)v =
N
,=1
a
(u)
2
v +
N
=0
c
(u)
v.
Therefore,
|G(u)v G(u
0
)v|
0,p,
N
,=1
|a
(u) a
(u
0
)|
0,,
|
2
v|
0,p,
+
N
=0
|c
(u) c
(u
0
)|
0,,
|
v|
0,p,
.
As already observed, the Nemytskii operators a
and c
(, IR
mm
). Accordingly, given > 0, there is > 0 such that
|a
(u) a
(u
0
)|
0,,
and |c
(u) c
(u
0
)|
0,,
,
whenever |u u
0
| . Consequently,
|G(u)v G(u
0
)v|
0,p,
const. |v|
2,p,
,
and therefore,
|G(u) G(u
0
)|
L(X
p
(),Y
p
())
const. .
The boundedness property follows from the boundedness of a
and c
(Lemma 2.2).
2.4 Properness
Lemma 2.13 Let u W
2,p
(, IR
m
) and (u
n
) W
2,p
(, IR
m
) be a bounded sequence converg-
ing to u in C
1
d
(, IR
m
). Then F(u
n
) F(u) DF(u)(u
n
u) 0 in L
p
(, IR
m
).
2.4. PROPERNESS 43
Proof. Note rst that u
n
u in W
2,p
(, IR
m
), by Note A2 of the appendix.
F(u
n
) F(u) DF(u)(u
n
u)
=
N
,=1
_
a
(u
n
) a
(u)
_
u
n
+
N
,=1
_
Da
(u)(u
n
u)
_
u
+ b(u
n
) b(u) Db(u)(u
n
u). (2.30)
As already observed in the proof of Lemma 2.8, v (Da
(u)v)
2
u is a compact linear
operator from X
p
() to Y
p
(), therefore
N
,=1
_
Da
(u)(u
n
u)
_
u 0 in L
p
(, IR
m
). (2.31)
By Lemma 2.6
b(u
n
) b(u) Db(u)(u
n
u) 0 in L
p
(, IR
m
). (2.32)
By Lemma 2.2 (i), a
(u
n
) a
(u) in L
(, IR
mm
), and since
2
u
n
is bounded in
L
p
(, IR
m
)
N
,=1
_
a
(u
n
) a
(u)
_
u
n
0 in L
p
(, IR
m
). (2.33)
r
= x : [x[ > r for every r > 0.
(i) (ii). It follows from note A2 in the appendix that u
n
u in X
p
(). Next let > 0
be given. There is r > 0 for which [u(x)[ + [u(x)[ /2 whenever x
r
. Let n
0
be
such that |u
n
u|
1,,
/2 for n n
0
. Then for every x
r
and n n
0
, we have
[u
n
(x)[ +[u
n
(x)[ .
(ii) (i). Let > 0 be given and let r > 0 and n
0
IN be such that [u
n
(x)[ +[u
n
(x)[
/2 whenever x
r
and n n
0
. After increasing r if necessary we may assume that
[u(x)[ +[u(x)[ /2. Hence
[u(x) u
n
(x)[ +[u(x) u
n
(x)[ , x
r
n n
0
.
Next, since X
p
(
r
)
comp
C
1
d
(
r
, IR
m
) there is n
1
IN such that
[u(x) u
n
(x)[ +[u(x) u
n
(x)[ , x
r
n n
1
.
Thus nally, we have |u
n
u|
1,,
for n max(n
0
, n
1
), which shows that u
n
u in
C
1
d
(, IR
m
) as claimed.
Corollary 2.1 Let have a C
2
boundary. Suppose there exists u
0
D
p
() for which
DF(u
0
)
+
(D
p
(), Y
p
()). The following conditions are equivalent.
(i) F : D
p
() Y
p
() is proper on the closed bounded subsets of D
p
().
(ii) Every bounded sequence (u
n
) D
p
() such that (F(u
n
)) converges in Y
p
(), vanishes
uniformly at innity in the sense of C
1
d
().
(iii) Every bounded sequence (u
n
) D
p
() such that (F(u
n
)) converges in Y
p
(), contains
a subsequence vanishing uniformly at innity in the sense of C
1
d
().
Proof. (i) (ii). Let (u
n
) be a bounded sequence from D
p
() such that (F(u
n
)) converges
in Y
p
(), and suppose that (u
n
) does not vanish uniformly at innity. Set
n
(x) = [u
n
(x)[ +
[u
n
(x)[. Then there are
0
> 0, a subsequence u
(n)
, and a sequence (x
n
) such that
[x
n
[ n and
(n)
(x
n
) = [u
(n)
(x
n
)[ + [u
(n)
(x
n
)[
0
. Now (u
(n)
) is also bounded and its
image by F convergent, so by Theorem 2.5 it contains a subsequence u
((n))
converging in C
1
d
and therefore vanishing uniformly at innity. Accordingly, there is n
0
IN, and r > 0 such that
((n))
(x) <
0
2
whenever [x[ r an n n
0
. So for n max(r, n
0
) (since (n) n), we have
0
((n))
(x
(n)
) <
0
2
. Contradiction.
(ii) (iii) is evident.
(iii) (i). Let (u
n
) be a bounded sequence from D
p
() such that (F(u
n
)) converges, by
assumption it contains a subsequence (u
(n)
) vanishing uniformly at innity in the sense of C
1
d
.
But this subsequence is also bounded in D
p
() and therefore it contains a subsequence (u
((n))
)
converging weakly to some u in D
p
(). So, by Lemma 2.14, u
((n))
u in C
1
d
(, IR
m
). Hence
F is proper by Theorem 2.5 (ii).
2.4. PROPERNESS 45
Lemma 2.15 Let (u
n
) W
2,p
(), be a sequence converging to zero in W
2,p
(
), for every
bounded and open subset
. Then, given v W
2,p
(), (0, 1) and n
0
IN, there is
n
1
IN, n
1
n
0
, such that for every n n
1
:
(i) |v|
p
2,p,
+|u
n
|
p
2,p,
|v +u
n
|
p
2,p,
|v|
p
2,p,
+|u
n
|
p
2,p,
+, and
(ii) |v +u
n
|
1,,
max (|v|
1,,
, |u
n
|
1,,
) +.
Proof. Let
r
= x : [x[ < r, and
r
= x : [x[ > r. Since v X
p
()
C
1
d
(, IR
m
), there is r > 0 such that
|v|
2,p,
r
and |v|
1,,
r
(2.34)
By assumption, we have u
n
0 in W
2,p
(
r
) and hence also in C
1
(
r
). Thus for n large enough
|u
n
|
2,p,
r
and |u
n
|
1,,
r
. (2.35)
Using the preceding inequalities and [a b[
p
a
p
p(a +b)
p1
b, we get
|v +u
n
|
p
2,p,
[|v|
2,p,
r
|u
n
|
2,p,
r
[
p
+
|v|
2,p,
r
|u
n
|
2,p,
p
|v|
p
2,p,
r
+|u
n
|
p
2,p,
r
2p(M +)
p1
,
where M is a bound for |v|
2,p,
and |u
n
|
2,p,
. We also deduce from (2.34) and (2.35) that
|v|
p
2,p,
|v|
p
2,p,
r
+
p
and |u
n
|
p
2,p,
|u
n
|
p
2,p,
r
+
p
.
Thus
|v|
p
2,p,
+|u
n
|
p
2,p,
2(p(M + 1)
p1
+ 1) |v +u
n
|
p
2,p,
. (2.36)
Analogously, using (2.34) and (2.35) and (a +b)
p
a
p
+p(a +b)
p1
b, we prove that
|v +u
n
|
p
2,p,
|v|
p
2,p,
+|u
n
|
p
2,p,
+ 2p(M + 1)
p1
.
This proves (i) since is arbitrary.
For (ii), we have
|v +u
n
|
1,,
= max
_
|v +u
n
|
1,,
r
, |v +u
n
|
1,,
r
_
max
_
|v|
1,,
r
+, |u
n
|
1,,
r
+
_
= max
_
|v|
1,,
r
, |u
n
|
1,,
r
_
+
max (|v|
1,,
, |u
n
|
1,,
) +.
,=1
A
(x)
2
+
N
=1
B
(x)
+C(x)
be a dierential operator which is strictly Petrovskii-elliptic on the compact subsets of , with
continuous and bounded coecients. Suppose that there is a sequence (u
n
) in D
p
() such that
u
n
0 in D
p
(), Lu
n
0 in Y
p
(), and (u
n
) contains no subsequence converging to 0 in
D
p
(). Then, there is a sequence (w
n
) D
p
() such that w
n
0 in D
p
() and Lw
n
0 in
Y
p
(), (w
n
) contains no subsequence converging to 0 in D
p
(), but furthermore, w
n
0 in
C
1
d
(, IR
m
).
46 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
Proof. For simplicity we denote by |u|
k,p
the norm of u in W
k,p
(, IR
m
).
Since (u
n
) contains no subsequence converging to 0, there exist > 0 and n
0
IN such
that |u
n
|
2,p
for n n
0
. Therefore, at least one component (u
l
n
n
) of (u
n
) veries |u
l
n
n
|
2,p
m
= . Since (l
n
) 1, . . . , m is nite, it contains a constant subsequence l
(n)
= l so that
|u
l
(n)
|
2,p
. In the remainder of the proof, l is xed, and for more simplicity we denote by
u
n
the subsequence u
(n)
.
Note that the hypotheses made about (u
n
) imply by Lemma 2.10 that u
n
0 in W
2,p
(
, IR
m
)
for every open and bounded subset
. Therefore u
j
n
0 in W
2,p
(
) j = 1, . . . , m.
Let
n
be a sequence from (0,1) such that
n=0
n
=
p
.
We construct a sequence (v
n
) in D
p
(), and a subsequence (u
(n)
) verifying v
n+1
= v
n
+
u
(n+1)
. Recall that Lu
n
0 so there is an integer (0) for which |Lu
(0)
|
0,p
0
. Set v
0
=
u
(0)
. In Lemma 2.15 let v = u
j
(0)
, =
1
and n
0
= (0). This produces a integer n
1
(j). Also
there is a n
2
IN such that k n
2
|Lu
k
|
0,p
1
. Set then (1) = maxn
2
, n
1
(j), 1 j
m+1, and v
1
= v
0
+u
(1)
. By induction suppose (v
n
) and (n) already constructed. Let then
v = v
j
n
and =
n+1
and n
0
= (n) in Lemma 2.15. This produces an integer n
1
(j) from which
the estimates of this Lemma hold. Also there is a n
2
IN such that k n
2
|Lu
k
|
0,p
n+1
.
Set then (n + 1) = maxn
2
, n
1
(j) : 1 j m + 1, and v
n+1
= v
n
+ u
(n+1)
. Note that by
construction |Lu
(n)
|
0,p
n
. Note also that the relation dening v
n
, shows by induction that
v
n
D
p
().
By Lemma 2.15 (i), we have
|v
j
k
|
p
2,p
+|u
j
(k+1)
|
p
2,p
k+1
|v
j
k+1
|
p
2,p
|v
j
k
|
p
2,p
+|u
j
(k+1)
|
p
2,p
+
k+1
k IN.
Thus by summation for n 1,
n
k=0
|u
j
(k)
|
p
2,p
n
k=1
k
|v
j
n
|
p
2,p
n
k=0
|u
j
(k)
|
p
2,p
+
n
k=1
k
. (2.37)
Now taking j = l in the above, we get for n n
0
(l and n
0
are dened in the beginning of the
proof) (n n
0
)
p
|v
l
n
|
p
2,p
and therefore
(n n
0
)
1/p
|v
l
n
|
2,p
|v
n
|
2,p
for n n
0
. (2.38)
Let M
j
1 be a bound for |u
j
n
|
2,p
, so that M =
m
j=1
M
j
is a bound of |u
n
|
2,p
. The second
inequality of (2.37) yields: |v
j
n
|
2,p
M
j
(n + 1 +
p
)
1/p
. Therefore
|v
n
|
2,p
M (n + 1 +
p
)
1/p
. (2.39)
From Lemma 2.15 (ii),
|v
j
n+1
|
1,
max|v
j
n
|
1,
, |u
j
(n+1)
|
1,
+
n+1
.
Hence, by induction
|v
j
n
|
1,
max|u
j
(k)
|
1,
: 0 k n +
n
k=1
k
,
and thus
|v
n
|
1,
m(C +
p
) , (2.40)
where C is a bound for |u
n
|
1,
.
2.4. PROPERNESS 47
Next,
|Lv
n
|
0,p
n
k=0
|Lu
(k)
|
0,p
n
k=0
k
.
Therefore,
|Lv
n
|
0,p
p
. (2.41)
Now set w
n
= n
1/p
v
n
. Then rst, by (2.38) |w
n
|
2,p
(1 (n
0
/n))
1/p
so that (w
n
) contains
no subsequence converging to 0 in D
p
(). Secondly, by (2.39) (w
n
) is bounded in D
p
().
Next by (2.40) |w
n
|
1,
const. n
1/p
, whence w
n
0 in C
1
d
(, IR
m
). Lastly by (2.41),
|Lw
n
|
0,p
p
n
1/p
which implies that Lw
n
0 in Y
p
(). That w
n
0 in D
p
() follows from
its boundedness in D
p
() and its convergence to 0 in C
1
d
(, IR
m
).
Theorem 2.6 Let have a C
2
boundary, and L be an elliptic operator as in the preceding
lemma. Then the following statements are equivalent.
(i) L
+
(D
p
(), Y
p
()).
(ii) Every bounded sequence (u
n
) D
p
() converging to zero in C
1
d
(, IR
m
) and such that
Lu
n
0 in Y
p
(), contains a subsequence converging to zero in D
p
()
6
.
Proof. (i) (ii). Recall that a bounded sequence in D
p
() converging to 0 in C
1
d
(, IR
m
), is
weakly convergent to zero in D
p
(), and by Yoods criterion L is proper on the closed bounded
subsets of D
p
(). Then, the result follows from Note C in 1.3.
(ii) (i). It suces to show (by the same note) that if (u
n
) is sequence in D
p
(), such
that u
n
0 in D
p
() and Lu
n
0 in Y
p
(), then u
n
0 in D
p
(). If this is false, then
there is a subsequence (u
(n)
) bounded away from zero in D
p
() (which implies that it contains
no subsequence converging to zero). Hence (u
(n)
) satises the conditions of Lemma 2.16, and
accordingly, there is a sequence (w
n
) having the same properties as (u
(n)
) and furthermore
converging to zero in C
1
d
(, IR
m
). Then by assumption (w
n
) contains a subsequence converging
to zero in D
p
(). But this is impossible since (w
n
) contains no subsequence converging to 0 in
D
p
().
Corollary 2.2 Let have a C
2
boundary. Suppose that every bounded sequence (u
n
)
D
p
() converging to zero in C
1
d
(, IR
m
) and such that F(u
n
) F(0) in Y
p
(), contains a
subsequence converging to zero in D
p
(). (It is so if F is proper on the closed bounded subsets
of D
p
()). Then DF(u)
+
(D
p
(), Y
p
()) for all u D
p
().
Proof. By Theorem 2.3 it suces to show that DF(0)
+
(D
p
(), Y
p
()). According to
Theorem 2.6 with L = DF(0), it is sucient to show that if (u
n
) is a bounded sequence from
D
p
() converging to zero in C
1
d
(, IR
m
) and (DF(0)u
n
) converges to 0 in Y
p
(), then (u
n
)
contains a subsequence converging to 0 in D
p
(). By Lemma 2.13 we have
F(u
n
) F(0) DF(0)u
n
0 in Y
p
().
But DF(0)u
n
0, and therefore F(u
n
) F(0). Hence by assumption (u
n
) contains a subse-
quence converging to 0 in D
p
().
6
which in turn implies that u
n
0 in D
p
(), by Note A3 of the appendix.
48 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
2.5 Operators with asymptotically periodic coecients
In this section, we consider the case where F has a limit operator with periodic coecients
in a sense precised below. Here is unbounded and so K =
B
r
= x IR
N
: [x[ > r,
r
= B
r
and
r
=
B
r
.
Assume that there are two families of matrix valued functions,
a
: IR
N
_
IR
m
IR
mN
_
IR
mm
, 1 , N,
and
c
i
: IR
N
_
IR
m
IR
mN
_
IR
mm
, 0 i N,
both continuous and periodic in x with the same period T, and satisfying
lim
|x|
(x, ) a
(x, )
= 0, (2.42)
lim
|x|
_
1
0
i
b(x, t) dt c
i
(x, )
= 0, (2.43)
the convergence being uniform on the compact subsets of IR
m
IR
mN
. We set
b
(x, ) =
N
i=0
c
i
(x, )
i
, (2.44)
so that b
(x, 0) = 0.
