UNIVERSIT

`
A DEGLI STUDI DI VERONA
FACOLT
`
A DI SCIENZE MATEMATICHE FISICHE E NATURALI
CORSO DI LAUREA IN MATEMATICA APPLICATA
A Poincare-Bendixson Theorem for
Scalar Reaction Diusion Equations
Relatore Laureando
Ch.mo Prof. Marco Squassina Isacco Perin
Anno Accademico 2011/2012
Contents
Introduction 5
1 Results & Preliminaries 7
1.1 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 The function z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 The Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Axiomatic Structure 13
3 Proofs 17
3.1 The Soft Version . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 More results about z and . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Proof of Proposition 1.1.3 . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Proof of Proposition 1.1.4 . . . . . . . . . . . . . . . . . . . . . . . . 28
3.5 Proof of Theorem 1.3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4 Other Results & Open Questions 33
4.1 Reaction Diusion Equation . . . . . . . . . . . . . . . . . . . . . . . 33
4.1.1 Dirichlet or Neumann boundary conditions . . . . . . . . . . . 33
4.1.2 Periodic boundary conditions and f independent of x . . . . . 33
4.2 Monotone Cyclic Feedback Systems . . . . . . . . . . . . . . . . . . . 34
4.3 Monotone Feedback Delay Equation . . . . . . . . . . . . . . . . . . . 35
4.4 Two Open Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3
Introduction
The Poincare-Bendixson theorem presents an important result about planar sys-
tems
1
of ordinary dierential equations. Thanks to this theorem is possible to an-
alyze the -limit sets
2
of vector elds in the plane. Let, in fact,
0
R
2
be an
initial condition to which it is associated a bounded solution (t) and let (
0
) be
the -limit set of
0
. The Poincare-Bendixson theorem states that the set (
0
)
contains, or an equilibrium point
3
or a periodic orbit
4
or a nite number of equilib-
rium points and a set of homoclinic
5
or heteroclinic
6
orbits connecting them to each
other. The result of this theorem is that the solution of a planar system of ODE
7
has a denite behavior for large times, regardless of the initial data, i.e. the system
1
The planar systems are systems of two equations with two unknowns.
2
Given a system of dierential equations, let : J R R
n
be the solution of the system
relative to a given initial data
0
. We dene the -limit set (
0
) R
n
of
0
as
(
0
) = (
1
, ...,
n
) R
n
: t
j
+ : lim
j
(
1
(t
j
), ...,
n
(t
j
)) = (
1
, ...,
n
).
3
We dene equilibrium point for a system of dierential equations each constant solution.
4
A solution is not constant of a system of dierential equations is called periodic orbit if exist
a time T
0
> 0 such that (t + T
0
) = (t), t > 0.
5
Let : R R
n
be a global solution and let
0
be an equilibrium point both related to the
same dierential equation system. We say that is an homoclinic orbit if
lim
t
(t) = lim
t+
(t) =
0
.
6
Let : R R
n
be a global solution and let
0
,
1
be two equilibrium points both related to
the same dierential equation system. We say that is an heteroclinic orbit if
lim
t
(t) =
0
, lim
t+
(t) =
1
7
ODE stands for Ordinary Dierential Equations
5
has not a chaotic behavior. Obviously the theorem breaks down for ows in space
of dimension 3 or higher, e.g. the Lorenz equations
8
. Surprisingly, a theorem of the
Poincare-Bendixson type does hold for the innite-dimensional dynamical system
given by the general scalar reaction diusion equation
u
t
= u
xx
+ f (x, u, u
x
) , x S
1
(1)
on the circle S
1
(i.e. with periodic boundary conditions) with f C
2
. This paper
aims to enunciate and demonstrate rigorously what was just said about the reaction
diusion equation.
8
The Lorenz equations are the rst example of a system with sensitive dependence on initial
data proposed by Lorenz in his famous article Deterministic Nonperiodic Flow (1963).
6
Chapter 1
Results & Preliminaries
In order to present our results, it is necessary to make some preliminary observations.
The solutions of partial dierential equations are usually dened in Sobolev Spaces
1
rather than in spaces of continuous functions with derivatives. Therefore, let X =
W
2,2
(S
1
) be the Sobolev space on S
1
of maps u : S
1
R with square integrable
second derivative. The solution of the equation (1) will be dened on X. Henceforth
let E X denote the set of equilibrium points of (1), in other words w E if and
only if
0 = w
xx
+ f (x, w, w
x
) , x S
1
. (1.1)
For any initial condition u
0
X is associated a maximal solution curve u(t) of (1)
with t [0, ) and hence an orbit

+
(u
0
) = u(t)[t [0, ). (1.2)
In the event that the orbit
+
(u
0
) is contained in a bounded subset of X, then =
and the -limit set (u
0
) is nonempty, compact, connected and invariant
2
. If we
consider any v
0
(u
0
), there exists a unique solution curve v(t) of (1) through v
0
and, by invariance, v(t) (u
0
) for all time t R. With a further view we can
note that (v
0
) and (v
0
) are both subsets of (u
0
), but for a stationary or periodic
1
We dene Sobolev space the space W
k,p
subset of L
p
such that for a given p 1
W
k,p
= f L
p
[D

f L
p
, k
and where with D

f we mean the weak derivative of the function f.

2
We say that X is an invariant set for the equation (1) if, starting from any initial condition
u
0
which is associated a maximal solution curve u(t), we have that u(t) , t R.
7
CHAPTER 1. Results & Preliminaries
solution u(t) we trivially have
(v
0
) = (v
0
) = (u
0
) =
+
(u
0
) = (u
0
)
for any v
0
(u
0
).
1.1 Main Result
Let X = W
2,2
(S
1
) be the Sobolev space on S
1
of maps u : S
1
R with square
integrable second derivative. Moreover, let E X denote the set of equilibrium
points of (1).
Theorem 1.1.1. (Main Result)
Let u
0
X be such that its maximal orbit
+
(u
0
) is bounded in X. Then the -limit
set (u
0
) satises exactly one of the following alternatives:
1. (u
0
) consists in precisely one periodic orbit;
2. (v
0
) E and (v
0
) E for any v
0
(u
0
).
The alternative 2 allows (u
0
) to consist of homoclinic or heteroclinic solutions
joining equilibrium points, i.e. elements of E (as in the classical Poincare-Bendixson
theorem). We will achieve the proof of Theorem 1.1.1, as a logical consequence,
proving the following three propositions:
Proposition 1.1.2. (Soft Version)
Let v
0
(u
0
). Then (v
0
) contains a periodic solution or an equilibrium point.
The same holds for (v
0
).
Remark 1. (About Soft Version)
By Proposition 1.1.2, the -limit set (u
0
) also contains a periodic solution or an
equilibrium point because (v
0
), (v
0
) are subsets of (u
0
).
Proposition 1.1.3.
Let v
0
(u
0
). Then two are the possible alternatives:
1. (v
0
) and (v
0
) consist entirely of equilibrium points;
2. The solution through v
0
is periodic.
8
1.2. The function z
Proposition 1.1.4.
If (u
0
) contains a periodic orbit then (u
0
) is a single periodic orbit.
The proofs of these propositions will be presented in chapter 3. For these proofs
will be necessary the function that will be presented in the following section.
1.2 The function z
For the proof of Theorem 1.1.1 we will adopt as principal tool the function z that
counts the number of sign changes z () N
0
of a continuous map X along
orbits of the semiow (1). We report, as a result, some results (for more details
about these see Matano 1, Angenent, Brunovsky & Fiedler 1) about the function z:
Proposition 1.2.1.
Consider the function z applied to u
1
(t) u
2
(t) where u
1
and u
2
are two solutions
of (1)
t z

u
1
(t) u
2
(t)

