Академический Документы
Профессиональный Документы
Культура Документы
Radoslaw Czaja
a,b
, Carlos Rocha
b,
a
Institute of Mathematics, University of Silesia, Bankowa 14, 40-007 Katowice, Poland
b
Centro de Anlise Matemtica, Geometria e Sistemas Dinmicos, Departamento de Matemtica,
Instituto Superior Tcnico, Av. Rovisco Pais, 1049-001 Lisboa, Portugal
Received 1 August 2007; revised 17 January 2008
Available online 6 March 2008
Abstract
We prove that stable and unstable manifolds of hyperbolic periodic orbits for general scalar reaction
diffusion equations on a circle always intersect transversally. The argument also shows that for a periodic
orbit there are no homoclinic connections. The main tool used in the proofs is Matanos zero number theory
dealing with the Sturm nodal properties of the solutions.
2008 Elsevier Inc. All rights reserved.
MSC: primary 35B10; secondary 35B40, 35K57
Keywords: Periodic orbit; Heteroclinic orbit; Transversality; Global attractor; Zero number
1. Introduction
We consider the scalar reactiondiffusion equation of the form
u
t
=u
xx
+f (x, u, u
x
) (1.1)
for one real variable u = u(t, x) on a circle x S
1
=R/2Z. In other words, we consider (1.1)
together with periodic boundary conditions
Partially supported by the Fundao para a Cincia e a Tecnologia through the Program POCI 2010/FEDER and by
the Project PDCT/MAT/56476/2004.
*
Corresponding author.
E-mail addresses: czaja@math.ist.utl.pt (R. Czaja), crocha@math.ist.utl.pt (C. Rocha).
0022-0396/$ see front matter 2008 Elsevier Inc. All rights reserved.
doi:10.1016/j.jde.2008.01.018
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 693
u(t, 0) =u(t, 2), u
x
(t, 0) =u
x
(t, 2)
and discuss (1.1) with initial condition
u(0, x) =u
0
(x), x S
1
. (1.2)
Belowwe use suitable assumptions on f so that the problem(1.1), (1.2) denes a global semiow
in X
=H
2
(S
1
),
3
4
< < 1, for which there exists a global attractor, i.e. a nonempty compact
invariant set attracting every bounded subset of X
,
3
4
< <1, then either its -limit set (u
0
) consists in precisely one periodic orbit or (v
0
) and
(v
0
) are subsets of the set of all equilibria for any v
0
(u
0
).
In this paper we investigate closely the situation when a bounded orbit from the global at-
tractor connects two hyperbolic periodic orbits. First, we exclude the existence of a homoclinic
connection for a hyperbolic periodic orbit (cf. [23]) in order to nally prove the main result of this
paper stating that the intersection of the global unstable manifold of a hyperbolic periodic orbit
) W
s
loc
(
+
).
The paper is organized as follows. In Section 2 we formulate the abstract Cauchy problem
for (1.1)(1.2) and using the theory from [14] we solve the problem locally. Further we obtain
a priori and subordination estimates, which ensure that the solutions exist globally in time. The
semiow of global solutions constructed in this way is point dissipative and compact, thus has a
compact global attractor. In Section 3 we examine the properties of the semiow and the evolu-
tion system for the linearization around a given solution. Moreover, we recall the properties of
the zero number of solutions of linear parabolic equations. Section 4 is devoted to the operator
called a period map for a periodic orbit. We describe its spectrum and decompose the phase space
694 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
according to the spectrum. We also recall the notions of local stable and unstable manifolds of a
hyperbolic periodic orbit and list their properties. In Section 5 we analyze the local stable mani-
fold of a hyperbolic periodic orbit and show that for any u
0
/ from the local stable manifold
of there exists a such that u(t ; u
0
) p(t ; a) tends exponentially to 0 as t and
z(u
0
a) i() +1 +
1 +(1)
i()
2
, (1.4)
where i() denotes the total algebraic multiplicity of eigenvalues of the period map for out-
side the closed unit ball. Similarly, in Section 6 we investigate the global unstable manifold of
a hyperbolic periodic orbit . We prove that for any u
0
/ from the global unstable manifold
there exists a such that u(t ; u
0
) p(t ; a) tends exponentially to 0 as t and
z(u
0
a) i() 1 +
1 +(1)
i()
2
. (1.5)
In Section 7 we combine the estimates (1.4) and (1.5) and nd, in particular, that there is no
homoclinic connection for a hyperbolic periodic orbit. Finally, in Section 8 we follow the ideas
from [6] and introduce ltrations of the phase space with respect to the asymptotic behavior of
solutions for the linearized equation around an orbit connecting two hyperbolic periodic orbits.