Note that by Remark 2.7, a
and b
are equicontinuous C
0
bundle maps. Now we dene
the limit operator F
by
F
(u) =
N
,=1
a
(., u, u)
2
u +b
(v) F
(0)
_
_
,=1
a
(., 0)
2
v +
N
=0
c
(., 0)
v
_
_
=
N
,=1
_
a
(., v, v) a
(., 0)
_
v +
N
=0
(c
(., v, v) c
(., 0))
v
So it follows from the equicontinuity of a
and c
at = 0, that F
is dierentiable at 0 with
derivative
DF
(0)v =
N
,=1
a
(., 0)
2
v +
N
=0
c
(., 0)
v (2.46)
2.5. OPERATORS WITH ASYMPTOTICALLY PERIODIC COEFFICIENTS 49
Lemma 2.17 Let
r
= x : [x[ > r and B X
p
() be a bounded subset. Then for every
> 0 there is an r > 0 such that for every u B, the following hold.
(i) |F(u) F(0) F
(u)|
0,p,
r
, and
(ii) |DF(0)u DF
(0)u|
0,p,
r
.
Proof. (i) Since B is bounded in X
p
() and therefore in C
1
d
(, IR
m
), there is a compact set
K IR
m
IR
mN
such that (u(x), u(x)) K for every u B and x . Since the limit in
(2.42) is uniform in K, [a
(x, ) a
(x, )[ x
r
K if r is large enough.
Thus
|a
(u) a
(u)|
0,,
r
u B. (2.47)
A similar argument based on (2.43) yields
_
1
0
i
b(x, t) dt c
i
(x, )
r
K
if r is large enough, and thus
[b(x, ) b(x, 0) b
(x, )[
N
i=0
[
i
[.
Therefore for k = 1, . . . , m,
|b
k
(u) b
k
(0) b
,k
(u)|
0,p,
r
N
i=0
|
i
u|
0,p,
r
N
i=0
m
j=1
_
_
i
u
j
_
_
0,p,
r
m(N + 1)|u|
2,p,
u B.
Thus
|b(u) b(0) b
(u)|
0,p,
r
m
2
(N + 1)|u|
2,p,
. (2.48)
With (2.47) we get
|F(u) F(0) F
(u)|
0,p,
r
m
2
(N
2
+N + 1)|u|
2,p,
u B.
Lastly note that is arbitrary and |u|
2,p,
is bounded. Hence the desired result follows.
(ii) The proof is similar. Recall that
DF(0)u DF
(0)u =
N
,=1
_
a
(., 0) a
(., 0)
_
u +
N
=0
(
b(., 0) c
(., 0))
u.
Thus |DF(0)u DF
(0)u|
0,p,
r
m
2
(N
2
+N + 1)|u|
2,p,
.
Corollary 2.3 Let (u
n
) be a bounded sequence from X
p
() such that u
n
0 in X
p
(
) for
every bounded open subset
(u
n
) 0 in Y
p
().
50 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
(ii) (DF(0) DF
(0))u
n
0 in Y
p
().
Proof. (i) Let > be given. Since (u
n
) is bounded in X
p
(), it follows from Lemma
2.17 (i) that for r > 0 large enough we have |F(u
n
) F(0) F
(u
n
)|
0,p,
r
n IN.
Next recall that u
n
0 in X
p
(
r
) by hypotheses and that F and F
(u
n
) 0
in Y
p
(
r
), which means that |F(u
n
) F(0) F
(u
n
)|
0,p,
r
for n large enough. Thus
|F(u
n
) F(0) F
(u
n
)|
0,p,
can be made arbitrary small for n large enough.
(ii) The proof is similar. First by Lemma 2.17 (ii) we have |DF(0)u
n
DF
(0)u
n
|
0,p,
r
.
Next DF(0) L(X
p
(
r
), Y
p
(
r
)) by Theorem 2.2 with =
r
and it is clearly seen from (2.46)
that DF
(0) L(X
p
(
r
), Y
p
(
r
)). Therefore (DF(0) DF
(0)) u
n
0 in Y
p
(
r
). And thus
| (DF(0) DF
(0)) u
n
|
0,p,
can be made arbitrary small for n large enough.
Given h IR
N
and a function f : IR
N
IR, we denote by
h
(f) : IR
N
IR the function
h
(f)(x) = f(x +h)
Corollary 2.4 Let B X
p
() be a bounded subset, and
IR
N
be a bounded open subset.
Then for every > 0, we have |
h
(F(u) F(0))
h
F
(u)|
0,p,
for every u B provided
[h[ is large enough.
Proof. Choose r > 0 as in Lemma 2.17 and increase it if necessary to have
B
r
. Then
for [h[ large enough we have
+ h
B
r
. By the translation invariance of the Lebesgue
measure,
|
h
(F(u) F(0))
h
F
(u)|
0,p,
= |F(u) F(0) F
(u)|
0,p,
+h
.
Now, by Lemma 2.17
|F(u) F(0) F
(u)|
0,p,
+h
|F(u) F(0) F
(u)|
0,p,
B
r
for every u B.
Lemma 2.18 (Shifted subsequence Lemma) Let T = (T
1
, . . . , T
N
) with T
i
> 0. If (u
n
) is
a bounded sequence from X
p
(), then either
(i) (u
n
) vanishes uniformly at innity in the sense of C
1
d
(), or
(ii) there exist a sequence (l
n
) Z
N
with lim
n
[l
n
[ = , a subsequence (u
(n)
), and
a nonzero element u X
p
= W
2,p
_
IR
N
, IR
m
_
such that the sequence u
n
dened by
u
n
(x) = u
(n)
(x +l
n
T) is weakly convergent to u in X
p
(B
k
) for all k IN
.
Proof. First of all the denition of u
n
makes sense. Indeed, the domain of such shifted
subsequence is l
n
T. Let B
R
be a ball containing K =
, and
T = minT
i
. Given k IN
there is n
k
IN
such that [l
n
[ >
k+R
T
for n n
k
. Then for x B
k
, [x +l
n
T[ [l
n
[
T [x[
[l
n
[
.
Now we go to the proof of the alternative. Let Q
0
= (0, T
1
) (0, T
N
). Suppose that (i)
does not hold, so that there is an
0
> 0, a sequence (x
n
) such that [x
n
[ n, and a subsequence
(u
(n)
) such that [u
(n)
(x
n
)[ + [u
(n)
(x
n
)[
0
for all n N. Since IR
N
=
lZ
N
_
Q
0
+lT
_
,
there is z
n
Z
N
for which y
n
= x
n
z
n
T Q
0
. Clearly lim
n
[z
n
[ = . Dene v
n
(x) =
u
(n)
(x +z
n
T).
According to what has been said at the beginning of the proof, for every k IN
there is
n
k
IN
from which v
n
X
p
(B
k
), and furthermore (v
n
)
nn
k
is bounded in X
p
(B
k
). In partic-
ular (v
n
)
nn
1
is bounded in X
p
(B
1
), and so there is a subsequence (v
1
(n)
) converging weakly
to some u
1
X
p
(B
1
). But for n n
2
, (v
1
(n)
) (v
n
) is bounded in X
p
(B
2
), and again there is
a subsequence (v
2
(n)
) (v
1
(n)
) converging weakly to some u
2
X
p
(B
2
). Clearly u
2
[
B
1
= u
1
.
Continuing the process, we construct a sequence of subsequences (v
k
(n)
), each of which con-
verging weakly to u
k
X
p
(B
k
), and furthermore u
k+1
is an extension of u
k
.
Now we dene ( u
n
) as the diagonal subsequence (v
n
(n)
), i.e.
u
n
(x) = u
(
n
(n))
(x +z
n
(n)
T) = u
(n)
(x +l
n
T)
if we set l
n
= z
n
(n)
and (n) = (
n
(n)). On the other hand, we see that there is a function
u : IR
N
IR
m
naturally dened by: u(x) = u
k
(x) if x B
k
.
Now since ( u
n
)
nn
k
is a subsequence of (v
k
(n)
), we have that ( u
n
) converges weakly to u in
X
p
(B
k
) for all k IN
, and so u X
p
.
It remains to show that u ,= 0. Choose k IN
such that Q
0
B
k
. By the compactness
of the embedding W
2,p
(Q
0
, IR
m
) C
1
(Q
0
, IR
m
) we have that u
n
u in C
1
(Q
0
, IR
m
), hence
| u
n
|
1,,Q
0
| u|
1,,Q
0
. But
| u
n
|
1,,Q
0
1
2
_
[ u
n
(y
n
(n)
)[ +[ u
n
(y
n
(n)
)[
_
=
1
2
_
[u
(
n
(n))
(x
n
(n)
)[ + [u
(
n
(n))
(x
n
(n)
)[
_
1
2
0
.
Therefore, | u|
1,,Q
0
1
2
0
, whence u ,= 0.
Theorem 2.7 Let have a C
2
boundary. Suppose that
(i) there is u
0
D
p
() for which DF(u
0
)
+
(D
p
(), Y
p
()), and
(ii) u X
p
[ F
(u) = 0 = 0.
Then F is proper on the closed bounded subsets of D
p
().
Proof. By Corollary 2.1 it suces to show that if (u
n
) is a bounded sequence from D
p
()
and (F(u
n
)) converges to some y in Y
p
(), then (u
n
) vanishes uniformly at innity in the sense
of C
1
d
(). After replacing F by F F(0) and y by y F(0), we can assume that F(0) = 0.
Let us show that the case (ii) of Lemma 2.18 cannot occur. By contradiction suppose there is a
52 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
sequence (l
n
) Z
N
with lim
n
[l
n
[ = and a subsequence (u
(n)
) such that the sequence ( u
n
)
dened by u
n
(x) = u
(n)
(x +l
n
T) has a nonzero weak limit u in X
p
(B
k
).
It is enough to show that F
( u) = 0. Let y
n
be dened by y
n
(x) = y(x+l
n
T) =
l
n
T
(y)(x).
According to the proof of Lemma 2.18, for all k IN
, y
n
X
p
(B
k
) from a certain rank n
k
,
and it is bounded by a constant independent of k.
Let C
0
(IR
N
), and choose k IN
such that B
k
contains the support of . Recall that
u
n
u in X
p
(B
k
) and F
: X
p
(B
k
) Y
p
(B
k
) is weakly continuous, and so F( u
n
) F( u) in
Y
p
(B
k
). Thus
_
IR
N
F
( u) dx =
_
B
k
F
( u) dx = lim
n
_
B
k
F
( u
n
) dx
= lim
n
_
B
k
l
n
T
F
(u
(n)
) dx. (2.50)
On the other hand according to Corollary 2.4 (we assumed that F(0) = 0 ),
l
n
T
F(u
(n)
)
l
n
T
F
(u
(n)
)
n
0 in Y
p
(B
k
). (2.51)
Next for n n
k
, and j = 1, . . . , m, we have
_
_
_
_
l
n
T
F(u
(n)
) y
n
_
j
_
_
_
p
0,p,B
k
=
_
B
k
l
n
T
F
j
(u
(n)
)(x)
l
n
T
y
j
(x)
p
dx
=
_
|zl
n
T|<k
F
j
(u
(n)
)(z) y
j
(z)
p
dz
_
_
F
j
(u
(n)
) y
j
_
_
p
0,p,
n
0 by assumption. (2.52)
Now together equations (2.50), (2.51) and (2.52), give
_
IR
N
F
( u) dx = lim
n
_
B
k
y
n
dx.
But for each component ( y
n
)
j
, j = 1, . . . , m, we have
_
B
k
( y
n
)
j
dx =
_
B
k
(x)y
j
(x +l
n
T) dx =
_
B(l
n
T,k)
(z l
n
T)y
j
(z) dz
_
_
B(l
n
T,k)
[y
j
[
p
_1
p
_
_
B(l
n
T,k)
[(z l
n
T)[
q
dz
_1
q _
1
p
+
1
q
= 1
_
_
_
B(l
n
T,k)
[y
j
[
p
_1
p
_
_
IR
N
[[
q
_1
q
n
0,
because B(l
n
T, k)
B
|l
n
T|k
so that the result follows from Note B3 in the appendix.
Thus, nally
_
IR
N
F
( u) dx = 0 for all C
0
(IR
N
),
2.5. OPERATORS WITH ASYMPTOTICALLY PERIODIC COEFFICIENTS 53
and consequently F
( u) = 0.
The last theorem shows that together the semi-Fredholmness of F and the nonexistence of
nontrivial solutions of the limit problem F
, IR
m
) for every
bounded open subset
IR
N
. In particular u
n
0 in W
k,q
(IR
N
, IR
m
).
Proof. For m = 1, this is Lemma 4.8 of [31]. The conclusion is then clear, since convergence,
respectively weak convergence in W
k,q
(IR
N
, IR
m
), is equivalent to this type of convergence of
each component in W
k,q
(IR
N
).
Theorem 2.8 The following statements are equivalent.
(i) F is proper on the closed bounded subsets of X
p
.
(ii) Every sequence (u
n
) X
p
such that u
n
0 in X
p
and F(u
n
) F(0) in Y
p
, contains a
subsequence converging in X
p
.
(iii) There is u
0
X
p
for which DF(u
0
)
+
(X
p
, Y
p
) and the equation F
(u) = 0 has no
nonzero solution in X
p
.
Proof. (i) (ii) is evident since a weakly convergent sequence is bounded. (iii) (i) is
Theorem 2.7 (with = IR
N
). It remains to prove (ii) (iii). That DF(0) : X
p
Y
p
is semi-
Fredholm follows from Corollary 2.2. Consider now an element u X
p
such that F
(u) = 0.
Set u
n
(x) = u(x + nT) so that F
(u
n
) = 0 by the periodicity of the the coecients of F
.
By Lemma 2.19, u
n
0 in X
p
(
IR
N
, and hence by
Corollary 2.3 (i) F(u
n
) F(0) 0 in Y
p
. Also u
n
0 in X
p
, and therefore by hypotheses (u
n
)
contains a convergent subsequence (u
(n)
). Its limit is necessarily 0 (by the uniqueness of the
weak limit in X
p
). But recall that |u
n
|
2,p,IR
N = |u|
2,p,IR
N, hence u = 0.
Remark 2.10 Note that strict Petrovskii-ellipticity on the compact subsets of IR
N
and strict
ellipticity on IR
N
are equivalent for F
.
On the other hand note that det
_
N
,=1
(x, 0)
_
is a homogeneous polynomial of
order 2m in , so it could be written as
7
P(x, ) =
N
||=2m
p
(x)
. Similarly ,
det
_
_
N
,=1
(x, 0)
_
_
= P
(x, ) =
N
||=2m
p
(x)
.
7
= (
1
, . . . ,
N
) IN
N
is a multi-index
54 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
Thus from (2.42) it follows that given > 0, [p
(x) p
(x, )[ N
2m
[[
2m
. And thus if P(x, ) [[
2m
we have
P
(x, ) ( N
2m
)[[
2m
.
Therefore the strict ellipticity condition (in IR
N
) for a
(., 0).
Note that this reasoning also proves the stability of the ellipticity condition i.e. an elliptic
system remains elliptic after a small enough perturbation of its leading coecients.
2.6 Linear systems
2.6.1 Linear systems on IR
N
Let S = S(IR
N
) denote the Schwartz space of rapidly decreasing functions and S
for all f L
p
(IR
N
) where T
f
is dened by
T
f
, ) =
_
IR
N
fdx for all S.
Multiplication by any function C
(IR
N
) which is slowly increasing (that is, and all its
partial derivatives have at most polynomial growth at innity) denes a continuous mapping of
S into itself and consequently a multiplication in S
, denoted by () : S
, can be dened
through
()u, ) = u, ) for all u S
and S,
and is a continuous mapping of S
N
2
_
IR
N
e
ix
(x) dx
which is a continuous bijection of S onto itself, induces a continuous bijection T : S
by
Tu, ) = u, T) for all u S
and S.
Recall also that (see for instance [38] Theorem 2.3.3. p. 177), for 1 < p < ,
W
2,p
(IR
N
) =
_
g L
p
(IR
N
)
T
1
(1 +[[
2
)TT
g
= T
f
for some f L
p
(IR
N
)
_
and that there is a constant K > 0 such that
|g|
2,p
K|f|
0,p
.
A function : IR
N
l C is called a multiplier in L
p
if u T
1
()Tu denes a continuous
linear map from L
p
(IR
N
) into itself. As a special case of a result due to Miklin (Theorem
2, Appendix [27]) we can formulate the following sucient condition for a function to be a
multiplier.
(M) C
(IR
N
) is a slowly increasing function and there exists a constant M > 0 such that,
for all k 0, 1, ..., N,
[[
k
k
()
j
1
...
j
k
M for all IR
N
,
2.6. LINEAR SYSTEMS 55
where 1 j
1
< j
2
< < j
k
N.