. (1.3)
Then the function (1.3) is nonincreasing with t along u
1
(t) u
2
(t).
Proposition 1.2.2.
The function z drops strictly at t
0
> 0 if u
1
(t
0
) u
2
(t
0
) has a multiple zero. Here,
a multiple zero means an x
0
S
1
such that
u
1
(t
0
, x
0
) u
2
(t
0
, x
0
) = 0
u
1
x
(t
0
, x
0
) u
2
x
(t
0
, x
0
) = 0
Remark 2. (About Proposition 1.2.2)
Note that the function u
x
is dened because C
1
(S
1
) X and u is dened on X.
Proposition 1.2.3.
The function z in (1.3) is nite for any t > 0 even if z (u
1
(0) u
2
(0)) = .
9
CHAPTER 1. Results & Preliminaries
1.3 The Projection
We dene a projection
x
0
from X onto R
2
:

x
0
: X R
2
(1.4)

x
0
() := ((x
0
),
x
(x
0
))
where x
0
is an arbitrary but henceforth xed point on S
1
. The reason to dene such
projection is that thanks to it, we can show that the -limit set (u
0
) of (1) is at
most two-dimensional. In fact, another remarkable result is the following:
Theorem 1.3.1.
Under the assumption of Theorem 1.1.1 the projection
: (u
0
) R
2
(1.5)
is a homeomorphism onto the compact subset ((u
0
)) of R
2
.
The proof of this theorem will be presented in chapter 3. We note that, in the
theorem just stated, the projection is dened independently from x
0
. The reason of
this notation is that the sets
x
0
((u
0
)) are essentially independent from the choice
of x
0
. In fact, any two such sets are homeomorphic and even homotopic by a fam-
ily of homeomorphism of the form
x
1

1
x
0

. These homeomorphism conjugate the

induced ows on
x
0
((u
0
)) and
x
1
((u
0
)). It should be emphasized the fact that
it is only the -limit set to be at most two-dimensional and not the solution of the
innite-dimensional dynamical system (1) as a whole. As a warning we mention a
linear example which shows that need not be injective on any semi-orbit
+

u

regardless of how large

t is chosen.
Example 1. (About Theorem 1.3.1)
Consider
u
t
= u
xx
+ u
x
+ u, x S
1
, (1.6)
with initial condition
u
0
(x) = cos(x) + cos(2x). (1.7)
Then the solution is given by
u(t, x) = cos (t + x) + e
3t
cos (2 (t + x)) .
10
1.3. The Projection
Picking x
0
= 0 we obtain the following projection
(u(t, x
0
)) = (cos(t), sin(t)) + e
3t
(cos(2t), 2 sin(2t)) .
Clearly, (u
0
) is a periodic orbit and projects this periodic orbit homeomorphi-
cally onto the unit circle, but (u(t, x
0
)) intersects itself as well as its -limit set
at arbitrarily large times.
All proofs on the next are based on an axiomatic structure which we develop in
the second chapter. The reason for this axiomatization is twofold. First, there are
standard, but technical, arguments which provide the axiomatic structure itself in
concrete applications like equation (1). Second, this axiomatic structure is the same
adopted by Mallet-Paret and H.Smith in the discussion of monotone cyclic feedback
systems (see Mallet-Paret & H.Smith) although such systems are very dierent from
those discussed in this paper.
11
CHAPTER 1. Results & Preliminaries
12
Chapter 2
Axiomatic Structure
Some basic considerations, valid for the rest of the paper, need to be stated before
presenting the axiomatic structure. The Sobolev space X equipped with the norm
| |
2,2
1
is a Banach space. Consider a strongly continuous local semiow dened
on X. The solutions, u(t) X, depend continuously on t 0 and on the initial
condition u
0
X, as long as they exist. The semiow should be regularizing, i.e.
u
t
(t) exists, and depends continuously on t and u
0
for t > 0. Assume, also, that
the following property of compactness is valid: if t
n
and u(t
n
) is a bounded
sequence, then u(t
n
) has a convergent subsequence. In particular, as a consequence,
the -limit set (u
0
) is compact for any bounded semi-orbit. Finally, we do not
require elements v
0
(u
0
) to have a unique global extension v(t), with t R,
as an orbit of the semiow. Then, of course, v
0
(u
0
) possesses some global
extensions. Lets see how to dene a global extension. We know that there exists a
sequence t
n
such that
v
0
= lim
n
u(t
n
).
We dene the global extension of v
0
as
v(t) := lim
n
u(t + t
n
). (2.1)
Trivially, we have that v(0) = v
0
. By property of compactness above, for any > 0
we can assume convergence of u( +t
n
), with t
n
. Therefore v(t) is dened for
each t 0. We can also assume the convergence of u