A proper choice of the spaces from the ltrations carefully combined with the corresponding zero
number estimates for the functions from these spaces yields the transversality of the intersection
of the stable and unstable manifolds of two hyperbolic periodic orbits. The transversal intersec-
tion of invariant manifolds of critical elements is one of the ingredients for genericity results (cf.
e.g. KupkaSmale theorem) or structural stability theorems (cf. [13, Chapter 10], [24]) in the
theory of dynamical systems. In Section 9 we make some concluding remarks about structural
stability for the semiow generated by (1.1).
Under different boundary conditions many authors have considered problems of the same type
as discussed here. For separated boundary conditions, the results of Henry [15] and Angenent
[1] on the transversality of the stable and unstable manifolds of stationary solutions constitute
obligatory references. A problem of this type has also been considered by Chen, Chen and Hale
in [6] for nonautonomous time periodic equations with f =f (t, x, u) under Dirichlet boundary
conditions. The effect of radial symmetry on the transversality of stable and unstable manifolds of
equilibria for problems dened on symmetric domains in R
n
has been studied by Pol cik in [26].
For special classes of ordinary differential equations on R
n
, Fusco and Oliva have considered the
transversality between stable and unstable manifolds of equilibria and periodic orbits (see [10,
11]). Here we extend the results of [11] realizing the plans sketched by these authors for further
possible extensions.
2. Abstract setting of the problem and existence of the global attractor
Assume that f : S
1
RRR is a C
2
function satisfying the following conditions
there exist 0 <2 and a continuous function k : [0, ) [0, ) such that
f (x, y, z)
k(r)
_
1 + |z|
_
, (x, y, z) S
1
[r, r] R for each r >0, (2.1)
yf (x, y, 0) <0, (x, y) S
1
R, |y| K for some K >0. (2.2)
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 695
In this paper we are going to use fractional Sobolev spaces of 2-periodic functions H
s
(S
1
),
s > 0, and their properties (cf. [29, Appendix A]). Among others we will frequently use the
Sobolev embedding
H
s
_
S
1
_
C
1
_
S
1
_
if s >
3
2
.
We consider the operator A: L
2
(S
1
) H
2
(S
1
) L
2
(S
1
) given by
Au =u
xx
+u, u H
2
_
S
1
_
.
Since A is a positive denite self-adjoint operator, it is a positive sectorial operator. Henceforth
we consider fractional power spaces
X
=D
_
A
_
, 0,
with norms u
X
= A
u
L
2
(S
1
)
, u X
=
_
L
2
_
S
1
_
, H
2
_
S
1
__
=H
2
_
S
1
_
, (0, 1)
(see [35, Section 1.18.10] and [32, Section 3.6.1]). Since H
2
(S
1
) is compactly embedded in
L
2
(S
1
), it follows that A has a compact resolvent.
We rewrite (1.1), (1.2) as an abstract Cauchy problem in X
0
_
u
t
+Au =F(u),
u(0) =u
0
,
(2.3)
where F is the Nemycki operator corresponding to
F(u)(x) =f
_
x, u(x), u
x
(x)
_
+u(x), x S
1
.
For a xed (
3
4
, 1), F takes X
into X
0
and is Lipschitz continuous on bounded subsets of X
.
By the theory presented in [14] it follows that for each u
0
X
_
C
1
_
(0,
u
0
), X
0
_
C
_
(0,
u
0
), X
1
_
and satises (2.3) on [0,
u
0
) in X
0
. Moreover, either
u
0
= or
u
0
< and limsup
t
u
0
_
_
u(t ; u
0
)
_
_
X
=.
Using assumption (2.2) and the maximum principle it follows that if for some R 0 we have
u
0
(S
1
)
K +R, then there exists a positive constant =(K, R) such that
_
_
u(t ; u
0
)
_
_
L
(S
1
)
K +Re
t
, t [0,
u
0
). (2.4)
This implies that each forward X
solution is bounded in L
(S
1
).
696 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
Observe that by using Youngs inequality we can assume without loss of generality that
1 < <2 in (2.1). Applying (2.4) to (2.1), we obtain
_
_
F
_
u(t ; u
0
)
__
_
X
0
c
_
u
0
(S
1
)
__
1 +
_
_
u(t ; u
0
)
_
_
W
1,2
(S
1
)
_
, t (0,
u
0
). (2.5)
Fix r >max{
1
2
, 2} and let be such that
1
2
_
+
1
r
( 1)
_
< <1.
Then for a chosen
_
1 +
1
r
1
2
2 +
1
r
1
2
,
1
_
such that
r2
(r2)
the following interpolation inequality is satised
_
_
u(t ; u
0
)
_
_
W
1,2
(S
1
)
c
_
_
u(t ; u
0
)
_
_
H
2
(S
1
)
_
_
u(t ; u
0
)
_
_
1
L
r
(S
1
)
, t (0,
u
0
), (2.6)
due to the embedding (cf. [32, Section 3.6.1])
_
L
r
_
S
1
_
, H
2
_
S
1
__
W
1,2
_
S
1
_
.