If satises (M), then, for 1 < p < , there is a constant C
p
> 0 such that
_
T
1
()TT
f
,
_
C
p
|f|
0,p
||
0,p
for all f L
p
(IR
N
) and S,
where
1
p
+
1
p
(f) L
p
(IR
N
) such that
_
T
1
()TT
f
,
_
=
_
IR
N
(f)
and
L(L
p
(IR
N
), L
p
(IR
N
)).
We can now formulate a condition ensuring that
L(L
p
(IR
N
), W
2,p
(IR
N
)).
(W) Let and satisfy the condition (M) where () = (1 +[[
2
)().
If (W) is satised by and f L
p
(IR
N
), we have that T
(f)
= T
1
()TT
f
S
and
T
1
(1 +[[
2
)TT
(f)
= T
1
(1 +[[
2
)()TT
f
= T
(f)
where
(f) L
p
(IR
N
) since also satises (M). Thus
(f) W
2,p
(IR
N
) and there exists a
constant K > 0 such that
|
(f)|
2,p
K|
(f)|
0,p
K
1
|f|
0,p
for all f L
p
(IR
N
),
showing that
L(L
p
(IR
N
), W
2,p
(IR
N
)) for 1 < p < .
All the above discussion extends to vector valued functions and distributions in an obvious
way.
For a second order linear mm system with constant coecients
Lu =
N
,=1
A
u +
N
=1
B
u +Cu (2.53)
where A
, B
,=1
+i
N
=1
+C.
Note the the characteristic polynomial of the formal adjoint operator
L
t
u =
N
,=1
A
T
u
N
=1
B
T
u +C
T
u
is the matrix S()
= S()
T
where T denotes the transpose of a matrix.
(S) There exist constants > 0 and > 0 such that
[det S()[ [[
2m
+ for all IR
N
.
56 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
Then, since
S()
1
=
1
det S()
[Cof S()]
T
,
and the elements of the cofactor matrix Cof S() are polynomials in of degree at most 2m2,
it follows that all the elements of the matrix S()
1
satisfy the condition (W) when (S) holds.
Recall that L acts on S
u, L
t
_
for all u S
and S.
Furthermore, TLu = S()Tu for all u S
and so
TL
t
u = S()
Tu.
Theorem 2.9 Let L be a linear dierential operator of the form (2.53) whose characteristic
polynomial satises (S). Then L : W
2,p
(IR
N
, IR
m
) L
p
(IR
N
, IR
m
) is an isomorphism.
Proof. Let f L
p
(IR
N
, IR
m
). The equation Lu = T
f
in S
is equivalent to
TLu, ) = TT
f
, ) for all S
S()Tu, ) = TT
f
, ) for all S
u, ) =
_
T
1
S()
1
TT
f
,
_
for all S.
Hence w := T
1
S()
1
TT
f
S
of the equation Lu = T
f
.
Since the elements of the matrix S()
1
satisfy the condition (W) it follows that there exists
g W
2,p
(IR
N
, IR
m
) such that T
g
= w. Thus, for all S,
_
IR
N
(Lg) dx =
_
IR
N
(L
t
) g dx =
T
g
, L
t
_
=
w, L
t
_
= Lw, ) = T
f
, ) =
_
IR
N
f dx,
showing that Lg = f.
Lemma 2.20 Let L be a linear dierential operator of the form (2.53) that is elliptic in the
sense of Petrovskii. Then its characteristic polynomial satises (S) if and only if det S() ,= 0
for all IR
N
.
Proof. If there exists a point IR
N
such that det S() = 0 then S() clearly cannot satisfy
(S).
If L is elliptic, it follows from (2.16) that there exists > 0 such that
det
_
_
N
,=1
_
_
[[
2m
for all IR
N
.
We claim that there exists a constant R > 0 such that
det
_
[[
2
S()
_
2
for all IR
N
with [[ R.
If not, there exists a sequence
_
k
_
IR
N
such that
[
k
[ and
det
_
[
k
[
2
S(
k
)
_
<
2
.
Setting
k
=
k
[
k
[
and passing to a subsequence, we have that
k
where [[ = 1, and
2.6. LINEAR SYSTEMS 57
1
[
k
[
2
S(
k
) =
N
,=1
+i
N
=1
[
k
[
B
+
1
[
k
[
2
C
N
,=1
.
By the continuity of the determinant this implies that
det
_
_
N
,=1
_
_
2
=
2
[[
2m
,
contradicting the choice of . It follows that, for all IR
N
with [[ R,
[det S()[
2
[[
2m
4
[[
2m
+
R
2m
4
.
If det S() ,= 0 for all IR
N
, there exists > 0 such that [det S()[ for all IR
N
with
[[ R. Thus, setting = /(2R
2m
), we have that
[det S()[
2
+R
2m
[[
2m
+
2
for all IR
N
with [[ R, showing that S() satises (S).
Corollary 2.5 Let L be a linear dierential operator of the form (2.53) that is elliptic in the
sense of Petrovskii and such that the determinant of its characteristic polynomial has no zeros
in IR
N
. Then L : W
2,p
(IR
N
, IR
m
) L
p
(IR
N
, IR
m
) is an isomorphism.
Finally we consider the case of linear systems with variable coecients.
Lu =
N
,=1
A
(x)
2
u +
N
=1
B
(x)
u +C(x)u (2.54)
where
(C) A
, B
, B
and C
such that
A
(x) A
, B
(x) B
and C(x) C
as [x[ .
Let
L
u =
N
,=1
A
u +
N
=1
B
u +C
u
and let S
() ,= 0 for
all IR
N
. Then L
0
(X
p
, Y
p
).
Proof. Our assumptions ensure that L
: X
p
Y
p
is an isomorphism. Furthermore L
t
: X
p
Y
p
is a bounded linear operator for
all t [0, 1].
We now consider T = L L
B
r
for all n. By Lemma 2.10, we have that u
n
0 in
X
p
(B
r
) and so Tu
n
0 in Y
p
(B
r
). Thus we may conclude that Tu
n
0 in Y
p
, showing that
T is compact modulo L
t
. Since L
: X
p
Y
p
is an isomorphism, L
0
= L
0
(X
p
, Y
p
). The
conclusion now follows from Lemma 2.11 (iii) since L = L
1
.
As already noticed in 2.2, all operators on the segment joining two operators which are strongly
elliptic in the sense of Legendre-Hadamard, are also elliptic. Therefore if this new condition is
used, then the assumption L
t
= tL+(1t)L
() ,= 0 for all IR
N
. Then L
0
(X
p
, Y
p
).
Proof. Letting [x[ in (2.17), we see that L
, i.e. L
L(D
p
(), Y
p
()),
and indeed L L(X
p
, Y
p
).
Then what is the connection between the Fredholmness of L and L
+
(D
p
(), Y
p
()) L
+
(X
p
, Y
p
).
Proof. (i) First let L
+
(D
p
(), Y
p
()). To prove the semi-Fredholmness of L, consider
a sequence (v
n
) from X
p
weakly convergent to 0, and such that Lv
n
0 in Y
p
. We show that
v
n
0 in X
p
.
Let B
r
be a ball containing (and consequently also K =
), R > r , and C
0
(IR
N
)
be such that = 1 outside B
R
, = 0 on B
r
. Then consider the restriction to of (v
n
) that
we denote by (u
n
). We have clearly (u
n
) D
p
(), and furthermore:
u
n
0 in D
p
(). Indeed v
n
0 in Y
p
so v
n
0 in Y
p
(because if f L
p
, then also
f L
p
, so that
_
fv
n
0 f L
p
). For 1 , N,
(v
n
) = (
)v
n
+
v
n
, and
so for the same reasons as above, we have
(v
n
) 0 in Y
p
. Next since
(v
n
) = (
2
)v
n
+ (
v
n
+ (
v
n
+
2
v
n
,
we have
2
(v
n
) 0 in Y
p
.
Now weak convergence to 0 in W
2,p
of a sequence is equivalent to the weak convergence to 0 in
L
p
of all its derivatives up to order 2 (see Note C in the appendix). Thus v
n
0 in X
p
and
so also in D
p
().
Since u
n
= v
n
outside B
R
, we have that L
u
n
0 in Y
p
(
B
R
). But a direct calculation
already used in the proof of Lemma 2.10 shows that
L
u
n
=
Lv
n
+
,
_
_
v
n
+
,
(
v
n
+
,
(
v
n
+
) v
n
.
2.7. EXPONENTIAL DECAY 59
Recalling that the imbedding W
2,p
(IR
N
) W
1,p
(
R
) is compact, we see that v
n
0 in
W
1,p
(
R
, IR
m
). This fact together with the boundedness of , its derivatives and of the coe-
cients of L, implies that L
u
n
0 in Y
p
(
R
). So in fact L
u
n
0 in Y
p
().
Now by hypothesis L
+
(D
p
(), Y
p
()), therefore u
n
0 in D
p
(). Next as already
observed v
n
= u
n
outside B
R
, so v
n
0 in X
p
(
B
R
). And by Lemma 2.10 (with = IR
N
)
v
n
0 in X
p
(B
R
). Thus nally v
n
0 in X
p
, and the proof of the rst part is complete.
(ii) Let now L
+
(X
p
, Y
p
), and (u
n
) be a sequence from D
p
() converging weakly to 0,
and L
u
n
converging to 0 in Y
p
() (by note A4 of the appendix, weak convergence in D
p
() is
the same as weak convergence in X
p
()). Take the same r, R and as in the proof of (i), and
dene the sequence (v
n
) by v
n
(x) = (x)u
n
(x) if x and 0 elsewhere.
Now one can check that v
n
X
p
, and furthermore as in the proof of the rst part, that
v
n
0 in X
p
.
On the other hand, by construction u
n
= v
n
outside B
R
, therefore L
u
n
0 in Y
p
(
B
R
) and
since
Lv
n
[
= L
u
n
+
,
_
_
u
n
+
,
(
u
n
+
,
(
u
n
+
) u
n
,
we deduce as in the proof of the rst part that Lv
n
0 in Y
p
(r < [x[ < R). Next, v
n
= 0 on
B
r
, so Lv
n
0 in Y
p
(B
r
). Therefore Lv
n
0 in X
p
.
The semi-Fredholmness of L implies that v
n
0 in X
p
, and so u
n
0 in X
p
(
B
R
). By
Lemma 2.10, u
n
0 in X
p
(
R
) . Thus nally u
n
0 in X
p
() and the proof is complete.
2.7 Exponential decay
Exponential decay of solutions is an important question in the eld of partial dierential equa-
tions. Consider the operator F in (1.2), and let f Y
p
() decays exponentially at innity
(in particular this happens when f = 0), then, does all possible solutions of F(u) = f have
exponential decay? In our context, this question is related to the properness issue. In [34],
by establishing exponential decay of possible solutions of quasilinear elliptic equations on IR
N
,
Rabier and Stuart prove identities of Pohozaev type for the limit problem, and in some sit-
uations, this implies that the limit problem has only the trivial solution. If in addition the
operator is Fredholm, then Theorem 2.7 ensures that F is proper on the closed bounded subset
of W
2,p
(IR
N
). But as we shall see, this question has also another strong connection with the
Fredholm property.
In a recent paper [30], Rabier noticed the following. For u L
p
(IR
N
), the intuitive idea that
u(x) decays exponentially as [x[ is usually captured by the condition that |u|
L
p
(|x|>r)
=
O(e
sr
) for some s > 0 as r , and this happens in particular if u = e
s|x|
v for some
v L
p
(IR
N
) and s > 0. The remark that multiplication by e
s|x|
generates a semigroup on
L(L
p
(IR
N
)) leads Rabier to study the problem in an abstract setting. We briey describe his
results.
Recall rst that given a Banach space X, a C
0
or strongly continuous semigroup of bounded
linear operators on X is a family (T(t))
t0
L(X) satisfying
(i) T(0) is the identity operator on X.
(ii) T(s +t) = T(t)T(s) for every t, s 0.
(iii) lim
t0
T(t)x = x for every x X.
60 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
The uniform boundedness principle then implies that the function t |T(t)|
L(X)
is bounded
on the bounded subsets of [0, ). In turn, this fact implies that, for every x X, the map
t T(t)x is continuous from [0, ) into X (see [28]).
Now let X and Y be two reexive Banach spaces. Assume that there are two injective C
0
semigroups (T(s))
s0
L(X) and (S(s))
s0
L(Y ). Let L L(X, Y ) be a Fredholm operator
such that for some > 0, we have
rge LT(s) rge S(s), s [0, ] (2.55)
S(s)
1
LT(s) L(X, Y ) s [0, ] (2.56)
lim
s0
|S(s)
1
LT(s) L|
L(X,Y )
= 0. (2.57)
Under the above assumptions Rabier proves the following theorem.
Theorem 2.12 There is s
0
(0, ], such that the following property holds. If f = S(s)g
for some g Y , and s > 0, and if u X satises Lu = f, then there is v X such that
u = T(min(s, s
0
))v.
The proof is given in three steps. First the result is proved when L is surjective, and in this case
the index of L is nonnegative. The second step treats the case when L is injective (and then
the index is nonpositive). Finally, the general case is reduced to one of the previews situations
(according to the sign of the index) by adding a nite dimensional operator, and using the
following lemma.
Lemma 2.21 Let E L(X, Y ) have nite rank. Then the mappings S(.)E : [0, ) L(X, Y )
and ET(.) : [0, ) L(X, Y ) are continuous.
Indeed the proof becomes simpler if the index is zero, since by adding a nite dimensional
operator, the problem is reduced to the case when L is an isomorphism.
Now Theorem 2.12 leaves us with the following question. Suppose that s > 0 is xed, but
g, and the possible solutions u of Lu = S(s)g are allowed to vary. Then how does the element
v behave with respect to u and g? Our next proposition answers the question in the zero index
case, and reproves by the way the existence of v. Note that since T(min(s, s
0
)) is injective by
hypothesis, v is uniquely determined by u (for each g).
Proposition 2.1 Let L be Fredholm of index zero in Theorem 2.12. Suppose that s > 0 is
xed, and S(s)g rge L. Then the following hold.
(i) For each g, the correspondence u v is an ane continuous map from L
1
(S(s)g) to X.
(ii) In particular, when g = 0, the map u v is linear and continuous from ker L to X.
(iii) Let g vary in a bounded subset B of Y , and u vary in a bounded subset of
_
gB
L
1
(S(s)g).
Then, v also varies in a bounded subset of X.
Proof. Since L is Fredholm of index zero, there is a nite rank operator E L(X, Y ) such
that L + E GL(X, Y ). It follows from (2.55) that LT(t) + S(t)ET(t) rge S(t) for all
t [0, ], therefore the operator
L
t
= S(t)
1
(L +S(t)E) T(t) (2.58)
is well dened. Indeed, L
t
= S(t)
1
LT(t) + ET(t), and it follows from (2.57) and lemma 2.21
that L
t
L(X, Y ) and
lim
t0
|L
t
(L +E)|
L(X,Y )
= 0. (2.59)
2.7. EXPONENTIAL DECAY 61
Since GL(X, Y ) is open in L(X, Y ), there is s
0
(0, ] such that both L
t
and L+S(t)E are in
GL(X, Y ) for all t [0, s
0
]. Take in particular t = min(s
0
, s) and let
v = L
1
t
(S(s t)g +Eu) , (2.60)
where Lu = S(s)g. Then
(L +S(t)E)T(t)v = S(t)S(s t)g +S(t)Eu = S(s)g +S(t)Eu = Lu +S(t)Eu,
which yields
(L +S(t)E)(T(t)v u) = 0.
Since L +S(t)E GL(X, Y ), we have u = T(t)v.
Therefore (i) and (ii) follow from (2.60), which also yields
|v|
X
|L
1
t
|
L(Y,X)
_
|S(s t)|
L(Y )
|g|
Y
+|E|
L(X,Y )
|u|
X
_
. (2.61)
Thus, (iii) follows from the boundedness of |g| and |u|.
Before we apply the preceding results to elliptic systems of second order, we extend Lemma
2.21 to the case when E is compact.
Lemma 2.22 Let E L(X, Y ) be compact. Then the mappings S(.)E : [0, ) L(X, Y )
and ET(.) : [0, ) L(X, Y ) are continuous.
Proof. The result can be deduced from Lemma 2.21 by approximating E with nite-rank
operators. But we give a direct proof.