t + t
n

for

t < 0 such that
the succession

t + t
n
tends to +. Now, in order to extend the convergence to all
1
|f|
2,2
=

2
k=0

[f
(i)
(t)[
2
dt
1
2
13
CHAPTER 2. Axiomatic Structure
negative times we adopt the Cantors diagonal argument
2
. Thanks to the latter we
can assume that convergence in (2.1) holds t R. This denes a global extension
v(t) and, obviously, v(t) (u
0
) for any t R. Given some such global extensions
v(), we can dene its -limit set (v
0
)
3
because v(t) is dened for all negative t.
So we understand that henceforth each time we write (v
0
) it means that has been
performed the above extension process.
Our abstract assumption on z and are the following: there exist two maps
z : X N
0
,
: X R
2
such that the axioms listed below all hold, whenever the functions u
1
(t) and u
2
(t)
are dened.
(A0) Finiteness: for any two solutions u
1
(t), u
2
(t) of (1) and for any positive time
t, z(u
1
(t) u
2
(t)) is nite.
(A1) z-dropping: let t
0
> 0. Then (u
1
(t
0
) u
2
(t
0
)) = 0 implies that
1. Either t z(u
1
(t) u
2
(t)) drops strictly at t = t
0
, or
2. u
1
(t
0
) u
2
(t
0
) = 0.
(A2) Continuity: if t z(u
1
(t) u
2
(t)) does not drops strictly at t = t
0
and if
u
1
(t
0
) u
2
(t
0
) ,= 0 then z is locally constant, i.e. exist a neighborhood U X
containing u
1
(t
0
) u
2
(t
0
) such that
z() = z(u
1
(t
0
) u
2
(t
0
))
2
The Cantors diagonal argument is a technique used for mathematical proofs. In our specic
case we adopt the same technique used in the proof of the Ascoli-Arzel`a Theorem. For more details
on this, see the rst chapter of M.Fusco & P.Marcellini & C.Sbordone.
3
Given the reaction diusion equation (1), let : R X be the solution of the equation relative
to a given initial data
0
. We dene the -limit set (
0
) X of
0
as
(
0
) = X : t
j
: lim
j
(t
j
) = .
14
for any U.
(A3) Regularity: axioms (A0 A2) also holds for u
t
replacing u
1
u
2
.
For the semiow generated by the scalar parabolic equation (1) we have already
dened z and with these properties above. Indeed, the axiom (A0) follows from
Proposition 1.2.3, the axioms (A1) and (A2) follow because z drops unless all zeros
of x u
1
(t
0
, x)u
2
(t
0
, x) are simple and because C
1
(S
1
) X (see Proposition 1.2.1
and Proposition 1.2.2). Finally the axiom (A3) follows by dierentiating equation
(1) with respect to t to obtain
(u
t
)
t
= (u
t
)
xx
+
f(x, u, u
x
)
u
u
t
+
f(x, u, u
x
)
u
x
(u
t
)
x
(2.2)
The fact that the axiom (A3) hold is proved in Agenent, in a work concerning
the equation (2.2). Axioms (A0 A3) also hold with Dirichlet conditions. For
certain technical reasons, to make that the axioms hold with more general boundary
conditions f must be analytic.
15
CHAPTER 2. Axiomatic Structure
16
Chapter 3
Proofs
3.1 The Soft Version
In this section we prove Proposition 1.1.2. In order to obtain this proof, we have to
present and prove three lemmas. Lemma 3.1.1 is about the projection and it states
that is injective if restricted to the closure of (v
0
), (v
0
), where v
0
is contained
in the -limit set of u
0
. As in the statement of Theorem 1.1.1, u
0
X is such that
its maximal orbit
+
(u
0
) is bounded in X. At the end of this chapter we will see
that, thanks to the proof of the Theorem 1.3.1, is injective on all of (u
0
). Lemma
3.1.2 states that (v(t)) has derivative with respect to t dierent from zero unless v
is an equilibrium point. Lemma 3.1.3 shows that the planar semiow on ((v
0
)),
induced by the semiow on (v
0
), is a solution to a continuous planar vector eld.
Henceforth, we assume axioms (A0 A3) and we keep u
0
xed for the remainder of
this paper. Now, we can state rigorously what we have just explained.
Lemma 3.1.1.
Let v
0
(u
0
) and let v
1
(), v
2
() be solutions of (1) in (v
0
) such that v
1
(0) ,= v
2
(0).
Then
z

v
1
() v
2
()

is locally constant in a neighborhood of = 0. In particular the restricted projection

: (v
0
) R
2
is injective, and hence is a homeomorphism onto its range ((v
0
)).
17
CHAPTER 3. Proofs
Proof.
We give an indirect proof. We deny the injectivity of on (v
0
) to reach a con-
tradiction. Suppose, in fact, there exist v
1
0
, v
2
0
(v
0
), with v
1
0
,= v
2
0
such that
(v
1
0
) = (v
2
0
). Let v
1
(), v
2
() the solution curves of (1) in (v
0
) through v
1
0
, v
2
0
respectively. The z-dropping assumption (A1) implies that z (v
1
() v
2
())
drops at = 0. Then, for > 0
z

v
1
() v
2
()

< z

v
1
() v
2
()

. (3.1)
By axiom (A0), z (v
1
() v
2
()) is nite and it is also constant for negative
respectively positive near 0 because of (A2). Thanks again to the axiom (A2),
we may perturb slightly v
1
, v
2
in (v
0
), v
1
(), v
2
() (v
0
), maintaining the
respective values of z. Therefore, there exist t
0
, t
1
R such that
z (v(t
0
+ t
1
+ ) v(t
0
+ )) = z

v
1
() v
2
()

v
1
() v
2
()

= z (v(t
0
+ t
1
) v(t
0
))
where v is the solution of (1) through v
0
. Then, for (3.1)
z (v(t
0
+ t
1
+ ) v(t
0
+ )) < z (v(t
0
+ t
1
) v(t
0
)) . (3.2)
Again because of (A2), t z (v(t + t
1
) v(t )) can be assumed to be locally
constant near t = t
0
. Furthermore, there exists a sequence t
n
such that
u(t
n
+ t
1
) v (t
0
+ t
1
)
u(t
n
) v (t
0
)
and hence, again by continuity (A2)
z (u(t
n
+ t
1
+ ) u(t
n
+ )) = z (v(t
0
+ t
1
+ ) v(t
0
+ ))
< z (v(t
0
+ t
1
) v(t
0
)) = z (u(t
n
+ t
1
) u(t
n
)) (3.3)
If necessary, passing to a subsequence we may assume that t
n+1
t
n
> . Then (3.3)
implies that
t z (u(t + t
1
) u(t)) (3.4)
drops innitely. This contradicts niteness of z, i.e. (A0). Then must be injective.
18
3.1. The Soft Version
Lemma 3.1.2.
Let v(t) be a solution curve of (1) in (u
0
) such that
d
dt

t=0
(v(t)) = 0.
Then v(0) E, i.e. v(0) is an equilibrium point.
Proof.
Again, the proof is indirect. We suppose by contradiction that v(0) / E. Then
v
t
(0) ,= 0 for the hypothesis just made. By the regularity assumption (A3) and
z-dropping (A1) relative to v
t
, there exists > 0 such that
z (v
t
()) < z (v
t
()) .
As in the previous proof, picking a suitable sequence t
n
and exploiting the
continuity assumption (A2) as done in (3.4), we conclude that
t z (u
t
(t))
drops strictly in each interval [t
n
, t
n
+ ]. This contradicts niteness of z, i.e.
(A0). Then v(0) E.
Lemma 3.1.3.
Let v
0
(u
0
). For each w
0
(v
0
) we dene the following function
F ((w
0
)) :=
d
dt

t=0
(w(t))
where w(t) is any solution curve through w
0
. Then
F :

(v
0
)