(i) We prove the continuity of t S(t)E. Let t 0 and > 0 be given. Consider a bounded
neighborhood I of t in [0, ), and let M > 0 be a bound of |T(s)| for s I. Let B denote
the closed unit ball in X. The set E(B) is relatively compact in Y and therefore it is totally
bounded. Accordingly, there exist u
1
, . . . , u
k
B such that E(B) is covered by the open balls
centered at Eu
j
with radius
3M
. It follows from the strong continuity of S(s) at Eu
j
that
|S(s)Eu
j
S(t)Eu
j
|
Y
3
,
for every s I close enough to t, and every j = 1, . . . , k. Now, for every u B there is some j
such that
|Eu Eu
j
|
Y
<
3M
.
Then,
|S(s)Eu S(t)Eu| |S(s)Eu S(s)Eu
j
| +|S(s)Eu
j
S(t)Eu
j
| +|S(t)Eu
j
S(t)Eu|
|S(s)||Eu Eu
j
| +|S(s)Eu
j
S(t)Eu
j
| +|S(t)||Eu
j
Eu|
M
3M
+
3
+M
3M
= .
Since this holds for every u B, we have |S(s)ES(t)E|
L(X,Y )
, and the proof is complete.
(ii) Let E
and T
is a C
0
semigroup on X
. Now,
|ET(s) ET(t)|
L(X,Y )
= |T
(s)E
(t)E
|
L(Y
,X
)
.
But E
,=1
A
(x)
2
v +
N
=1
B
(x)
v +C(x)v
with its coecients being continuous and bounded on .
We already know that L L(X
p
(), Y
p
()).
Let B
r
be a ball containing (and consequently also K =
), R > r, and C
0
(IR
N
)
be such that = 1 outside B
R
, = 0 on B
r
. Set (x) = (x)[x[. Then C
(IR
N
), (x) = [x[
when [x[ R, and (x) = 0 when [x[ r.
Furthermore for x ,= 0
x
i
=
x
i
[x[
+
i
[x[ (2.62)
x
j
x
i
=
_
ij
[x[
x
i
x
j
[x[
3
_
+
j
x
i
[x[
+
i
x
j
[x[
+
2
ij
[x[ (2.63)
When [x[ r, = 0 and when [x[ > R, = 1 so its derivatives are zero. Therefore from (2.62)
and (2.63), we deduce that all the derivatives of are bounded on IR
N
.
Dene for every s 0 and u Y
p
(), S(s)u := e
s
u. Let T(s) be the restriction of S(s)
to X
p
(). Now we check that T and S satisfy all the assumptions of Theorem 2.12.
We clearly have that e
s
C
(IR
N
) and is bounded by 1 (because 0), therefore
S(s) maps Y
p
() into itself, T(s) maps X
p
() into itself, and D
p
() into itself. T and S clearly
satisfy conditions (i) and (ii) in the denition of a semigroup. They are also injective since
e
s
> 0.
We check condition (iii). Let u Y
p
() be xed, then it follows from the Lebesgue dominated
convergence that
lim
s0
e
s
u = u in Y
p
() (2.64)
Therefore S is a C
0
semigroup on L(Y
p
()).
But for the same reason when u X
p
()
lim
s0
e
s
u =
u in Y
p
() (2.65)
lim
s0
e
s
u =
2
u in Y
p
(). (2.66)
for all 1 , N. Now a direct calculation (in Y
p
()) shows that
(e
s
u) e
s
u = s
e
s
u, (2.67)
(e
s
u) e
s
u = se
s
_
u +
u +
2
u s
u
_
. (2.68)
Therefore
lim
s0
_
(e
s
u) e
s
u
_
= 0 (2.69)
lim
s0
_
(e
s
u) e
s
u
_
= 0. (2.70)
Now, recalling (2.65) and (2.66), we see that
(e
s
u)
u, and
2
(e
s
u)
2
u in
Y
p
() when s 0. All this means that e
s
u u in X
p
(), that is lim
s0
T(s)u = u in X
p
().
This proves the strong continuity the semigroups T and S.
2.7. EXPONENTIAL DECAY 63
Now, we prove that condition (2.55) is satised. Let u X
p
(), from (2.67) we see that
B
(e
s
u) = e
s
where
is some function in Y
p
() (because B
is bounded). From
(2.68) we see that A
(e
s
u) = e
s
where
Y
p
(). Hence
L(e
s
u) = e
s
_
_
N
,=1
+
N
=1
+Cu
_
_
.
That is LT(s)u = L(e
s
u) = e
s
g = S(s)g for some g in Y
p
(), and every s 0.
We prove (2.56) and (2.57). Again using (2.67) an (2.68) we see that for all u X
p
()
|S(s)
1
LT(s)u Lu|
0,p,
sM|u|
2,p,
(2.71)
where M is a suitable bound of the coecients A
, B
b(., 0)
bounded, the number t in Proposition 2.3 can be chosen independently of u.
Proof. We begin with some notation. For a scalar equicontinuous C
1
bundle map h :
IR
m
IR
mN
IR, we know from the proof of Lemma 2.3 that h can be written in the
form (see (2.2))
h(x, ) = h(x, 0) +
N
=0
__
1
0
h(x, t) dt
_
. (2.75)
If now h takes values in IR
m
, applying the above equality to each component h
k
and letting
h
k
, then (2.75) also holds for h in this case. We make one
more step in the matrix-vector notation. If h takes values in IR
mm
, we apply what has been
said to each column h
j
. Let then
h
j
(x, ). Then (2.75) still holds for h in this case as well.
Now we go to the proof of the Theorem. Since b is an equicontinuous C
1
=0
c
(x, )
.
But b is supposed to be C
2
is an equicontinuous C
1
bundle map.
According to what have been said, there is a family of equicontinuous C
0
bundle map c
such
that
c
(x, ) = c
(x, 0) +
N
=0
c
(x, )
.
Note that b
(., 0) = c
(., 0), and condition (2.13) implies that they are bounded functions.
Now, again, we can write
a
(x, ) = a
(x, 0) +
N
=0
a
(x, )
,
where a
are equicontinuous C
0
bundle maps.
Then
F(u) =
N
,=1
_
a
(0) +
N
=0
a
(u)
u
_
u +
N
=0
_
c
(0) +
N
=0
c
(u)
u
_
u,
and
G(0)u = DF(0)u =
N
,=1
a
(0)
2
u +
N
=0
b
(0)
u.
Therefore, since b
(0) = c
(0), we get
G(0)u f = G(0)u F(u)
=
N
,=1
N
=0
_
a
(u)
u
_
u
N
=0
N
=0
_
c
(u)
u
_
u. (2.76)
Now, let s
0
be the number given by Theorem 2.12, for the linear operator G(0). Indeed it
suces to prove the result, when s s
0
, for if s > s
0
, then the condition e
s
f Y
p
(), implies
that e
s
0
f Y
p
().
2.7. EXPONENTIAL DECAY 65
Proposition 2.3, already ensures that u = e
t
v, for some t (depending on u) and some
v D
p
(). Recalling (2.67) and (2.68), we see that the the derivatives of u have also an
exponential decay i.e. they have the form e
t
y where y Y
p
(). Therefore, replacing u by
e
t
v in the right hand side of (2.76), we see that we get the factor e
t
e
t
= e
2t
, and what
is left is a function we call g
0
, i.e.
G(0)u f = e
2t
g
0
.
We claim that g
0
Y
p
(). First, conditions (2.11) and (2.13) imply that a
(., 0), c
(., 0) are
bounded. The assumption that
2
(u), and c
and in L
p
, and the second derivatives are in L
p
. Therefore,
g
0
is a sum of functions in Y
p
(), and the claim is proved.
Consequently, we can write
G(0)u = f +e
2t
g
0
= e
min(s,2t)
_
e
min(s,2t)
f +e
(2t+min(s,2t))
g
0
_
.
Now, [e
min(s,2t)(x)
f(x)[ [e
s(x)
f(x)[ for every x and so e
min(s,2t)
f Y
p
(). And similarly,
e
(2t+min(s,2t))
g
0
Y
p
().
As a result, Theorem 2.12 for G(0) yields an element v
1
D
p
(), such that
u = e
min(s
0
,min(s,2t))
v
1
= e
min(s,2t)
v
1
,
since min(s, 2t) s s
0
by assumption.
If s 2t, then u = e
s
v
1
and we are done. If not, replace u by e
2t
v
1
in (2.76), to get
G(0)u = f +e
4t
g
1
= e
min(s,4t)
_
e
min(s,4t)
f +e
(4t+min(s,4t))
g
0
_
.
Then, again, Theorem 2.12 yields an element v
2
D
p
() such that
u = e
min(s,4t)
v
2
.
If s 4t, then u = e
s
v
2
and this is the claimed result. If not, then u = e
4t
v
2
, and replacing
this value of u in (2.76), leads again to an alternative. But this procedure ends in a nite
number k of steps (depending on u), for which s 2
k
u, and we have
u = e
s
v
k
,
for some v
k
D
p
().
Proposition 2.4 S(s)
1
G(u)T(s) converges to G(u) as s 0, uniformly with respect to u in
bounded subsets.
Proof.
G(u)T(s)v = G(u)(e
s
v) =
N
,=1
a
(u)
2
(e
s
v) +
N
=0
c
(u)
(e
s
v)
=
N
,=1
a
(u)
_
se
s
_
v +
v +
2
v s
v
_
+e
s
v
_
+
N
=1
c
(u)
_
s
e
s
v +e
s
v
_
+c
0
(u)e
s
v
66 CHAPTER 2. FREDHOLM AND PROPERNESS PROPERTIES
Therefore
S(s)
1
G(u)T(s)v G(u)v = s
N
,=1
a
(u)
_
v +
v +
2
v s
v
_
s
N
=1
c
(u)(
v).
Let u belong to a bounded subset of X
p
(). It follows that there is a positive constant M such
that
|S(s)
1
G(u)T(s)v G(u)v|
0,p,
sM|v|
2,p,
.
Chapter 3
Existence from a priori bounds
3.1 A general result
Let be an open subset of IR
N
with a C
2
bounded boundary. Consider the quasilinear operator
F dened by (1.2), under the assumptions (2.10)(2.13) and (2.18). The purpose of this section
is to give a condition ensuring that for every h Y
p
(), the equation F(u) = h has at least one
solution u D
p
(). To simplify the writing and since no other domain than is involved in
this part, we write D
p
and Y
p
instead of D
p
() and Y
p
().
In order to use the degree for Fredholm maps dened in [29], DF(u) must be invertible at
some point u
0
D
p
. Otherwise F has no base point and so the absolute degree [ deg [(F, B, h)
(when dened) is always zero, and therefore no conclusion about existence can be derived using
the degree in the usual way. But in our setting this ensures that F is Fredholm of index zero
(Theorem 2.3). It is not restrictive then to assume that u
0
= 0, and F(u
0
) = F(0) = 0. Indeed
set
F(u) = F(u + u
0
) F(u
0
). Then the properness of F is equivalent to the properness of
F. Also D
F(u) = DF(u + u
0
) so the Fredholmness of F is equivalent to the Fredholmness (of
the same index) of
F . And of course the solvability of the equation F(u) = h for every h is
equivalent to the solvability of
F(u) = h for every h.
We assume that
DF(0) GL(D
p
, Y
p
).
F is proper on the closed bounded subsets of D
p
. This happens for instance if is
unbounded and F has a periodic limit operator F
such that u X
p
(IR
N
); F
(u) = 0 = 0.
But F may be proper for dierent reasons
1
.
We consider the homotopy H dened by
H(t, u) =
_
_
_
1
t
F(tu) th if t ,= 0
DF(0)u if t = 0
(3.1)
Lemma 3.1 H has the following properties.
1. For every t IR, H(t, .) C
1
(D
p
, Y
p
), and H C
1
(IR
D
p
, Y
p
).
2. D
u
H(t, u) = DF(tu)
0
(D
p
, Y
p
), (t, u) IR D
p
.
3. For every bounded B D
p
, the collection (H(., u))
uB
is equicontinuous.
4. The restriction of H to any closed bounded subset of IR D
p
is proper.
1
Theorem 2.8 states that when = IR
N
and F has a limit operator, then the properness of F implies the
nonexistence of nontrivial solutions of F
D
p
and this is equivalent to the second statement
of point 1.
3. The equicontinuity at t = 0 is a consequence of the dierentiability of F at 0. Indeed
G(t, u) G(0, u) =
1
t
F(tu) DF(0)u. Now for every > 0 there is > 0 such that
|F(v) DF(0)v|
p
|v|
2,p
if |v|
2,p
.
Therefore if M > 0 is a bound for u B, and [t[ /M, we have |G(t, u) G(0, u)|
p
M.
Let now t
0
,= 0, then
G(t, u) G(t
0
, u) =
1
t
F(tu)
1
t
0
F(t
0
u)
=
N
,=1
_
a
(tu) a
(t
0
u)
_
u +
N
=0
_
c
(tu) c
(t
0
u)
_
u
where c
(x, ) =
_
1
0
b(x, ) d.
Let M > 0 be a bound for u B, so in particular [u(x)[ and [u(x)[ are bounded by M for
every u B and every x . Now by Lemma 2.1 (i), (a
(x, .))
x
is uniformly equicontinuous
on the bounded subsets of IR
m
IR
mN
. Accordingly for every > 0, there is > 0 such that
[a
(x, t
0
u(x), t
0
u(x))[ if [t t
0
[ /M, that is
|a
(tu) a
(t
0
u)|
.
On the other hand using the equicontinuity for c
, we see that |c
(tu)c
(t
0
u)|
if [tt
0
[
is suciently small. Therefore for [t t
0
[ suciently small and all u B, we have
|G(t, u) G(t
0
, u)|
p
m
2
M(N
2
+N + 1).
Since is arbitrary, this inequality means that (G(., u))
uB
is equicontinuous at t
0
,= 0. Together
with point 1., this also shows that G C(IR D
p
, Y
p
).
4. We begin by proving the properness of G(t, .). If t = 0 then G(0, .) is an isomorphisms and
so it is proper. If t ,= 0, let (u
n
) D
p
be a bounded sequence such that (G(t, u
n
)) converges.
Then (F(tu
n
)) converges. But (tu
n
) is bounded, so the properness of F implies that it has a
convergent subsequence. Since t ,= 0, (u
n
) has a convergent subsequence.
Since for every bounded subset B D
p
, (G(., u))
uB
is equicontinuous, the conclusion
follows from Lemma 1.1.
Remark 3.1 H need not be dierentiable at zero. Take for example F(u) = u + [u[
1/2
u,
then
1
t
F(tu) = u +[t[
1/2
[u[
1/2
u. Indeed this situation does not occur if more regularity on
F is assumed. For instance if F is C
2
, it follows from the Taylor expansion of F about zero
3.1. A GENERAL RESULT 69
that
1
t
2
F(tu)
1
t
DF(0)u
1
2
D
2
F(0)(u, u) as t 0 (uniformly with respect to u on bounded
subsets), and so D
t
G(0, u) =
1
2
D
2
F(0)(u, u). On the other hand as t 0,
D
t
G(t, u) =
1
t
2
F(tu) +
1
t
DF(0)u
1
t
DF(0)u +
1
t
DF(tu)u
1
2
D
2
F(0)(u, u) + D
2
F(0)(u, u) =
1
2
D
2
F(0)(u, u).
Therefore G is C
1
in this case.
Now since F is only C
1
, H[
[0,1]
need not be an admissible homotopy in the sense of Denition
4.2 of [29]. However it is admissible in the sense of Benevieri-Furi.
Proposition 3.1 Suppose that there is R > 0 such that
|u|
2,p
R whenever H(t, u) = 0 for some t [0, 1]. (E)
Then F(u) = h is solvable in D
p
.
Proof. Let B D
p
be the ball of center 0 and radius R + 1. The condition (E) ensures that
0 / H([0, 1], B). Next by Lemma 3.1 (2.), D
u
H(t, 0) = DF(0) GL(D
p
, Y
p
), i.e. 0 is a base
point of H(t, .) for all t IR. Therefore, using Theorem 1.2, we get
deg
0
(F h, B, 0) = deg
0
(H(1, .)), B, 0)
= deg
0
(H(0, .), B, 0)
= deg
0
(DF(0), B, 0)
= 1.
Hence the existence of an element u D
p
such that F(u) h = 0.
Remark 3.2 We can still prove Proposition 3.1 without appealing to Theorem 1.2 (which uses
the Benevieri-Furi degree). Indeed, note rst that, as a consequence of Lemma 3.1, H[
[,1]
B
is
B-admissible for every (0, 1) and every bounded subset B D
p
. In particular, taking B as
in the rst proof above, we get
deg
0
(H(1, .), B, 0) = deg
0
(H(, .), B, 0).
So let us calculate deg
0
(H(, .), B, 0) for small enough.
From the inverse function theorem, there is a neighborhood U
0
of zero in D
p
and a neigh-
borhood V
0
of zero in Y
p
such that F : U
0
V
0
is a dieomorphism. Now choose
0
,
1
(0, 1)
small enough so that
2
0
h V
0
, and
1
B U
0
, and let = min(
0
,
1
). Then there exists u
0
U
0
for which F(u
0
) =
2
h. Letting then u
1
=
1
u
0
, this is equivalent to H(, u
1
) = 0. Furthermore
u
1
is the unique solution of H(, u) = 0 (for this choice of indeed). For if u
2
is another solution,
then rstly, F(u
2
) =
2
h, and secondly, u
2
B by (E) and so u
2
U
0
. But F[
U
0
is injective,
so u
1
= u
2
and therefore u
1
= u
2
.
Now since F : U
0
V
0
is a dieomorphism, DF(u) GL(D
p
, Y
p
) for all u U
0
, thus
D
u
H(, u) = DF(u) GL(D
p
, Y
p
) for all u B. This means that 0 is regular value of H(, .).
Therefore from the very denition of the degree at regular values
deg
0
(H(, .), B, 0) = (D
u
H(, (), [0, 1]),
where (() = u
1
is the segment joining 0 to u
1
. But from the above, D
u
H(, u
1
) is an
isomorphism, and since the parity of a path of isomorphisms is 1, we conclude that
deg
0
(H(, .), B, 0) = 1.
be real numbers
satisfying
0
> 0 IR
N
N
,=1
A
k
0
[[
2
. (3.3)
Let also d
k
IR
N
be given vectors. Dene for each k the (scalar) elliptic dierential operator
L
k
by
L
k
u
k
=
N
,=1
A
k
u
k
+d
k
u
k
for u
k
W
2,p
, (3.4)
and set for u = (u
1
, . . . , u
m
)
T
X
p
Lu = L
_
_
_
u
1
.
.
.
u
m
_
_
_
=
_
_
_
L
1
u
1
.
.
.
L
m
u
m
_
_
_
. (3.5)
Let f : IR
N
IR
m
IR
m
be an equicontinuous C
1
f(, 0)
bounded
2
. We assume that there is a matrix-valued function g
C(IR
m
, IR
mm
) such that
lim
|x|
_
1
0
f(x, t) dt g
()
= 0, (3.6)
uniformly for in compact subsets of IR
m
, and we set f
() = g
().
Consider then the operator F dened by (3.2). It has the form (1.2)
F(u) =
N
,=1
a
(., u, u)
2
(x,
0
, ..,
N
) =
_
_
_
_
_
A
1
0 0
0 A
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 A
m
_
_
_
_
_
, (3.8)
and
b(x,
0
, ..,
N
) =
_
_
_
_
_
_
_
N
=1
d
1
+f
1
(x,
0
)
.
.
.
N
=1
d
m
+f
m
(x,
0
)
_
_
_
_
_
_
_
. (3.9)
2
here replace the original
0
.
3.2. A SPECIAL CASE (REACTION - DIFFUSION SYSTEMS) 71
One can check that L is strongly elliptic in the sense of Legendre-Hadamard. This, together
with the assumptions made on the nonlinearity f, ensure that conditions (2.10)(2.13) and
(2.18), are satised. Furthermore F has a limit operator F
(u) = Lu + f
(u) (3.10)
which is dierentiable at 0 with (according to (2.46))
DF
(0)u = Lu +g
(0)u. (3.11)
Therefore, DF
,=1
p
(x)
2
v +B(x) v,
where the matrix (p
f(x, t) dt
_
g
() = f
().
So letting [x[ in (3.12), we get
f
k
()
k
[
k
[
2
IR
m
. (3.13)
Now if u = (u
1
, . . . , u
m
) X
p
satises F
k
(u)u
k
(u
k
)
2
0.
Therefore u
k
= 0 from Lemma 3.2 (iii). From Theorem 2.7 (with = IR
N
), we conclude that
F is proper on the closed bounded subsets of X
p
.
Proposition 3.2 If DF(0) GL(X
p
, Y
p
), then for every h Y
p
, F(u) = h is solvable in X
p
.
Proof. F satises all the properties required in 3.1. So to prove solvability it is enough
to obtain a priori estimates. Set for t [0, 1], E
t
= u X
p
, H(t, u) = 0. Since E
0
=
DF(0)
1
(h), we need to show that E
=
0<t1
E
t
is bounded.
We know that L
k
+ is an isomorphism from W
2,p
(IR
N
) to L
p
(IR
N
) (apply for example
Corollary 2.5 with m = 1). So let for k = 1, . . . , m,
k
= (L
k
+)
1
([h
k
[). Then by Lemma 3.2
(i),
k
0. Now consider an element u E
. We claim that [u
k
(x)[
k
(x) for every x IR
N
and every k = 1, . . . , m. Suppose rst that for some k 1, . . . , m and some x IR
N
, we have
u
k
(x) >
k
(x). Let then z = u
k
k
, and =
_
x IR
N
, z(x) > 0
_
. Then, for all x
L
k
z = L
k
u
k
L
k
k
=
1
t
f
k
(tu) +th
k
(L
k
+)
k
+
k
=
1
t
f
k
(tu) +th
k
[h
k
[ +
k
.
But for x , tu
k
(x) > 0, and so by condition (3.12)
f
k
(x, tu(x)) tu
k
(x) hence
1
t
f
k
(x, tu(x)) u
k
(x)
Therefore
L
k
z u
k
+
k
< 0.
3.3. A SECOND HOMOTOPY 73
If = IR
N
, then by Lemma 3.2 (i) z 0 on IR
N
, contradiction. If ,= IR
N
, then z(x) = 0 for
x , and z 0 on from Lemma 3.2 (ii). But this contradicts the denition of . Thus
u
k
(x)
k
(x) for every x IR
N
and every k = 1, . . . , m.
Next, suppose that u
k
(x) <
k
(x) for some k 1, . . . , m and some x IR
N
, and let
w = u
k
+
k
, and =
_
x IR
N
, w(x) < 0
_
. Then
L
k
w = L
k
u
k
+L
k
k
=
1
t
f
k
(tu) +th
k
+ (L
k
+)
k
k
=
1
t
f
k
(tu) +th
k
+[h
k
[
k
u
k
k
> 0,
because f
k
(x, tu(x)) tu
k
(x) since u
k
(x) < 0 when x . Once again by Lemma 3.2
(i) and (ii), w 0 on , and we have a contradiction. Consequently [u(x)[ [(x)[ and so
|u|
||
and |u|
0,p
||
0,p
.
Now as already observed several times f can be written as f(x, ) = g(x, ) , where g is the
equicontinuous C
0
bundle map dened by g(x, ) =
_
1
0
f(., 0)
is bounded. Accordingly, for every bounded K IR
M
there is a constant M
K
such that
[f(x, )[ M
K
[[ for all x IR
N
and K. Consequently, since |u|
||
, there is
M = M
(L + 1)u =
1
t
f(tu) +th +u,
and since (L + 1) GL(X
p
, Y
p
), there is C > 0 such that |u|
2,p
C|(L + 1)u|
0,p
, and
consequently
|u|
2,p
C|(L + 1)u|
0,p
C ((M + 1)||
0,p
+|h|
0,p
) .
t
() ,= 0 for all IR
N
and t [0, 1]. (3.15)
Lemma 3.4 H has the following properties.
74 CHAPTER 3. EXISTENCE FROM A PRIORI BOUNDS
(i) H C
1
(IR X
p
, Y
p
).
(ii) D
u
H(t, u)
0
(X
p
, Y
p
) (t, u) [0, 1] X
p
.
(iii) For every bounded B X
p
, the collection (H(., u))
uB
is equicontinuous.
(iv) The restriction of H to any closed bounded subset of [0, 1] X
p
is proper.
Proof. Indeed it is sucient to prove the above properties for G(t, u) = H(t, u) +th.
(i) is straightforward.
(ii) For each t [0, 1], G(t, .) has the same form of F with f(x, ) replaced by tf(x, )+(1t)
which satises exactly the same assumptions on f. Thus, once again, all theorems of chapter 2
apply to G(t, .). In particular by Theorem 2.3, it is enough to prove that D
u
G(t, 0)
0
(X
p
, Y
p
).
But D
u
G(t, 0)u = Lu + tDf(0)u + (1 t)u, and it has (3.14) as a limit operator, which is
assumed to be an isomorphism for all t [0, 1]. Therefore the result follows from Corollary 2.6.
(iii) Let t, t
0
[0, 1], then G(t, u) G(t
0
, u) = (t t
0
)f(u) +(t
0
t)u. And the result follows
from the fact that f(u) is bounded in Y
p
when u is bounded in X
p
(see for example Lemma 2.3
(i)).
(iv) We prove that each partial G(t, .) map is proper. Then recalling point (iii), we proceed
exactly as in the proof of Lemma 3.1 (4.). Now for each t [0, 1], G(t, .) has a limit problem
G
(t, u) = Lu +tf
(u) + (1 t)u
Let u X
p
satises G
(t, u) = 0. Then
(L
k
u
k
)u
k
= tf
k
(u)u
k
(1 t)(u
k
)
2
0
as it follows from (3.12). Therefore the maximum principle implies u
k
= 0. By Theorem 2.7
(recalling (ii)), G(t., ) is proper on the closed bounded subsets of X
p
.
Corollary 3.1 For every bounded subset B in X
p
, H[
[0,1]
B
is B-admissible.
Lemma 3.5 Let E = u X
p
, H(t, u) = 0 for some t [0, 1]. Then E is bounded.
Proof. Let for k = 1, . . . , m,
k
= (L
k
+ )
1
([h
k
[). Then by Lemma 3.2 (i),
k
0. We
claim that [u
k
(x)[
k
(x) for every x IR
N
and every k = 1, . . . , m. Suppose rst that for
some k 1, . . . , m and some x IR
N
, we have u
k
(x) >
k
(x). Let then z = u
k
k
, and
=
_
x IR
N
, z(x) > 0
_
. Then for all x
L
k
z = L
k
u
k
L
k
k
= tf
k
(u) (1 t)u
k
+th
k
(L
k
+)
k
+
k
= tf
k
(tu) (1 t)u
k
+th
k
[h
k
[ +
k
u
k
+
k
< 0,
which leads to a contradiction by the maximum principle. Therefore u
k
(x)
k
(x) for every
x IR
N
and every k = 1, . . . , m.
A similar discussion as in the proof of Proposition 3.2 shows that u
k
(x)
k
(x) for every
x IR
N
and every k = 1, . . . , m.
Consequently, [u(x)[ [(x)[ and so |u|
||
and |u|
0,p
||
0,p
. The conclusion
follows exactly as in the end of Propositions 3.2 proof.
3.3. A SECOND HOMOTOPY 75
Proposition 3.3 For every h Y
p
, F(u) = h is solvable in X
p
.
Proof. Let B be a ball of center 0 and containing E. Then indeed 0 / H([0, 1], B). Now
H(0, .) = L + GL(X
p
, Y
p
), therefore 0 is a base point of H(0, .) and deg
0
(H(0, .), B, 0) = 1.
The contraposition of Theorem 1.1 (ii) ensures the existence of an element u B for which
H(1, u) = F(u) h = 0.
Remark 3.3 The contraposition of Theorem 1.1 (ii) ensures also the existence of a base point
u
0
of H(1, .), and consequently of F. Then, deg
u
0
(F h, B, 0) = 1.
76 CHAPTER 3. EXISTENCE FROM A PRIORI BOUNDS
Chapter 4
Global continuation in nonlinear
elasticity
4.1 Setting of the problem
In three dimensional elasticity the following notation is often used when formulating the basic
eld equations.
M is the set of all real 3 3 matrices,
M
+
= E M [ det E > 0 ,
O
+
=
_
E M [ E
T
E = EE
T
= I and det E = 1
_
,
where I is the identity matrix, and E
T
is the transpose of E. A scalar product on M is dened
by
E, G) = trace EG
T
= trace E
T
G.
and the associated norm is denoted by | |. This norm satises |E| = |E
T
|, and |QE| =
|EQ| = |E| for all E M, and Q O
+
.
We consider an elastic body in an unstressed reference conguration which lls IR
3
. In this
case a deformation of the body is described by a map such that
(i) C
1
(IR
3
, IR
3
).
(ii) (x) M
+
for all x IR
3
.
(iii) : IR
3
IR
3
is injective.
We suppose that the mass density in the reference conguration is given by a function
: IR
3
IR such that
C(IR
3
) and (x) > 0 for all x IR
3
.
The external forces acting on the body are given by a density f : IR
3
IR
3
IR
3
such that
f : IR
3
IR
3
IR
3
is an equicontinuous C
1
bundle map.
A particular example reminiscent of the Hooks law is
f(x, ) = k( x) +f
0
(x),
77
78 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
where k > 0, and f
0
: IR
3
IR
3
is a continuous function with compact support. Under these
assumptions, the equations of equilibrium for a deformation are
div T(x) +(x)f(x, (x)) = 0 for x IR
3
, (4.1)
T(x)[(x)]
T
= [(x)][T(x)]
T
for x IR
3
, (4.2)
where T(x) denotes the rst Piola-Kirchho stress tensor at x.
We suppose that the body is asymptotically undeformed in the sense that
(x) x 0 as [x[ . (4.3)
To complete the specication of the problem (4.1) (4.3), we must say how the stress in the
material is determined by its deformation. This is done through a constitutive relation of the
form
T(x) =
T((x)),
where
T : M
+
M is the elastic response function, which depends on the material. The
frame indierence requires that
T(QE) = Q
T(E), G
_
= DW(E)G for all E M
+
and G M.
Remark 4.1 Here, DW(E) is the Frechet derivative of W at E, and so it is a linear function
from M to IR, whereas
T(E) is a matrix.
Remark 4.2 Materials for which the response function derives from a potential energy are
called hyperelastic.
If we write in the form (x) = x + u(x), (the function u : IR
3
IR
3
is called the
displacement corresponding to the deformation ), the equations (4.1), (4.2) and (4.3) are
then equivalent to
div
T (I +u(x)) +(x)f (x, x +u(x)) = 0 for x IR
3
, (4.5)
lim
|x|
u(x) = 0. (4.6)
4.1. SETTING OF THE PROBLEM 79
We note that, for i = 1, 2, 3,
_
div
T (I +u(x))
_
i
=
3
=1
T
i
(I +u(x))
_
=
3
=1
3
,j=1
T
i
(I +u(x))
E
j
[u(x)]
j
=
3
=1
3
,j=1
T
i
(I +u(x))
E
j
u
j
(x).
Before going further, let us write down explicitly the expression of
T, for a Mooney-Rivlin
material, In doing so, we are led naturally to establish some useful identities that will be needed
in the sequel.
Lemma 4.1 For E M
+
,
j=1
E
ij
(Cof E)
ij
(expanding det E along the i
th
row).
Therefore,
det E
E
ik
=
3
j=1
E
ij
E
ik
(Cof E)
ij
+E
ij
(Cof E)
ij
E
ik
=
3
j=1
jk
(Cof E)
ij
+E
ij
0, since (Cof E)
ij
contains no terms from row i
= (Cof E)
ik
.
80 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
Hence,
D(det E)G =
3
i,k=1
det E
E
ik
G
ik
=
3
i,k=1
(Cof E)
ik
G
ik
= trace (Cof E)
T
G = Cof E, G).
= (det E)E
T
, G).
Now, by dierentiating the identity EE
1
= I, we get D(E
1
)G = E
1
GE
1
, and by the
chain rule, D(E
T
)G =
_
E
1
GE
1
_
T
= E
T
G
T
E
T
. Since Cof E = (det E)E
T
, we get
D(Cof E)G = (det E)E
T
, G)E
T
+ (det E)(E
T
G
T
E
T
).
Replacing Cof E and D(Cof E)G by their values in (4.8), using the properties of the scalar
product , ), and the identity E
T
, E
T
G
T
E
T
) = E
T
E
1
E
T
, G), we nally obtain
DB(E)G = 2b(det E)
2
_
_
_
E
1
_
_
2
E
T
, G)
E
T
E
1
E
T
, G
_
_
.
Equation (4.9) becomes
DC(E)G = h
E
T
E
1
E
T
, G
_
_
+h
Remark 4.3 In fact, the identities of the previous lemma, hold for arbitrary n n matrices
with nonzero determinant, and not just for E M
+
.
It follows that
T(I) = (2a+4b+h
(1) = 0. (4.10)
Corollary 4.1
T has the following properties.
R(i)
T C
2
(M
+
, M), and
T(I) = 0.
R(ii)
T(QE) = Q
T(E)]
T
for all E M
+
.
Remark 4.4 The property R(ii) means that the response function is frame-indierent. By
R(iii), the equation (4.2) for the balance of the moments of the forces is always satised.