R
2
is a well-dened continuous function on the compact set

(v
0
)

R
2
. If w(t)
is any solution curve in (v
0
) then (t) := (w(t)) is a solution of the ordinary
dierential equation

= F().
Proof.
The fact that F is well-dened follows from Lemma 3.1.1 (injectivity of on (v
0
)).
It is also clear that (t) = (w(t)) satises

= F() by construction. Now, the
19
CHAPTER 3. Proofs
only thing that need to be proved is the continuity of F. In order to prove this,
consider a sequence w
n
() (v
0
) of solutions of (1) such that (w
n
(0)) converges
in

(v
0
)

. By compactness of

(v
0
)

, we may assume that the sequence w

n
(t)
converges uniformly to w

(t) for t in a compact set and for a solution w

() (v
0
)
of (1). The functions w
n
are derivable and by continuous dependence, we have
w
n
t
(0) w

t
(0) because w
n
(0) w

(0). But then F ( (w

n
(0))) F ( (w

(0))) by
denition on F, completing the proof.
Thanks to Lemma 3.1.3 we have obtained a planar vector eld F from the semi-
ow on (v
0
) for v
0
(u
0
). This fact strongly indicates why a Poincare-Bendixson
result (Soft Version) should hold on (v
0
). However, the only continuity of F is not
enough to have a unique solution of F. In particular, there might be trajectories of
F which do not correspond to trajectory in (v
0
) through use of the projection .
Then, in addition to the lemmas, we give a detailed proof of Proposition 1.1.2.
Proof. (Proposition 1.1.2)
Proceed proving the proposition for (v
0
). The proof for (v
0
) is analogous, replac-
ing (v
0
) by (v
0
) everywhere. Suppose that (v
0
) does not contain any equilib-
rium point. Then, we will show that (v
0
) contains a periodic solution. We pick
any w
0
(v
0
) and any w

0
(w
0
) and, in addition, we dene the global solution
curves through v
0
, w
0
, w

0
by v(t), w(t), w

(t). These solutions are obtained by an

extension process of the type (2.1). Lemma 3.1.2 implies that
d
dt

t=0
(w

(t)) ,= 0
because w

0
(w
0
) (v
0
) (w
0
can not be an equilibrium point). Let S be a
suciently small line segment in R
2
through (w

0
), transverse to the curve (w

(t))
at t = 0. Choosing a small enough neighborhood U R
2
of w

0
, we note that the
following fact holds:
1. If a solution v() (v
0
) is such that ( v(0)) U, then the curve ( v(t))
crosses S exactly once near t = 0. Furthermore, the crossing is in the same
direction as that of (w

(t)).
This fact follows directly from Lemma 3.1.3 and observing that the continuous vec-
tor eld F is transverse to S at (w

0
). Let t
n
denote a succession of
20
3.2. More results about z and
negative times for which (w(t
n
)) S. If for any two times , we have that
(w()) = (w( )), then w(t) is a periodic solution by Lemma 3.1.1 and we
have nished this proof. Therefore, we suppose that (w(t
n
)) are mutually dis-
tinct to obtain a contradiction to the Jordan curve Theorem
1
. Hence we consider
the closed Jordan curve formed by (w([t
n+1
, t
n
])) and the interval in S with end-
points (w(t
n
)) , (w(t
n+1
)).
Now, we make some remarks about (v(t)). As seen above, (v(t)) has to cross S in a
given direction, whenever it enters in U. By Lemma 3.1.1, (v(t)) / (w([t
n+1
, t
n
]))
for suciently large and negative times t. Therefore, either (v(t)) stays in the
interior of the Jordan curve dened above or else remains in the exterior of it, for
suciently large and negative times t. However, the curve (v(t)), when t ,
has both (w(t
n
)) and (w(t
n+1
)) as accumulation points by denition. Clearly
this contradicts the properties above and consequently w(t) is a periodic solution in
(v
0
).
3.2 More results about z and
In this section we prepare the remaining proofs, introducing some elementary tech-
nical facts on the behavior of z and along orbits of the semiow (1). The rst
result, Lemma 3.2.1, is needed to prove the other lemmas contained in this section.
Lemma 3.2.2 strengthens the result of Lemma 3.1.1, under an additional assumption.
Thanks to Lemma 3.2.3, we will prove that distinct stationary or periodic orbits have
disjoint projections. Finally, Lemma 3.2.4 states that the function z(v
0
e) worth a
certain integer number for any v
0
(u
0
) and for any equilibrium point e (u
0
).
We proceed to formally expose what we said above.
Lemma 3.2.1.
For any v
0
(u
0
) there exists k
v
0
Z such that
z

v
1
0
v
2
0

= k
v
0
for any v
1
0
, v
2
0
(v
0
) such that v
1
0
,= v
2
0
.
1
Let be a Jordan curve in R
2
. Then exist two sets , simply connected and disjoint,
= , such that R
2
.
21
CHAPTER 3. Proofs
Proof.
In the case when v
0
is an equilibrium point, then the set (v
0
) = v
0
has only one
element and, trivially, the lemma is void. Therefore, by Proposition 1.1.2, it remains
only to prove the case in which (v
0
) is periodic. Then (v
0
) is homeomorphic to
S
1
. Let d(v
0
) := (v, v)[v (v
0
) denote the diagonal in the 2-torus (v
0
) (v
0
).
Consider the following map
Z : ((v
0
) (v
0
)) d(v
0
) Z,

v
1
0
, v
2
0

v
1
0
v
2
0

.
By Lemma 3.1.1 and the continuity assumption (A2), there is a region of ((v
0
) (v
0
))
d(v
0
) where Z is constant. But the 2-torus minus the diagonal d(v
0
) is a connected
set. So, Z is identically constant on all ((v
0
) (v
0
)) d(v
0
).
In the case when t v(t) is injective, the set ((v
0
) (v
0
)) d(v
0
) is formed by
two connected components. The components are the sets of points (v(t
1
), v(t
2
)) with
t
1
> t
2
one, and the other with t
1
< t
2
. However Z (v(t
1
), v(t
2
)) = Z (v(t
2
), v(t
1
))
by injectivity of v. Hence Z is identically constant again and this proves the claim
for v
1
0
, v
2
0
(v
0
).
Finally, we prove indirectly that this lemma still holds for v
1
0
, v
2
0
(v
0
). Suppose
there exist v
1
0
, v
2
0
(v
0
) and also v
0
1
, v
0
2
(v
0
) such that v
1
0
,= v
2
0
, v
0
1
,= v
0
2
, and
z (v
1
0
v
2
0
) < z