Dierentiating the terms of
T and using the identities of the previous lemma, we get
4.1. SETTING OF THE PROBLEM 81
Lemma 4.2 Let
B(E) = (det E)
2
_
_
_
E
1
_
_
2
E
T
E
T
E
1
E
T
_
C(E) = h
B(E)G = 2(det E)
2
E
T
, G)
_
|E
1
|
2
E
T
E
T
E
1
E
T
_
(det E)
2
_
2E
1
, E
1
GE
1
)E
T
+|E
1
|
2
E
T
G
T
E
T
_
+ (det E)
2
E
T
_
G
T
E
T
E
1
+E
1
GE
1
+E
1
E
T
G
T
_
E
T
and
D
C(E)G = h
(det E)(det E)
2
E
T
, G)E
T
+h
(det E)(det E)
_
E
T
, G)E
T
E
T
G
T
E
T
_
= h
(det E)(det E)
_
E
T
, G)E
T
E
T
G
T
E
T
_
Since
T(E) = 2aE +2b
B(E) +
C(E), we have D
T(E)G = 2aG+2bD
B(E)G+D
T(I)G = [4b +h
(1) +h
(1))G
T
.
But from (4.10), it follows that 2(a +b) = 2(b +h
(1) +h
(1) = 2a +h
(1),
= 2(a +b) > 0,
are the Lame constants, and satisfy
+ 2 = 2a + 4b +h
(1) > 0.
At this point, two issues remain to be resolved before we can use the results of chapter
2 in this context. First of all, the equation (4.5) is not dened for all smooth functions u :
IR
3
IR
3
but only for those such that I +u(x) M
+
for all x IR
3
. Secondly, amongst all
possible solutions of (4.5), only those such that x x +u(x) is injective on IR
3
correspond to
deformations of the elastic body.
To deal with the rst point, we clearly have to extend the denition of
T from M
+
to all of M.
For most response functions, this cannot be done in a smooth way because
_
_
_
T(E)
_
_
_ when det E 0.
Hence we must restrict
T to a slightly smaller set and then nd a smooth extension of this
restriction.
82 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
Lemma 4.3 Let h C
3
(0, ) satisfy h
C
3
(IR) such that
h
on (, ) as follows
h
(t) =
_
_
h() +h
()(t ) +
1
2
h
()(t )
2
+
1
6
h
()(t )
3
if h
() 0,
h() +
_
h
()
h
()
2
h
()
_
(t ) +
h
()
3
h
()
2
_
exp
_
h
()
h
()
(t )
_
1
_
if h
() > 0.
Note that in the second case h
() > 0, because h
() = h
(i)
() for i = 0, 1, 2, 3, and that h
(t) 0
for all t .
Setting
W
(E) = a |E|
2
+b |Cof E|
2
+h
C
3
(M, IR) and W
, where
M
C
2
(M, M) by
_
(E), G
_
= DW
(E) =
T(E) for all E M
.
We x > 0 and dene a matrix valued function a
: IR
3
_
IR
3
M
_
M by
_
a
(x,
0
,
)
_
i,j
=
i
(I +
)
E
j
(4.13)
where , , i, j = 1, 2, 3. We also dene a function b : IR
3
IR
3
M IR
3
by
b(x, ) = (x)f (x, x +
0
) (4.14)
We replace (4.5) by the equation
div
T
(I +u(x))
_
i
=
3
=1
3
,j=1
i
(I +u(x))
E
j
u
j
(x)
=
_
_
3
,=1
a
u
_
_
i
.
We prove that this system satises the ellipticity condition (2.18), and we rst claim that
_
D
(E)( ),
_
2a[[
2
[[
2
, (4.16)
4.1. SETTING OF THE PROBLEM 83
for all , IR
3
and all E M, and where ( )
ij
=
i
j
.
One way to show this, is to make a direct calculation based on Lemma 4.2, after noting that
some of the identities hold in fact for every E M, because the set E M [ det E ,= 0 is
dense in M
1
.
We give however, a more conceptual proof. In doing so, we introduce two fundamental
concepts in elasticity that we need later.
Note that the function W
= 0), however, if we
dene
W : M M IR IR by
W(E
1
, E
2
, ) = a |E
1
|
2
+b |E
2
|
2
+h
(),
then
W is convex in its arguments, and we have W
(E) =
W(E, Cof E, det E). A function
having this property is called polyconvex, a concept introduced by J. Ball in the context of
the calculus of variations.
Now, it can be shown (using the Jensen inequality), that a polyconvex function P : M IR
satises
P(E)
1
meas D
_
D
P(E +v) dx,
for every bounded measurable subset D IR
3
, every E M and every v W
1,
0
(D, IR
3
). A
function satisfying the preceding inequality is called quasiconvex.
Finally a C
2
quasiconvex function P satises the (not strong) Legendre-Hadamard condi-
tion
D
2
P(E)( , ) 0 for all E M, and all , IR
3
.
For more details and proofs, see for instance [10].
Now we prove inequality (4.16). Note that W
(E) where
A(E) = a |E|
2
, B(E) = b |Cof E|
2
, C
(E) = h
(det E).
For any E, G M,
D
2
A(E)(G, G) = 2a |G|
2
,
and so,
D
2
A(E)( , ) = 2a | |
2
= 2a[[
2
[[
2
for all , IR
3
.
Note that the parts B and C
(E)( , ) 0,
for all , IR
3
. But
_
D
(E)G, G
_
= D
2
W
(E)(G, G) 2a |G|
2
= 2a [[
2
[[
2
.
We now show that the ellipticity condition (2.18) is satised. Since
_
_
3
,=1
(x, )
_
_
ij
=
3
,=1
i
(I +
)
E
j
1
Since the equation det(E I) = 0 has nitely many solutions in IR, then det(E
1
n
I) = 0 for n IN
large enough, and indeed E
1
n
I E as n .
84 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
we have that
3
i,j=1
_
_
3
,=1
(x, )
_
_
ij
j
=
3
i,j=1
3
,=1
i
(I +
))
E
j
j
=
_
D
_
I +
_
( ),
_
= D
2
W
(I +
)( , ) 2a[[
2
[[
2
,
for all , IR
3
0 and (x, ) IR
3
_
IR
3
M
_
.
Thus, (since a > 0), the Strong Legendre-Hadamard condition (2.17) is satised and so also
(2.18).
Henceforth, we assume that the mass density (x) and the body forces f(x, ) have the
following property.
(B) There exist a real number p > 3 such that the function
x (x)f(x, x) belongs to L
p
(IR
3
, IR
3
),
and the function
x (x)
f(x, x) belongs to L
(IR
3
, M).
Then the hypotheses (2.10) to (2.13) are satised by (4.13) and (4.14). Solutions of the system
u X
p
= W
2,p
(IR
3
, IR
3
) and (4.17)
div
=
_
u X
p
[ I +u(x) M
for all x IR
3
_
for 0 < < 1.
Lemma 4.4 For any real number p > 3, the sets O and O
for all x IR
3
. Thus O =
_
0<<1
O
.
(ii) For any u O, the function dened by (x) = x +u(x) is a deformation.
(iii) For any (0, 1) there exists > 0 such that u X
p
: |u|
< O
.
Proof. Since det I = 1, there exists > 0 such that
det E
1
2
for all E Z = E M : |E I| < .
(i) Fix u O. Since u X
p
C
1
d
(IR
3
, IR
3
), there exists R > 0 such that
[u(x)[ +|u(x)| <
2
whenever [x[ R.
4.1. SETTING OF THE PROBLEM 85
In particular, I + u(x) Z for all [x[ R and consequently det(I + u(x))
1
2
whenever
[x[ R.
But det(I + u(x)) > 0 for all x and hence infdet(I + u(x)) : [x[ R > 0 by the
continuity of u. Thus we see that I +u(x) M
for all x IR
3
for any < min
1
2
, , and
hence u O
.
Furthermore, there exists > 0 such that det(I + E)
2
provided that |E u(x)| <
for some x with [x[ R. By the continuity of the embedding of X
p
in C
1
d
(IR
3
, IR
3
), there
exists > 0 such that |v(x) u(x)| < min
2
, for all x IR
3
and all v X
p
such that
|v u|
2,p
< . Then for v X
p
with |v u|
2,p
< , we have that
|v(x)| |u(x)| +|v(x) u(x)| <
2
+
2
for [x[ R,
and so I + v(x) Z for all [x[ R, whereas |v(x) u(x)| < for [x[ R, and so
det(I +v(x))
2
for all [x[ R.
It follows that det(I +v(x)) min
1
2
,
2
for all x IR
3
and all v X
p
with |v u|
2,p
< .
This proves that O is an open subset of X
p
. The proof that O
is open is similar.
(ii) Fix u O and set (x) = x +u(x). There exists r > 0 such that
[u(x)[ +|u(x)| <
1
2
whenever [x[ r.
We show rst that is injective on x IR
3
: [x[ r + 1.
Suppose that [x[ , [y[ r + 1 and that (x) = (y). Then
[x y[ = [u(x) u(y)[ [u(x)[ +[u(y)[ 1.
But
u(y) u(x) =
_
1
0
d
dt
u(ty + (1 t)x) dt =
_
1
0
u(ty + (1 t)x) dt[y x]
and so
[x y[ = [u(y) u(x)[ [y x[ max
0t1
|u(ty + (1 t)x)| .
However
[ty + (1 t)x[ = [x t(y x)[ [x[ t [y x[ r
since [x[ r + 1 and [y x[ 1. Therefore, |u(ty + (1 t)x)| <
1
2
for all t [0, 1] and
consequently,
[x y[ = [u(y) u(x)[
1
2
[y x[ ,
so we must have x = y, proving the injectivity of on the region x IR
3
: [x[ r + 1. From
this we can now deduce that is injective on all of IR
3
by appealing to a well-known result due
to Meisters and Olech [26],[9]. Indeed, given two elements x, y IR
3
such that (x) = (y),
there exists R > r + 1 such that x, y B, where B = z IR
3
: [z[ R. Now as we have
just shown, restricted to B is injective and det (x) > 0 for all x since u O. Hence
restricted to B is injective by the result of Meisters and Olech, and therefore, x = y.
(iii) Let U = E M : |E I| < 1 and recall that for all E U, E is invertible with
det E > 0. For any < 1,
() infdet E : |E I| > 0.
Now consider an element u X
p
such that |u|
and satises
(4.17),(4.18). Since O
is an open neighborhood of 0 in W
2,p
(IR
3
, IR
3
), solutions of (4.18) that
are obtained by continuation from u 0 will, at least initially lie in the subset O
.
Now, we are ready to study the Fredholm and properness properties of the map
F
(u)(x) = div
.
(T)
T
: M IR dened by
W
(E) = a |E|
2
+b |Cof E|
2
+h
(det E),
where a > 0, b 0 are constants, and h
: IR IR is a C
3
convex function. Furthermore,
W
(I) = 0 and
T
(I) = 0.
4.2 Fredholm and properness properties
So far, we assumed the following about the density and the forces.
C(IR
3
) and (x) > 0 for all x IR
3
, (4.20)
f : IR
3
IR
3
IR
3
is an equicontinuous C
1
(IR
3
, M) and L
p
(IR
3
, IR
3
) for some p (3, ). (4.22)
Now, we add the following assumptions.
lim
|x|
(x) =
> 0, (4.23)
there exists a function f
C
1
(IR
3
, IR
3
) with f
f(x, x +) f
(). (4.26)
(G) g (or equivalently f
() k[[
2
. (4.27)
Under the previous assumptions we have a limit operator in the sense of 2.5.
F
(u) = div
T(I +u)
(u) (4.28)
4.2. FREDHOLM AND PROPERNESS PROPERTIES 87
Remark 4.5 It follows that := inf
xIR
3
(x) > 0. Indeed, from (4.23), there is R > 0 such that
(x)
1
2
, whenever [x[ > R. From (4.20), attains its (positive) minimum on the compact
set B
R
at a point x
0
. But then min(
1
2
, (x
0
)) > 0.
Remark 4.6 Under assumptions (4.20) and (4.23), condition (4.22) is equivalent to
x
f(x, x) L
(IR
3
, M), and x f(x, x) L
p
(IR
3
, IR
3
). (4.29)
Indeed, if (4.22) holds, then (4.29) holds as well because |
f(x, x)|
1
|(x)
f(x, x)| ||
0,
|
T
f
(0) k[[
2
.
Remark 4.8 Assumption (G) is weaker than the fact that f is conservative with respect to
. Indeed, the condition that there is : IR
3
IR
3
IR such that
(x, ) = f(x, ), is
equivalent to the condition that rot
() = 0 and so f
is conservative.
Remark 4.9 The eld f(x, ) = k( x) +f
0
(x), given in 4.1 satises all our assumptions,
with f
() = k.
Proposition 4.1 Under assumptions (T) and (4.20)(4.22), the operator F
in (4.19) is of
class C
1
from X
p
to Y
p
. Its derivative at zero is
DF
is C
1
, follows from Theorem 2.2, since the coecients of F
satisfy the
required assumptions (2.10)(2.13). In particular, from the chain rule and the linearity of the
operators div and , it follows that
DF
(0)u = div D
(I)u (x)
f(x, x)u,
From Corollary 4.2, we have, since
T
(I) =
T(I),
D
(I)u
_
i
=
3
j=1
j
(div u)
ij
+
3
j=1
j
(
i
u
j
+
j
u
i
)
=
i
(div u) +
i
(div u) +u
i
= ( +) ((div u))
i
+u
i
.
(IR
3
) satisfy (x) = [x[ for [x[ > 2, and (x) = 0 when [x[ < 1. Let
u = e
s
v for some v X
p
and some s > 0. Then the following hold.
(i) u W
1,q
for all q [1, ].
(ii)
u
i
u
j
W
1,1
(IR
3
), for every i, j = 1, 2, 3 and , = 0, 1, 2, 3. In particular,
|u|
2
, (div u)
2
, |rot u|
2
, u u, (div u) u L
1
(IR
3
),
and
(div u)u, (u)
T
u W
1,1
(IR
3
, IR
3
).
(iii)
_
IR
3
w = 0 for every w W
1,1
(IR
3
), and every = 1, 2, 3.
Proof. (i) Note rst, that since v L
and e
s
L
q
for any q [1, ], we have u L
q
.
Second, by a direct computation already made in 2.7, we have for = 1, 2, 3,
(e
s
v) = e
s
v s
e
s
v.
Using the fact that the derivatives of are bounded functions, we see that
u belongs to L
q
as well.
(ii) In particular
u
i
L
p
, where
1
p
+
1
p
= 1. And since
u
j
L
p
, we have from Holder
inequality
u
i
u
j
L
1
. Next,
k
(
u
i
u
j
) =
k
u
i
u
j
+
u
i
k
u
j
. And for the same
reason as before, we have
k
(
u
i
u
j
) L
1
.
(iii) follows from the density of C
0
(IR
3
) in W
1,1
(IR
3
).
Proposition 4.2 Retain assumptions (T) and (4.20)(4.25). Then, DF
(0) GL(X
p
, Y
p
).
Proof. The proof is made in two steps. First, we show that DF
(0)
0
(X
p
, Y
p
), because it
is a compact perturbation of an isomorphism, and then we show that ker DF
(0) = 0.
First step.
DF
(0)u = ( +)(div u) u
(0)u + (
(0) (x)
f(x, x)) u.
Now, let
Lu = ( +)(div u) u
(0)u
= ( +)
_
_
2
1
1
2
1
1
2
2
2
1
3
2
2
3
_
_
_
_
u
1
u
2
u
3
_
_
_
_
0 0
0 0
0 0
_
_
_
_
u
1
u
2
u
3
_
_
(0)
_
_
u
1
u
2
u
3
_
_
.
Then, the characteristic polynomial of L is
S() = ( +)
_
_
2
1
1
2
1
1
2
2
2
1
3
2
2
3
_
_
+
_
_
[[
2
0 0
0 [[
2
0
0 0 [[
2
_
_
(0)
We claim that the matrix S() is positive denite. Indeed,
_
1
2
3
_
_
_
2
1
1
2
1
1
2
2
2
1
3
2
2
3
_
_
_
_
3
_
_
=
T
( ) = ( )
2
,
4.2. FREDHOLM AND PROPERNESS PROPERTIES 89
and
_
1
2
3
_
_
_
[[
2
0 0
0 [[
2
0
0 0 [[
2
_
_
_
_
3
_
_
= [[
2
[[
2
.
Therefore,
T
S() = ( +)( )
2
+[[
2
[[
2
T
f
(0)
By Cauchy inequality, [[
2
[[
2
( )
2
, and from Remark 4.7
T
f
(0) k[[
2
.
So
T
S() ( + 2)( )
2
+k[[
2
k[[
2
,
which proves the claim. But then S() has a positive determinant for all IR
3
, and it follows
from Corollary 2.5 that L is an isomorphism from X
p
and Y
p
.
On the other hand, the operator u (
(0) (x)
(0)
0
(X
p
, Y
p
).
Second step. Let u = (u
1
, u
2
, u
3
) ker DF
(0). Since DF
_
IR
3
(div u) udx =
_
IR
3
(div u)
2
dx.
For the same reasons, we deduce from the identity
div
_
(u)
T
u
_
= u u +u, u) ,
that
_
IR
3
u udx =
_
IR
3
u, u) dx =
_
IR
3
|u|
2
dx.
By a direct calculation,
|rot u|
2
+ (div u)
2
= |u|
2
+ 2
_
2
(u
2
1
u
1
)
1
(u
2
2
u
1
) +
3
(u
3
1
u
1
)
1
(u
3
3
u
1
)
+
3
(u
3
2
u
2
)
2
(u
3
3
u
2
)
_
.
It follows from Lemma 4.5, that all terms in the preceding identity belong to L
1
(IR
3
) and
furthermore
_
IR
3
i
(u
j
l
u
k
) dx = 0 for all i, j, k, l 1, 2, 3. Therefore
_
IR
3
|rot u|
2
dx +
_
IR
3
(div u)
2
dx =
_
IR
3
|u|
2
dx
90 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
and so,
_
IR
3
div D
T(I)u udx = ( +)
_
IR
3
(div u)
2
dx +
_
IR
3
|u|
2
dx
= ( +)
_
IR
3
(div u)
2
dx +
_
IR
3
|rot u|
2
+ (div u)
2
dx
= ( + 2)
_
IR
3
(div u)
2
dx +
_
IR
3
|rot u|
2
dx
min(, + 2)
_
IR
3
|u|
2
dx.
Now, since
f(x, x) L
and u L
2
, we get from assumption (4.25)
_
IR
3
(x) (
(0) = 0. Thus, DF
(0) GL(X
p
, Y
p
).
Proposition 4.3 (Pohozaev identities) Retain assumptions (T), (4.20)(4.25) and (G).
Let u X
p
be a solution of F
(u) = 0. Then,
(i) the functions
_
(I +u), u
_
, W
(I +u), u
_
dx (4.31)
= 3
_
IR
3
W
(I +u) dx 3
_
IR
3
((u) dx. (4.32)
Proof. (i) Observe rst, that, under our assumptions, the operator F
is of class C
1
between
X
p
and Y
p
, and its derivative at zero is the operator
DF
(0) = Lu = ( +)(div u) u
(0)u
But we showed in Proposition 4.2, that L an isomorphism from X
p
to Y
p
. Since the coecients
of F
have all the properties required in chapter 2, we deduce from Theorem 2.3 that F
is
Fredholm of index zero. Because F
(I +)[ = [W
(I +) W
(I)[ max
B
|DW
(I +)|||
for all B. In particular, since u(x) belong to the ball B = B(0, |u|
1,
), for all x IR
3
,
we have
[W
(I +u(x))[ max
B
|DW
(I +)||u(x)|.
4.2. FREDHOLM AND PROPERNESS PROPERTIES 91
Since u decays exponentially we get
[W
(I +u(x))[ const. e
s(x)
, (4.33)
for all x IR. In particular the function x W
(I +u(x)) belongs to L
1
(IR
3
).
Remark. The number s and the constant appearing in the last inequality depend on u, but
note that for the purpose of this proposition, we do not need any kind of uniform decay as in
Theorem 2.14 or Proposition 2.2.
For the same reasons, i.e. the mean value theorem, the exponential decay of u, and the
condition ((0) = 0, the function ((u) has exponential decay, and so belongs also to L
1
(IR
3
).
For the function g(u) u, note that since [u(x)[ |u|
1,
for all x IR
3
, it follows from
the continuity of g, that x g(u(x)) belongs to L
(IR
3
, IR
3
). Since u has exponential decay,
g(u) u has also an exponential decay, and therefore belongs to L
1
. Similarly, the function
_
(I +u), u
_
has exponential decay.
(ii) The idea to prove these identities, is to multiply the equation F
(u) = 0 i.e.
div
T
(I +u) +g(u) = 0,
successively by u and (x )u = (u)x, and integrate by parts.
As before, it follows from the mean value theorem, and the exponential decay of u and u
that
_
(I +u)
_
T
u W
1,1
(IR
3
, IR
3
). Therefore,
_
IR
3
div
_
_
(I +u)
_
T
u
_
dx = 0,
by Lemma 4.5 (iii). But
div
_
_
(I +u)
_
T
u
_
=
_
div
T
(I +u)
_
u +
_
(I +u), u
_
,
with all terms in L
1
(IR
3
). Hence,
_
IR
3
_
div
T
(I +u)
_
udx =
_
IR
3
_
(I +u), u
_
dx,
and equation (4.31) follows. Now,
g(u) (u)x =
3
i,j=1
g
i
(u)(
j
u
i
)x
j
=
3
j=1
3
i=1
i
((u)(
j
u
i
)x
j
=
3
j=1
x
j
x
j
((u).
Note that for any polynomial map P : IR
3
IR, the function P(x)e
s(x)
, decays exponentially
at innity. Consequently, all the terms above are in L
1
, and the function x
j
((u) belongs to
W
1,1
(IR
3
). Therefore,
0 =
_
IR
3
j
[x
j
((u)] dx =
_
IR
3
((u) dx +
_
IR
3
x
j
x
j
((u) dx,
92 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
and so
3
j=1
_
IR
3
x
j
x
j
((u) dx = 3
_
IR
3
((u) dx.
Similarly,
_
IR
3
div
T
(I +u) (u)xdx =
3
i,j,k=1
_
IR
3
x
j
ij
(I +u)x
k
k
u
i
dx
=
3
i,j,k=1
_
IR
3
ij
(I +u)
j
(x
k
k
u
i
) dx
=
3
i,j,k=1
_
IR
3
ij
(I +u)(
kj
k
u
i
+x
k
jk
u
i
) dx
=
_
IR
3
3
i,j=1
ij
(I +u)
j
u
i
dx +
3
k=1
_
IR
3
x
k
k
W
(I +u) dx
=
_
IR
3
(I +u), u) dx 3
_
IR
3
W
(I +u) dx.
Therefore,
3
_
IR
3
((u) dx =
_
IR
3
(I +u), u) dx 3
_
IR
3
W
(I +u) dx,
and equation (4.32) follows.
Theorem 4.1 Retain assumptions (T), (4.20)(4.25), (G), and (4.27). Then F
: X
p
Y
p
is
Fredholm of index zero, and proper on the closed bounded of X
p
.
Proof. The only point that remains to prove is the nonexistence of nontrivial solutions of
F
(I +u) dx = 3
_
IR
3
((u) dx +
_
IR
3
g(u) udx.
From assumption (4.27), we have g() k
[[
2
. As a consequence,
(() =
_
1
0
d
dt
((t) dt =
_
1
0
((t) dt
=
_
1
0
g(t) dt
_
1
0
k
t[[
2
dt
=
1
2
k
[[
2
.
Therefore,
3
_
IR
3
W
(I +u) dx
3
2
k
_
IR
3
[u[
2
dx k
_
IR
3
[u[
2
dx
=
5
2
k
_
IR
3
[u[
2
dx 0 (4.34)
4.3. GLOBAL CONTINUATION 93
so
_
IR
3
W
(I +u) dx 0.
We show that the reverse inequality also holds. It follows from the quasiconvexity of W
0
(IR
3
, IR
3
) such that suppv B
R
_
IR
3
W
(I +v) dx =
_
B
R
W
(I +v) dx (meas B
R
)W
(I) = 0.
Let : IR IR be C
function satisfying (t) = 1 for t < 0, and (t) = 0 for t > 1. Dene
for r > 0,
r
: IR
3
IR by
r
(x) =
_
1 if [x[ < r
([x[ r) if [x[ r.
Then, indeed,
r
is a C
_
I +(
r
u(x))
_
Ce
s(x)
,
for all x IR
3
and r > 0. Consequently,
_
|x|>r
[W
(I + (
r
u))[ dx tends to zero as r .
Given > 0, choose r large enough such that both
_
|x|>r
[W
(I +u)[ dx and
_
|x|>r
[W
(I +(
r
u))[ dx
are less than . Then
_
IR
3
W
(I +u) dx =
_
IR
3
W
(I +u) W
(I +(
r
u)) dx +
_
IR
3
W
(I +(
r
u)) dx
=
_
|x|>r
W
(I +u) dx
_
|x|>r
W
(I +(
r
u)) dx +
_
IR
3
W
(I +(
r
u)) dx
2.
But this is true for any > 0, and therefore,
_
IR
3
W
(I +u) dx 0. Hence,
_
IR
3
W
(I +u) dx = 0.
But then, it follows from (4.34) that
5
2
k
_
IR
3
[u[
2
dx = 0,
which implies u = 0.
4.3 Global continuation
Theorem 4.2 (Global continuation) Let X, Y be two Banach spaces, and J be an open
interval of the real line containing 0. Consider a continuous operator F : J X Y having
the following properties.
F is dierentiable with respect to the second variable, D
u
F(t, u)
0
(X, Y ), for all
(t, u) J X, and the map (t, u) D
u
F(t, u) is continuous.
94 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
For any compact subset A of J and any closed bounded subset B X, the restriction
F[
AB
is proper.
D
u
F(0, 0) GL(X, Y ) and F(0, 0) = 0.
Let O be an open subset of X containing 0, and ( be the maximal connected subset of (J O)
F
1
(0) which contains (0, 0). Let P
IR
and P
X
denote respectively the projections of IRX on
IR and X. Then, at least one of the following holds.
(i) ( is unbounded.
(ii) There exists u X, u ,= 0 such that (0, u) (.
(iii) The closure in IR of P
IR
(() intersects J.
(iv) P
X
(() intersects O, where ( is the closure of ( in IR X.
Remark 4.10 If alternative (iii) holds, then at a point t J, the map F(t, .) may no longer
be dened, or it may loose its properness or Fredholm properties. The meaning of alternative
(iv) will be clear in Theorem 4.3 (see Remark 4.13).
Remark 4.11 If (J O) = (this happens if and only if J = IR and O = X), then neither
of alternative (iii) or (iv) can hold. Therefore, we prove the theorem under the assumption that
(J O) ,= . Note that (J O) =
_
J O
_
_
J O
_
.
Proof. Suppose that none of the alternatives hold. Then,
1. Claim. ( is compact. Indeed, P
IR
(() is contained in J but does not meet J since (iii)
does not hold. Hence it is contained in J. Similarly, P
X
(() O, because (iv) does not hold.
Now observe that ( P
IR
(() P
X
(() P
IR
(() P
X
(() J O. But (, being a connected
component of J O, is closed in J O. Therefore, ( = (, that is, ( is closed in IR X.
Because alternative (i) does not hold, ( is bounded, and accordingly, P
IR
(() is compact. It
follows then from the properness assumption on F, that F
1
(0)
_
P
IR
(() P
IR
(()
_
is compact.
But this set contains the closed set (, which is, therefore, compact. Therefore, in fact, P
IR
(() is
a compact interval [a, b] J, and we can choose two points a
, b
, b
).
2. Since ( (J O) = , we have
d
1
:= dist ((, (J O))
= inf
(t,u)C
dist ((t, u), (J O)) > 0,
because the innimum is attained at a point (t
0
, u
0
) in the compact set (, and (t
0
, u
0
) / (JO).
3. The implicit function theorem ensures that there exist > 0, a C
1
map u : (, ) X
and a neighborhood V
) V
and let (
1
= B((0, 0),
r
2
) (, so that (
1
is open in (
and hence ((
1
is compact.
Therefore, since (ii) does not hold,
d
2
:= dist (((
1
, 0 X) > 0,
4.3. GLOBAL CONTINUATION 95
because ((
1
is compact and 0 X is closed.
Dene
= (t, u) IR X [ dist ((t, u), () < (a
, b
) O
=
_
(t
0
,u
0
)C
B((t
0
, u
0
), ) (a
, b
) O,
where =
1
2
min
_
d
1
, d
2
,
r
2
_
.
2
Then, indeed, J O, and is open and bounded.
Claim. F(0, u) = 0 and (0, u) u = 0.
If not, there exist u ,= 0 such that F(0, u) = 0 and (0, u) . Since the unique solution of
F(0, u) = 0 in V
). Therefore
|u| = |(0, u)| r
/2
_
2
,
and since (
1
B
r
2
, we have
dist ((0, u), (
1
)
r
2
.
(0, u) 0 X dist ((0, u), ((
1
) d
2
. Therefore,
dist ((0, u), () = mindist ((0, u), (
1
) , dist ((0, u), ((
1
)
min
_
r
2
, d
2
_
2.
But (0, u) (t, u) IR X[ dist ((t, u), () , so dist ((0, u), () , contradiction.
4. Let
K = F
1
(0),
which is compact by the properness of F. Since ( , then ( = .
If there were a closed connected subset
( of K intersecting both ( and , then, (
( would
be connected (as the union of two connected sets with non empty intersection), contradicting
the maximality of (.
Therefore, the celebrated separation theorem
3
implies that there exist two disjoint closed
subset A and B of K such that
K = A B,
( A,
F
1
(0) B.
Let
=
_
(t, u) [ dist ((t, u), A) <
1
2
dist (A, B)
_
,
which is open and bounded and contains A. So A = and indeed B = , hence
K = , i.e. F
1
(0) = . Since , we get
F
1
(0) = .
2
By taking smaller, we can ensure that
(t
0
,u
0
)C
B((t
0
, u
0
), ) (a
, b
t
= u O [ (t, u) .
Claim. 0 / F(t,
t
), for all t J.
One can show that u
t
(t, u) .
If 0 F(t,
t
) for some t J, then there exists u
t
such that F(t, u) = 0, so
(t, u) and (t, u) F
1
(0). But F
1
(0) = , a contradiction which proves the claim.
6. It follows from the claim in point 3., that u = 0 is the unique solution of F(0, u) = 0
in
0
. Therefore, by the denition of the base point degree at regular values
deg
0
(F(0, .),
0
, 0) = 1.
By the generalized homotopy invariance of the absolute degree (since 0 / F(t,
t
) for all
t J), we should have
[ deg [(F(t, .),
t
, 0) = 1, for all t J.
However, since (a
, b
) O, we have
t
= , for t (b
, supJ), and so
[ deg [(F(t, .),
t
, 0) = 0,
contradiction.
Remark 4.12 Consider the homotopy (3.1) dened in chapter 3. It satises all the assumptions
of the previous theorem with J = IR and O = D
p
(). Therefore, alternatives (iii) and (iv) can-
not hold. Alternative (ii) cannot hold either because the only solution of H(0, u) = DF(0)u = 0
is u = 0. Then, we deduce that there exists an unbounded branch of solutions (t, u) of
F(tu) = t
2
h emanating from (0, 0).
Consider now the operator F
(., u))
uX
p
is equicontinuous.
Thus, by Lemma 1.1, F
C
1
(IR X
p
, Y
p
), and D
u
F
satises the required conditions of Theorem 4.2. Accordingly, we have 4 alternatives, but
alternative (iii) of that theorem cannot happen since J = IR. Alternative (ii) cannot occur
either because F
(0, u) = F
(0, u), and we showed in Theorem 4.1 that the only solution of
F
(0, u) = 0 is u = 0.
Now we go back to the original problem (without restriction and extension). It turns out
that if we dene the operator F(t, u), by
F(t, u)(x) = div
T(I +u(x)) +k(x)u(x) t(x)f
0
(x),
then F maps IR O into L
p
(IR
3
, IR
3
).
Indeed, for every u O, there exists (0, 1) such that u O
, that F(t, u) = F
(t, u) Y
p
.
We formulate a global continuation result for this original operator.
Theorem 4.4 Let ( be the connected component of F
1
(0) (IR O) which contains (0, 0).
Then either ( is unbounded, or the projection onto X
p
of ( meets O
(0)
(IR O
) which contains (0, 0), then for every (0, 1), either (
is unbounded or P
X
p
((
)
O
,= .
Let ( =
0<<1
(
IR O
(t, u) = 0.
Now we have two cases. Either (i) there is (0, 1) s.t. (
)O
,=
. But since P
X
p
((
) P
X
p
((), we get P
X
p
(() O
: IR X
p
Y
p
is that
F(t, u) = F
(t, u) whenever u O
,
and so F coincides with F
on a open subset of X
p
and this has the following implications.
(a) F C
1
(IR O, Y
p
).
(b) For every u O, there is (0, 1) such that u O
, and so D
u
F(t, u) = DF
(t, u) is
Fredholm of index zero.
(c) For every (0, 1) (and every t IR), the restriction of F(t, .) to any closed bounded
subset A O
is proper.