v
0
1
v
0
2

. The following solution curves v

1
(), v
2
(), v
1
(), v
2
() are
associated respectively, to these points. We may assume without loss of generality
that z determined in v
1
0
v
2
0
, v
0
1
v
0
2
does not drop at t = 0. By (A2), we may
therefore pick w
1
0
, w
2
0
, w
0
1
, w
0
2
near v
1
0
, v
2
0
, v
0
1
, v
0
2
respectively, such that w
1
0
,= w
2
0
,
w
0
1
,= w
0
2
and
z

w
1
0
w
2
0

= z

v
1
0
v
2
0

< z

v
0
1
v
0
2

= z

w
0
1
w
0
2

.
On the other hand, because of w
1
0
, w
2
0
, w
0
1
, w
0
2
/ (v
0
) (v
0
), we know that
z

w
1
0
w
2
0

= z

w
0
1
w
0
2

.
With this contradiction, the proof is complete.
Lemma 3.2.2.
Let v
0
(u
0
) be such that (v
0
) (v
0
) = and let k
v
0
be dened as in Lemma
22
3.2. More results about z and
3.2.1. Then there exists

t 0 such that
z

u
1
0
v
1
0

= k
v
0
(3.5)
for any u
1
0

+

and any v
1
0
(v
0
) such that u
1
0
,= v
1
0
.
In particular (u
1
0
) = (v
1
0
), for some u
1
0

+

and some v
1
0
(v
0
), implies
that u
1
0
= v
1
0
. Therefore, (u(t)) /

(v
0
)

, t

t.
Proof.
We prove that (3.5) holds for u
1
0

+

u

. This proof is indirect. Suppose, in

fact, there exist sequences v
0
n
(v
0
) with t
n
such that
u(t
n
) ,= v
0
n
,
k
n
:= z (u(t
n
) v
0
n
) ,= k
v
0
. (3.6)
We will get a contradiction by denying the hypothesis (3.6).
Passing to a subsequence, if necessary, we may assume that either k
n
< k
v
0
for all
n, or else k
n
> k
v
0
for all n (thanks to Propositions 1.2.1, 1.2.2 and the fact that
u(t
n
) ,= v
0
n
). We prove separately each of these two cases.
Case 1. k
n
< k
v
0
, n.
Passing to a subsequence again, we have that either v
0
n
(v
0
) for all n, or else
v
0
n

(v
0
)

(v
0
) for all n, by denition of v
0
n
. Now, we can choose either one
of the following alternatives:
1. v
1
0
(v
0
) in case v
0
n
(v
0
).
2. v
1
0
(v
0
) in case v
0
n
(v
0
) (v
0
).
Here we have used the fact that the assumption (v
0
) (v
0
) = implies

(v
0
)

(v
0
) = (v
0
) (v
0
).
Consider that v
1
0
(v
0
). We may pass to a subsequence once more to guarantee
that there exists a sequence

t
n
with t
n
<

t
n
< t
n+1
for which
v
1
0
= lim
n
u

t
n

.
23
CHAPTER 3. Proofs
We also dene
v
2
0
:= lim
n
v
n

t
n
t
n

(3.7)
where v
n
is the solution curve through v
0
n
. The convergence in (3.7) is obtained, if
necessary, considering a further subsequence. In case v
0
n
(v
0
) (v
0
) we may
even assume that

t
n
t
n
is large enough to have v
n

t
n
t
n

+
(v
0
). Note that
v
1
0
, v
2
0
(v
0
). Moreover, v
1
0
,= v
2
0
by the hypothesis (v
0
) (v
0
) = . Indeed, by
construction either v
1
0
(v
0
) and v
2
0
(v
0
) or else v
1
0
(v
0
) and v
2
0

+
(v
0
) =

+
(v
0
) (v
0
). In both cases we have z (v
1
0
v
2
0
) = k
v
0
, by Lemma 3.2.1. Now, we
conclude the proof obtaining a contradiction. Using the axioms (A1) and (A2) we
have that, for suciently large n
k
v
0
= z

v
1
0
v
2
0

= z

t
n

v
n

t
n
t
n

t
n

v
n

t
n
t
n

= z

t
n
+

t
n
t
n

v
n

t
n
t
n

t
n
+

t
n
t
n

v
n

t
n
t
n

z (u(t
n
) v
0
n
) = k
n
.
This contradicts the hypothesis (3.6) for this case, k
n
< k
v
0
, n.
Case 2. k
n
> k
v
0
, n.
For this case we give a simplied version of the proof because it is analogous to the
previous one. Henceforth we pass to subsequences without further notice.
Either v
0
n
(v
0
) for all n, or v
0
n
(v
0
) (v
0
) for all n. Now, we can choose
either one of the following alternatives:
1. v
1
0
(v
0
) in case v
0
n
(v
0
).
2. v
1
0
(v
0
) in case v
0
n
(v
0
) (v
0
).
For a sequence

t
n
with t
n
<

t
n
< t
n+1
we have the convergence of
v
1
0
= lim
n
u

t
n

v
2
0
= lim
n
v
n+1

t
n
t
n+1

,
and in case v
0
n
(v
0
) (v
0
) also v
n+1

t
n
t
n+1

v(t)[t 0. Again v
1
0
, v
2
0

(v
0
) and v
1
0
,= v
2
0
by construction. As before, for large n
k
v
0
= z

v
1
0
v
2
0

= z

t
n

v
n+1

t
n
t
n+1

24
3.2. More results about z and
z

t
n

v
n+1

t
n
t
n+1

= z

t
n+1
+

t
n
t
n+1

v
n+1

t
n
t
n+1

t
n+1
+

t
n
t
n+1

v
n+1

t
n
t
n+1

u(t
n+1
) v
0
n+1

= k
n+1
.
This contradicts the hypothesis (3.6) for this case, k
n
> k
v
0
, n.
As a conclusion of the proof, we observe that the statement for u
1
0

+

follows by continuity (A2).

Lemma 3.2.3.
Let
1
and
2
be stationary or periodic solutions not necessarily distinct. Then there
exists a constant k

1
,
2
such that
z

p
1
0
p
2
0

= k

1
,
2
. (3.8)
for any p
1
0

1
, p
2
0

2
which satises p
1
0
,= p
2
0
.
In particular, the projections of disjoint periodic orbits are disjoint.