4
C IRO
.
98 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
We do not know if the set O is connected or simply connected, but in the proof of the
global continuation theorem, all we used is the absolute degree, and this does not requires any
condition of orientability. Accordingly, the theorem applies to the operator F and so we recover
the alternatives.
Generalization. We can treat more general body forces f that depend on a real parameter
t which lies in an interval J. But the list of technical assumptions increases.
Consider a continuous map f : J IR
3
IR
3
IR
3
with the following properties. For
each t J, f(t, .) : IR
3
IR
3
IR
3
is an equicontinuous C
1
f is an
equicontinuous C
0
bundle maps. Furthermore,
f(0, x, x) = 0, for all x IR
3
.
For each t J, the function x
f(t, x, x) belongs to L
(IR
3
, M), and
the function x f(t, x, x) belongs to L
p
(IR
3
, IR
3
). (4.36)
For each t J, the function x
t
f(t, x, x) belongs to L
(IR
3
, M). (4.37)
For every t J,
_
IR
3
[f(t +h, x, x) f(t, x, x)[
p
dx 0 as h 0. (4.38)
For every t J, there exists a function f
t
C
1
(IR
3
, IR
3
) with f
t
(0) = 0 for all t J
and such that lim
|x|
|
f(t, x, x +)
t
()| = 0, (4.39)
uniformly for in compact subsets of IR
3
.
For each t J, there is k
t
> 0 such that (
f(t, x, x)) k
t
[[
2
, x, IR
3
. (4.40)
Let
g
t
() =
t
(). (4.41)
We assume that for each t J, there exists a function (
t
C
1
(IR
3
) such that (
t
() = g
t
()
for all IR
3
. And the last hypothesis is
f
t
() k
t
[[
2
. (4.42)
Under the above conditions, we consider the elasticity operator which depends now on a
parameter t J,
F
(0, 0) = 0.
Then our assumptions ensure that for each t J, the map F
[
AB
is proper
whenever A is compact in J and B X
p
is closed and bounded, it is enough to show by Lemma
1.1 that the collection (F
(., u))
uB
is equicontinuous. This will follow from
Lemma 4.6 Let f(t, u)(x) = f(t, x, x + u(x)), be the Nemytskii operator associated with f.
Then, for every bounded subset B X
p
, the collection (f(., u))
uB
is equicontinuous.
4.3. GLOBAL CONTINUATION 99
Proof.
f(t +h, u)(x) f(t, u)(x) =
_
1
0
d
d
f(t +h, x, x +u(x)) d
= h
_
1
0
t
f(t +h, x, x +u(x)) d
= h
_
1
0
__
1
0
d
ds
t
f (t +h, x, x +su(x)) ds +
t
f(t +h, x, x)
_
d
= h
__
1
0
_
1
0
t
f (t +h, x, x +su(x)) ds d
_
u(x)
+
_
1
0
d
d
f(t +h, x, x) d. (4.44)
The assumptions made on f ensure (as in Lemma 2.1) that there is M > 0 such that
|
t
f (t +h, x, x +su(x)) | M
for all s, [0, 1], x IR
3
, and u B. The last term in (4.44) is f(t +h, x, x) f(t, x, x), and
it norm in L
p
tends to zero as h 0 by assumption (4.38). Therefore, given any > 0, we have
|f(t +h, u) f(t, u)|
0,p
[h[M|u|
0,p
+,
for [h[ small enough.
100 CHAPTER 4. GLOBAL CONTINUATION IN NONLINEAR ELASTICITY
Appendix A
Some results about sequences and
function spaces
In what follows X, Y and Z are real Banach spaces.
Note A1. Let X be reexive, u X and (u
n
) be a bounded sequence from X. Suppose that
every weakly convergent subsequence of (u
n
) converges weakly to u (i.e. the limit is independent
of the subsequence, or (u
n
) has a unique weak cluster point). Then u
n
u.
Proof. If not, there exist f X
,
0
> 0, and a subsequence (u
(n)
) such that
[f, u
(n)
) f, u)[
0
for all n IN.
But (u
(n)
) is bounded, therefore it contains a subsequence (u
((n))
) converging weakly to
some l. By assumption, l = u, so that u
((n))
u. But this contradicts the above inequality.
,
0
> 0 and a subsequence (u
(n)
) such that
f, u
(n)
) f, u)
0
for all n IN.
But (u
(n)
) is bounded so it contains a subsequence (u
((n))
) converging weakly to some v in
X (and hence in Y ). By the uniqueness of the weak limit in Y , we have v = u. Therefore
u
((n))
u in X. But this contradicts the denition of (u
(n)
).
Application. X = W
2,p
(, IR
m
), Y = C
1
d
(, IR
m
), with N < p < , and bounded and
Lipschitz.
Note A3. Let L : X Z have the following property: if (u
n
) is bounded in X, u
n
0 in Y
and L(u
n
) converges in Z then (u
n
) contains a subsequence converging to zero in X. Then for
(u
n
) as above, we have in fact u
n
0 in X.
Proof. If not, there is a subsequence whose norm is bounded away from zero. But this sub-
sequence has all the properties of (u
n
), so by hypotheses, it contains a subsequence converging
to zero in X, which contradicts its denition.
Note A4. Let E be a subspace of X. Consider a sequence (u
n
) and an element u from E.
Then u
n
u in E u
n
u in X.
101
102 APPENDIX A. SOME RESULTS ABOUT SEQUENCES AND FUNCTION SPACES
Proof. Since X
, if u
n
u in E, then u
n
u in X. Conversely let u
n
u in
X, and f E
. Then
f, u
n
u) =
f, u
n
u) 0. Thus u
n
u in E.
Note B1. Let A IR
N
be a measurable set, and 1 q . Let u
n
u in L
(A) and
v
n
v in L
q
(A). Then u
n
v
n
uv in L
q
(A).
Proof. This is because |u
n
v
n
uv|
q
|u
n
u|
|v
n
|
q
+|u|
|v
n
v|
q
.
Note B2. Let u
n
u in L
(A) and v
n
v in L
q
(A), for 1 q < . Then u
n
v
n
uv in
L
q
(A).
Proof. Let f L
q
(A), where
1
q
+
1
q
= 1. Then
_
A
f(u
n
v
n
uv) =
_
A
f(u
n
v
n
uv
n
) +
_
A
fu(v
n
v).
The result follows from the facts that fu L
q
(A) and
_
A
f(u
n
v
n
uv
n
)
|f|
q
|v
n
|
q
|u
n
u|
.
Note B3. Let f L
1
(A). Then the functional dened on the measurable subsets of A by
(G) =
_
G
[f[, is a measure on A, and as any measure it satises lim
n
(G
n
) = (
G
n
) for
every decreasing family of subsets (G
n
) from A. This is why
lim
n
_
|x|>n
[f[ =
_
B
n
[f[ = 0,
since
B
n
= .
C. Let G IR
N
be an open set, k IN, 1 q < , and (u
n
) W
k,q
(G) such that u
n
u.
Then for every multi-index with [[ k, D
: W
k,q
W
k||,q
is linear and bounded and
hence weakly continuous, therefore D
u
n
D
u in W
k||,q
(G) L
q
(G).
The converse results from a theorem on the representation of elements of
_
W
k,q
_
, see for
instance Adams [1] Theorem 3.8, which states that if f (W
k,q
(G))
) L
q
||k
_
G
(D
u) v
.
Note D. Let IR
N
be an open set, with bounded and Lipschitz and N < p < .
Then the functions of W
1,p
() are bounded and Holder continuous on (Adams [1], Theorem
5.4). Now let u W
1,p
(). Since C
0
(IR
N
) is dense in W
1,p
() ([1], Theorem 3.18), there is a
sequence (u
n
) in C
0
(IR
N
) converging to u in W
1,p
() and hence in L
(). Dene
n
for n IN by
n
(x) =
1 if [x[ n and zero elsewhere. Let T
n
v =
n
uv, and
n
= x : [x[ < n. Since
W
1,p
(
n
)
comp
L
(
n
), T
n
: W
1,p
() L
p
() is compact. Now
|Tv T
n
v|
p
p
=
_
|x|>n
[u[
p
[v[
p
|v|
p
_
|x|>n
[u[
p
const. |v|
p
1,p
_
|x|>n
[u[
p
.
This means that T is the uniform limit of a sequence of compact operators, and so it is itself
compact.
Note E2. Let A be an open unbounded subset of IR
N
, and L
2K
for almost every
x A
B
r
. By the compactness of the imbedding W
1,q
(A) L
q
(A B
r
), we have u
n
0 in
L
q
(A B
r
), and so for n large enough we have ||
q
_
|x|<r
[u
n
[
q
2
. Therefore
_
A
[u
n
[
q
=
_
|x|<r
[u
n
[
q
+
_
|x|>r
[u
n
[
q
for n large enough. This means that Mu
n
0 in L
q
(A) and so M is completely continuous
and hence compact because W
1,q
is reexive.
104 APPENDIX A. SOME RESULTS ABOUT SEQUENCES AND FUNCTION SPACES
Bibliography
[1] R. A. Adams, Sobolev spaces (Academic press, 1975).
[2] Yu. Agranovich, Elliptic boundary problems, Encyclopedia of Mathematical Sciences,
vol. 79 (Springer-Verlag, 1997).
[3] P. Benevieri and M. Furi, A simple notion of orientability for Fredholm maps of
index zero between Banach manifolds and degree, Ann. Sci. Math. Quebec 22 (1998),
131-148.
[4] P. Benevieri and M. Furi, On the concept of orientability for Fredholm maps between
real Banach manifolds, Top. Methods Nonlinear Anal. 16 (2000), 279-306.
[5] P. Benevieri and M. Furi, Bifurcation results for families of Fredholm maps of index
zero between Banach spaces, Nonlinear Anal. Forum 6 (2001), no. 1, 35-47.
[6] H. Br ezis, Analyse fonctionnelle, theorie et applications (Masson, Paris, 1983).
[7] R. F. Brown, A topological introduction to nonlinear analysis (Birkhauser, Boston,
2004).
[8] Y.-Z. Chen and L.-C. Wu, Second order elliptic equations and elliptic systems, ams
Transaction of Mathematical Monogograps, vol. 174 (American Mathematical Society,
Providence, RI 1995).
[9] P. G. Ciarlet, Mathematical Elasticity, Vol. I : Three-Dimensional Elasticity (North
Holland, Amsterdam, 1988).
[10] B. Dacorogna, Direct methods in the calculus of variations (Springer-Verlag, Berlin,
1989).
[11] K. Deimling, Nonlinear functional analysis (Springer-Verlag, Berlin, 1985).
[12] P. M. Fitzpatrick and J. Pejsachowicz, An extension of the Leray-Schauder degree
for fully nonlinear elliptic problems, Proc. Symp. Pure Math. 45, Part I (1986), 425-438.
[13] P. M. Fitzpatrick, J. Pejsachowicz and P. J. Rabier, The degree of proper C
2
Fredholm mappings, J. reine angew. Math. 427 (1992), 1-33.
[14] M. Freidlin, Reaction-diusion in incompressible uid: asymptotic problems, J. Di.
Equ, 179 (2002), 44-96.
[15] J. P. Galdi and P. J. Rabier, Functional properties of the Navier-Stokes operator
and bifurcation of stationary solutions: planar exterior domains, Progress in nonlinear
dierential equations and their applications vol. 35, Birkhauser verlag, Basel (1999),
273-303.
105
106 BIBLIOGRAPHY
[16] J. P. Galdi and P. J. Rabier, Sharp existence results for the stationary Navier-Stokes
equations in three dimensional exterior domains, Arch. Rational Mech. Anal. 154 (2000),
343-368.
[17] H. G. Gebran and C. A. Stuart, Fredholm and Properness properties of quasilinear
elliptic systems of second order, Proc. of Edinburgh Math. Soc. 48 (2005), 91-124.
[18] D. Gilbarg and N. S. Trudinger, Elliptic partial dierential equations of second
order (Springer-Verlag, 1983).
[19] T. J. Healey, Global continuation in displacement problems of nonlinear elastostatics
via the Leray-Schauder degree, Archive Rational Mech. Anal., 152 (2000), 273-282.
[20] T. J. Healey and H. C. Simpson, Global continuation in nonlinear elasticity, Archive
Rational Mech. Anal. 143 (1998), 1-28.
[21] H. Jeanjean, M. Lucia and C. A. Stuart, Branches of solutions to semilinear elliptic
equations on IR
N
, Math. Z. 230 (1999), 79-105.
[22] A. I. Koshelev, A priori estimates in L
p
and generalized solutions of elliptic equations
and systems, (ams translations, series 2, vol. 20, 1962).
[23] J. Lindentrauss and L. Tzafriri, Classical Banach spaces I (Springer-Verlag, 1977).
[24] N. G. Lloyd, Degree Theory (Cambridge University Press, 1978).
[25] W. C. McLean, Strongly elliptic systems and boundary integral equations, (Cambridge
University Press, 2000).
[26] G. H. Meisters and C. Olech, Locally one-to-one mappings and a classical theorem
on schlicht functions, Duke Math. J. 30 (1963), 63-80.
[27] S. G. Mikhlin, Multidimensional Singular Integrals and Integral Equations (Pergamon
Press, Oxford, 1965).
[28] A. Pazy, Semigroups of linear operators and applications to partial dierential equations
(Springer-Verlag, New York, 1983).
[29] J. Pejsachowicz and P. J. Rabier, Degree theory for C
1
Fredholm mappings of
index 0, J. Anal. Math. 76 (1998), 289-319.
[30] P. J. Rabier, Fredholm operators, semigroups and the asymptotic and boundary be-
havior of solutions of PDEs, J. Di. Equ. 193 (2003), 460-480.
[31] P. J. Rabier and C. A. Stuart, Fredholm and properness properties of quasilinear
elliptic operators on IR
N
, Math. Nachr. 231 (2001), 129-168.
[32] P. J. Rabier and C. A. Stuart, Global bifurcation for quasilinear elliptic equations
on IR
N
, Math. Z. 237 (2001), 85-124.
[33] P. J. Rabier and C. A. Stuart, Exponential decay of the solutions of quasilinear
second order equations and Pohozaev identities, J. Di. Equ. 165 (2000), 199-234.
[34] P. J. Rabier and C. A. Stuart, C
1
Fredholm maps and bifurcation for quasilinear
elliptic equations on IR
N
, Progress in nonlinear dierential equations and their applica-
tions vol. 40, Birkhauser verlag, Basel (2000), 249-264.
BIBLIOGRAPHY 107
[35] P. H. Rabinowitz, Some global results for nonlinear eigenvalue problems, J. Funct.
Anal. 7 (1971), 487-513.
[36] Ya. Rotberg, Elliptic boundary value problems in the spaces of distributions (Kluwer
Academic, Dordrecht, 1996).
[37] C. A. Stuart, Introductory lectures on nonlinear elliptic equations on IR
N
, (2003).
[38] H. Triebel, Interpolation theory, function spaces, dierential operators (North-
Holland, Amsterdam, 1978).
[39] T. Valent, Boundary Value Problems of Finite Elasticity (Springer Tracts in Natural
Philosophy, Vol. 31, Springer 1988).
[40] A. Volpert and V. Volpert, Normal solvability of linear elliptic problems, C. R.
Acad. Sci. Paris, t. 332, Serie I (2001), 1-6.
[41] A. Volpert and V. Volpert, Normal solvability and properness of elliptic problems,
AMS translations, series 2, vol. 206 (American Mathematical Society, Providence, RI,
2002).
[42] V. Volpert and A. Volpert, Properness and topological degree for general elliptic
operators, Abstract and applied analysis, Vol 3 (2003), 129-182.
[43] J. T. Wloka, B. Rowley and B. Lawruk, Boundary value problems for elliptic
systems (Cambridge University Press, 1995).
108 BIBLIOGRAPHY
Curriculum Vitae
I was born the 27th of March 1976 in Beirut, Lebanon. I studied at the Carmel Saint Joseph
school until I passed the French Baccalaureat (serie C), and the Lebanese Baccalaureat (Mathem-
atiques elementaires) in 1994. Then, I attended the Lebanese University, Faculty of Sciences,
where I obtained a Master in Mathematics in 1998.
In the same year, I attended a joint program between several French-speaking universities
(including the EPFL), at the end of which I obtained a Diplome detudes approfondies in
mathematical modelling and scientic computations (1999). In the context of this program, I
made a training course with Prof. Jean Descloux at EPFL. The following months, I worked as
an assistant at the American University of Beirut, where I did some numerical simulations with
Prof. M. El Ghoul.
I work in the group of Prof. C. A. Stuart since October 2000, where I have some teaching
activities, in parallel to the preparation of the present dissertation.