1

2
= = (
1
) (
2
) = . (3.9)
Proof.
Let
1
,
2
be periodic solutions. The other cases are analogous and even easier.
The proof presented below is along the lines of that concerning Lemma 3.2.1. The
following function
t

p
1
(t), p
2
(t)

induces a ow on the 2-torus

1

2
. This linear ow is periodic or ergodic. In
either case, for any p
1
0
, p
2
0
such that p
1
0
,= p
2
0
as in the assumption of this lemma, we
have that
k

1
,
2
:= z

p
1
(t) p
2
(t)

does not depend on t, by the assumption (A2). Then the map

p
1
0
, p
2
0

p
1
0
p
2
0

is locally constant on

p
1
0
, p
2
0

2
[p
1
0
,= p
2
0
, (3.10)
by continuity (A2). But the set (3.10) is connected even if
1
=
2
. This proves
equation (3.8). Finally, by assumption (A1), also (3.9) holds, and the proof is
complete.
25
CHAPTER 3. Proofs
Before stating and prove the last result of this section, we briey present a simple
application of the lemma just proved.
Example 2. (About Lemma 3.2.3)
Consider
u
t
= u
xx
+ f (u, u
x
) x S
1
. (3.11)
Taking advantage of the Lemma 3.2.3, we can say that the only periodic solutions
of (3.11) are rotating waves, u = u(x ct). In fact, by contradiction, suppose that
u = u(x, t) is a periodic solution, but is not a rotating wave. Then
T
2
:= u(, + ) [ R, S
1
X
is a 2-torus foliated by periodic solutions. Indeed, the ow on T
2
is given by a shift
in . By Lemma 3.2.3, the projection
: T
2

T
2

R
2
would be a homeomorphism onto its planar image. But, this is clearly impossible.
Therefore periodic solutions are rotating waves for the equation (3.11).
We conclude this section with the following result.
Lemma 3.2.4.
There exists k
0
Z such that
z (v
0
e) = k
0
holds for any v
0
(u
0
) and for any equilibrium point e (u
0
) such that v
0
,= e.
Proof.
Dene
k
e
:= lim
t
z (u(t) e) .
By continuity (A2), it is sucient to prove that k
e
does not depend on e (u
0
)E.
Therefore, let e (u
0
) E be another equilibrium point. Then, by Lemma 3.2.3,
z is locally constant at e e and also at e e. Thus there exist t and

t large enough
that
k
e
= z

= z ( e e) = z (e e) = z (u(t) e) = k
e
.
This proves the lemma.
26
3.3. Proof of Proposition 1.1.3
3.3 Proof of Proposition 1.1.3
This section provides the proof of Proposition 1.1.3.
Proof.
By Proposition 1.1.2, (v
0
) contains a periodic solution or an equilibrium point.
Therefore, suppose that v(t) is not periodic and proceed with an indirect proof.
Suppose, in fact, there exists w

0
(v
0
) such that w

0
/ E, i.e. w

0
is not an
equilibrium point. We omit the analogous case w

0
(v
0
) for brevity. Note that,
by Lemma 3.1.1, t (v(t)) is also not periodic, but rather it is bijective for t R.
As in the proof of Proposition 1.1.2, let S denote a suciently small line segment
in R
2
through (w

0
), transverse to (w

(t)) at t = 0. In the proof of Proposition

1.1.2, we have already seen that (v(t)) crosses S in the same direction as (w

(t))
near t = 0. To conclude this proof we strengthen this result.
1. Let S be a suciently small line segment in R
2
through (w

0
) and let R be
suciently large with (u()) S. Moreover, we suppose that (v
0
)(v
0
) =
. Then (u()) crosses S at t = . Furthermore, the crossing is in the same
direction as that of (w

(t)) at t = 0.
Indeed, (u(
n
)) S converge to (w

0
) for a sequence
n
. Then
d
dt

t=
n
(u(t))
d
dt

t=0
(w

(t)) , n
as seen in the proof of the Lemma 3.1.3. Thanks to the hypothesis (v
0
)(v
0
) = ,
that allows us to exploit the result of the Lemma 3.2.2, we conclude that u(
n
)
converge to w

0
.
Now, before applying this result, we show that (v
0
) (v
0
) = holds under the
assumptions made in this demonstration. In fact, let t
n
denote a succession
of positive times for which (v(t
n
)) S. Recalling that t (v(t)) is bijective
and that (v(t)) is not periodic, we note that (v(t
n
)) are pairwise disjoint. Now
consider the Jordan curve consisting of (v ([t
n
, t
n+1
])) together with the subinterval
of S with endpoints (v(t
n
)) , (v(t
n+1
)). By Lemma 3.1.1 and for the fact that
(v(t)) crosses S exactly once near t = 0 , one of the sets ((v
0
)) , ((v
0
)) is in
the interior of that Jordan curve while the other set is in the exterior. Therefore
(v
0
) (v
0
) = .
27
CHAPTER 3. Proofs
Now, we can apply Lemma 3.2.2 and the results presented in this proof. In particular
we consider the curve (
+
(u(t

))). As t , this curve has accumulation points

both in ((v
0
)) and in ((v
0
)). But, by Lemma 3.2.2, (
+
(u(t

))) can not

intersect

(v
0
)

. This contradiction concludes the proof.

3.4 Proof of Proposition 1.1.4
This section consists of the proof of Proposition 1.1.4.
Proof.
This proof is indirect. Suppose that (u
0
) contains a periodic solution (p
1
0
). Ex-
ploiting Proposition 1.1.3, we can construct two other periodic solutions, (p
2
0
) , (p
3
0
)
(u
0
), which are also contained in a small closed neighborhood V of (p
1
0
). We
choose V so small that the projections

p
j
0

, j = 1, 2, 3
are nested Jordan curves. These curves are pairwise disjoint by Lemma 3.2.3. We
will prove that (u(t)) can not intersect any of the curves

p
j
0

for t suciently
large and, obviously, this contradicts

p
j
0

(u
0
) for j = 1, 2, 3.
We dene p
1
0
, p
2
0
, p
3
0
in such a way as to obtain the contradiction. The set V should
not contain equilibrium points and (u
0
) should still have elements outside of V .
Therefore u(t) must enter and leave V innitely often. Consider a sequence t
n

such that
p
1
0
= lim
n
u(t
n
)
and such that u(t) leaves V between any two consecutive times. Then let I
n
=
[t
n

n
, t
n
+
n
] be the maximal time interval containing t
n
such that
u(t) V, t I
n
.
Passing to a subsequence, if necessary, we may assume the convergence of u(t
n

n
)
along the border of V , V . Note that t
n

n
. Moreover, since (u
0
) contains
a periodic orbit in the interior of V , we have that
n
+
n
(because
n
is nite
and
n
). We dene
p
2
0
= lim
n
u(t
n

n
) (u
0
).
28
3.5. Proof of Theorem 1.3.1
Note that p
2
0
V and, since
n
+
n
and V is closed,

p
2
0

+
(p
2
0
) V. (3.12)
Then, as V does not contain any equilibrium point, exploiting (3.12) and Proposition
1.1.3 we have that (p
2
0
) is periodic. Obviously (p
1
0
) and (p
2
0
) are distinct. Finally,
we can construct the periodic solution (p
3
0
) in the same way as (p
2
0
).
Now, to complete the proof, it remains to show that there exists a t

> 0 such that

(u(t)) /

p
j
0

, for any t t

and for any j = 1, 2, 3. We shall prove this fact

only for j = 1 because the other cases are analogous. Applying Lemma 3.2.3 to the
disjoint periodic orbits (p
1
0
) , (p
2
0
) we conclude that
z

q
1
0
q
2
0

= k
independently of the choice of q
1
0
(p
1
0
) , q
2
0
(p
2
0
) and for k R. By (A2) we
have that
z

q
1
0
u(t
n

n
)

= k (3.13)
for all q
1
0
(p
1
0
) and for all suciently large n, since u(t
n

n
) p
2
0
(p
2
0
). By
assumption (A1) and by (3.13), the projections of (p
1
0
) and u(t) are disjoint when
t
n

n
. Indeed, (u(t)) ( (p
1
0
)) can not happen for any open interval with
endpoints t
n

n
, t
m

m
, provided n, m are chosen large enough. This completes
the proof of Proposition 1.1.4, and thereby the proof of Theorem 1.1.1.
3.5 Proof of Theorem 1.3.1
This section presents the proof of Theorem 1.3.1.
Proof.
By the z-dropping assumption (A1) it is sucient to show that there exists a k

such that
z

v
1
0
v
2
0

= k

(3.14)
for all v
1
0
, v
2
0
(u
0
) with v
1
0
,= v
2
0
.
Applying Theorem 1.1.1, we can consider two alternative situations related to (u
0
).
If (u
0
) consists in a single periodic orbit, then (3.14) holds by Lemma 3.2.3.
29
CHAPTER 3. Proofs
It remains to prove only that (3.14) holds with (v
0
), (v
0
) E for any v
0
(u
0
)
(E is the set of equilibrium points). Consider k

= k
0
where k
0
is the same constant
dened in Lemma 3.2.4. Moreover, thanks to Lemma 3.2.4, we know that (3.14)
holds whenever v
1
0
or v
2
0
is an equilibrium point. Therefore, we assume that v
1
0
, v
2
0
/
E. In this case we prove (3.14) indirectly.
Suppose that
z

v
1
0
v
2
0

,= k
0
. (3.15)
Following the time ow t (v
1
(t) v
2
(t)) in a suitable direction, we may assume
z to be constant near v
1
0
v
2
0
by axioms(A1), (A2). We have to consider two cases.
Case 1. z (v
1
0
v
2
0
) > k
v
0
.
Let t
n
be a succession such that 0 < (t
n+1
t
n
) as n , and
v
1
0
= lim
n
u(t
n
).
Let
n
= (t
n+1
t
n
) . Passing to subsequence, if necessary, we may assume
that
e = lim
n
v
2
(
n
)
exists. Note that e (v
2
0
) E, therefore e is an equilibrium point. For n large
enough, this implies
z

v
1
0
v
2
0

= z

u(t
n+1
) v
2
(0)

u(t
n+1
+
n
) v
2
(
n
)

u(t
n+1
+
n
) v
2
(
n
)

= z

u(t
n
) v
2
(
n
)

= z

v
1
0
e

= k
0
where we have used the fact that z is constant near v
1
0
e by continuity (A2) and
by Lemma 3.2.4. This contradicts z (v
1
0
v
2
0
) > k
v
0
.
Case 2. z (v
1
0
v
2
0
) < k
v
0
.
This case is analogous to the previous one. Dene t
n
as before and let
n
=
t
n+1
t
n
+. Passing to a subsequence, if necessary, there exists
e = lim
n
v
2
(
n
).
Note that e (v
2
0
) E, therefore e is an equilibrium point. For n large enough,
this implies
z

v
1
0
v
2
0

= z

u(t
n
) v
2
(0)

u(t
n
+
n
) v
2
(
n
)

30
3.5. Proof of Theorem 1.3.1
z

u(t
n
+
n
) v
2
(
n
)

= z

u(t
n+1
) v
2
(
n
)

= z

v
1
0
e

= k
0
contradicting z (v
1
0
v
2
0
) < k
v
0
. This complete the proof of Theorem 1.3.1.
31
CHAPTER 3. Proofs
32
Chapter 4
Other Results & Open Questions
In this chapter we briey introduce some other results for equations which admit a
discrete Lyapunov functional like z. We will see that, under suitable assumptions,
some of these equations have additional special properties. Lets start by treating
the case about the reaction diusion equation (1).
4.1 Reaction Diusion Equation
4.1.1 Dirichlet or Neumann boundary conditions
Consider
u
t
= u
xx
+ f (x, u, u
x
) (4.1)
with Dirichlet or Neumann boundary conditions. The set (u
0
) consists of a single
equilibrium point if
+
(u
0
) is any bounded orbit. For further details about these
results see Zelen-Yak and Matano 2. Again for Dirichlet or Neumann conditions,
but for f depends on u only, f = f(u), the question of connecting heteroclinic orbits
between equilibrium points has been studied by use of the functional z in Henry,
Brunovsky & Fiedler 2,3, Fusco & Rocha.
4.1.2 Periodic boundary conditions and f independent of x
For equation (4.1) with periodic boundary conditions and for f = f (u, u
x
) inde-
pendent of x and analytic, the soft version of the Poincare-Bendixson theorem was
proved in Agenent & Fiedler, and stronger versions were suspected to hold. In this
33
CHAPTER 4. Other Results & Open Questions
particular case all periodic solutions turn out to be rotating waves, i.e. solutions of
the form u = u(x ct). Indeed, it is shown in Massatt that either (u
0
) is a single
rotating wave, or a set of equilibrium points which diers only by shifting x (the
same result has also been obtained in Matano 3). The fact that f does not depend
explicitly on x is crucial for the previous results. This, in fact, makes the equation
(4.1) S
1
-equivariant
1
. Therefore, we may transform the equation (4.1) to rotating
coordinates, u(t, x) = u(t, x + ct), and apply Theorem 1.1.1 to the equation
u
t
= u
xx
+ f ( u, u
x
) + c u
x
, x S
1
.
Then the stationary solutions become rotating waves for c ,= 0 and Massatts result
follows because if the set ( u
0
) contains some periodic solutions, then it contains
only a single periodic solution. In the case that the system (4.1) is O(2)-equivariant,
periodic solutions cannot occur. Therefore (u
0
) is a single equilibrium point, up
to x-shift (this result is presented in Matano 3). With Theorem 1.1.1 at hand, the
results about connections of heteroclinic orbit between rotating waves, carry over to
the more general case f = f (x, u, u
x
), see again Angenent & Fiedler. Moreover, in
Angenent is proved that analyticity of f can be weakened to f C
2
.
4.2 Monotone Cyclic Feedback Systems
As previously mentioned (see the end of chapter 1), there are systems of equations
that ts in our axiomatic setting (A0 A3). These are called monotone cyclic
feedback systems and are of the type
u
i
= f
i
(u
i
, u
i1
) , u
i
R, 0 i N (4.2)
where N is the dimension of the system. For the smooth nonlinearities of these
systems we require

i
f
i
(, )

> 0, , , i
where
i
+1, 1, i. Note that in case the product
1

2
. . .
N
= +1, the system
(4.2) denes monotone semiows. The solutions of these systems are dened on
X = R
N
. In order to adapt these systems to our axiomatic structure must be
1
Equivariant dynamical systems are dynamical systems that have symmetries (a symmetry of
a dynamical system is a transformation that sends solutions to other solutions).
34
4.3. Monotone Feedback Delay Equation
dened a function z and a projection . Let z be such that z(0) = 0 and for u ,= 0
the function is given by
z(u) = #i [ k 1 such that u
i
,= 0, u
ik
,= 0, u
ij
= 0 j with 1 j k 1
and also (
i

i1
. . .
ik1
) u
i
u
ik
< 0.
We also dene the projection
: X R
2
u (u
i1
, u
i
)
for some xed i. Then (A0 A3) hold. For the systems (4.2) in Mallet-Paret &
H.Smith are shown directly Theorems 1.1.1 and 1.3.1. Therefore, a theorem of the
Poincare-Bendixson type does hold for the monotone cyclic feedback systems.
4.3 Monotone Feedback Delay Equation
Consider the monotone feedback delay equation
u(t) = f (u(t), u(t 1)) , u R. (4.3)
For the smooth nonlinearities we require

f (, )

> 0 ,
where 1, +1. In case = +1, equation (4.3) denes a monotone semiow.
The set X, on which are dened the solutions, consists of C

functions with t R.
For the delay equation (4.3) a discrete Lyapunov functional z was studied in Mallet-
Paret 1. In that z (u()) is dened as the number of zeros of u() in [1, ], counting
multiplicity, and 0 denote the smallest zero of u(). If does not exists then
z (u()) := 1. We dene also the following projection
: X R
2
u (u(0), u(1))
In this specic case, z and are not adaptable to axioms (A0 A3). However, after
appropriate modications, Theorems 1.1.1 and 1.3.1 should be true, see Mallet-
Paret & Sell 1 for those. For results on heteroclinic orbits in delay equations with
a functional z see Mallet-Paret 2, Mischaikow and Fiedler & Mallet-Paret 1.
35
CHAPTER 4. Other Results & Open Questions
4.4 Two Open Questions
As informational, we would like to mention an open question about the reaction
diusion equations. Consider
u
t
=
2
u
xx
+ f (x, u, u
x
) , x S
1
(4.4)
and let tend to zero. Regardless of the choice of f we do not know if it is worth a
theorem of the Poincare-Bendixson type or not for the limiting nonlinear hyperbolic
equation (4.4).
Another open question concerns the dynamics of
u
t
= u + f (x, u, u) (4.5)
for x in some connected domain of dimension at least 2. What is still not clear is
the relationship between the domain and the dynamics of (4.5). In fact, in contrast
to the case of scalar x, no analogue of z is known for equation (4.5) and we noticed
the importance of this function in the study of the dynamic of (1) throughout this
document.
36
Bibliography
[1] S.Angenent, The zeroset of a solution of a parabolic equation, J. reine angew.
Math. 390 (1988)
[2] S.Angenent & B.Fiedler, The dynamics of rotating waves in scalar reaction
diusion equations, Trans. AMS 307 (1988)
[3] P.Brunovsky & B.Fiedler 1, Zero numbers on invariant manifolds in scalar
reaction diusion equations, Nonlinear Analysis TMA 10 (1986)
[4] P.Brunovsky & B.Fiedler 2, Connecting orbits in scalar reaction diusion equa-
tions, Dynamics Reported 1 (1988)
[5] P.Brunovsky & B.Fiedler 3, Connecting orbits in scalar reaction diusion equa-
tions 2: the complete solution, J. Dierential Equations 81 (1989)
[6] B.Fiedler & J.Mallet-Paret 1, Connections between Morse sets for delay-
dierential equations (1987)
[7] B.Fiedler & J.Mallet-Paret 2, A Poincare-Bendixson theorem for scalar reaction
diusion equations (1988)
[8] G.Fusco & C.Rocha, A permutation related to the dynamics of a scalar parabolic
PDE, J. Dierential Equations 91 (1991)
[9] M.Fusco & P.Marcellini & C.Sbordone, Analisi Matematica due, Liguori Edi-
tore, S.r.l. (1996)
[10] D.Henry, Some innite dimensional Morse-Smale systems dened by parabolic
dierential equations, J. Dierential Equations 59 (1985)
37
BIBLIOGRAPHY
[11] J.Mallet-Paret 1, Morse decompositions for delay-dierential equations, J. Dif-
ferential Equations 72 (1988)
[12] J.Mallet-Paret 2, Morse decompositions and global continuation of periodic
solutions for singularly perturbed delay equations, in Systems of Nonlinear
Partial Dierential Equations, J.M.Ball, Dordrecht (1983)
[13] J.Mallet-Paret & H.Smith, The Poincare-Bendixson theorem for monotone
cyclic feedback systems, J. Dierential Equations 4 (1990)
[14] J.Mallet-Paret & G.R.Sell 1, Inertial manifolds associated to partly dissipative
reaction-diusion systems, J. Math. Analysis and Applications 143 (1989)
[15] J.Mallet-Paret & G.R.Sell 2, The Poincare-Bendixson theorem for monotone
cyclic feedback systems with delay, J. Dierential Equations (1994)
[16] P.Massat, The convergence of scalar parabolic equations with convection to
periodic solutions (1986)
[17] H.Matano 1, Nonincrease of the lap-number of a solution for a one-dimensional
semilinear parabolic equation, J. Fac. Sci. Univ. Tokyo Sec. IA 29 (1982)
[18] H.Matano 2, Convergence of solutions of one-dimensional parabolic equations,
J. Math. Kyoto Univ. 18 (1978)
[19] H.Matano 3, Asymptotic behavior of solutions of semilinear heat equations on
S
1
, in Nonlinear Diusion Equations and Their Equilibrium States 2, W.-M.
Ni & B.Peletier & J.Serrin, Springer-Verlag, New York (1988)
[20] K.Mishaikow, Conleys connection matrix, in Dynamics of Innite Dimensional
Systems , S.N. Chow & J.K.Hale, Springer-Verlag, Berlin (1987)
[21] M.Squassina & S.Zuccher, Introduzione allanalisi qualitativa delle equazioni
dierenziali ordinarie, APOGEO S.r.l. (2008)
[22] T.I.Zelen-Yak, Stabilization of solutions of boundary value problems for a sec-
ond order parabolic equation with one space variable, J. Dierential Equations
4,1 (1968)
38