Since X
= H
2
(S
1
) and L
(S
1
) is continuously embedded in L
r
(S
1
), we obtain from (2.5),
(2.6) and again (2.4) the following subordination condition
_
_
F
_
u(t ; u
0
)
__
_
X
0
c
_
u
0
(S
1
)
__
1 +
_
_
u(t ; u
0
)
_
_
_
, t (0,
u
0
), (2.7)
with <1.
By [7, Theorem 3.1.1] it follows that each forward X
(S
1
) is asymptotically inde-
pendent of initial conditions
limsup
t
_
_
u(t ; u
0
)
_
_
L
(S
1
)
K.
Then by [7, Theorem 4.1.1] there exists a constant K
1
>0 such that
limsup
t
_
_
u(t ; u
0
)
_
_
X
K
1
. (2.8)
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 697
Therefore the semiow {S(t ): t 0} is point dissipative in X
for each t > 0 by [7, Theorem 3.3.1], since A has a compact resolvent. Thus the
semiow {S(t ): t 0} has a global attractor A in X
(t, ) u(t ; s, ) X
u(t ; s, ) X
u(t ; s, ) X
, t s,
is a unique global forward X
, (3.4)
where w(t ; s, ) is a unique global forward X
, X
and some
t
0
>0 we have
S(t
0
)u
1
=S(t
0
)u
2
.
Dene v(t ) =S(t )u
1
S(t )u
2
, t 0. Then we have
_
v
t
+Av =F
_
S(t )u
1
_
F
_
S(t )u
2
_
, t >0,
v(0) =u
1
u
2
.
(3.5)
Moreover, we know that v(t
0
) = 0. Note that A is a positive denite self-adjoint operator in the
Hilbert space X
0
=L
2
(S
1
) and X
X
1
2
. Furthermore,
v C
_
[0, ), X
_
C
1
_
(0, ), X
0
_
C
_
(0, ), X
1
_
and
_
_
F
_
S(t )u
1
_
F
_
S(t )u
2
__
_
X
0
L
_
_
S(t )u
1
S(t )u
2
_
_
X
1
2
=L
_
_
v(t )
_
_
X
1
2
, t [0, ),
where L is a constant depending on
sup
t [0,)
_
_
S(t )u
i
_
_
X
<, i = 1, 2.
By [7, Proposition 7.1.1] (see also [34, Lemmas 6.1, 6.2]) we get
v(t ) = 0, t [0, t
0
].
In particular, we obtain u
1
=u
2
. This proves the injectivity of the semiow.
Suppose now that
T (t
0
, s
0
) = 0
for some t
0
>s
0
and X
. Dene w(t ) =T (t +s
0
, s
0
) , t 0, and choose any T
0
>t
0
s
0
.
Then we have
_
w
t
+Aw =D
u
F
_
u(t +s
0
; s
0
, )
_
w, 0 <t T
0
,
w(0) =.
(3.6)
Moreover, we know that w(t
0
s
0
) = 0. For t [0, T
0
] we estimate
_
_
D
u
F
_
u(t +s
0
; s
0
, )
_
w(t )
_
_
X
0
C
1
_
_
w(t )
_
_
X
0
+C
2
_
_
w(t )
_
_
X
1
2
M
_
_
w(t )
_
_
X
1
2
,
where C
1
and C
2
depend on
sup
(t,x)[0,T
0
]S
1
f
y
_
x, u(t +s
0
; s
0
, )(x), u
x
(t +s
0
; s
0
, )(x)
_
,
sup
(t,x)[0,T
0
]S
1
f
z
_
x, u(t +s
0
; s
0
, )(x), u
x
(t +s
0
; s
0
, )(x)
_
,
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 699
respectively. Thus the assumptions of [7, Proposition 7.1.1] are fullled again and
w(t ) = 0, t [0, t
0
s
0
].
In particular, we obtain = 0. This proves the injectivity of T (t, s), t s.
Observe also that by the backward uniqueness of the adjoint equation of (3.3) (see [17]) the
adjoint operator T (t, s)
is injective and by [14, Theorem 7.3.3] each operator T (t, s), t s, has
a dense range.
In what follows we are going to use frequently the properties of the zero number of a C
1
func-
tion referring to the Sturm nodal properties of the solutions of (3.7) (see [33]) so successfully
reintroduced by Matano (cf. [19]) as an essential tool for the description of the dynamics of
scalar semilinear parabolic equations. We denote by z() the number of strict sign changes of
a C
1
function : S
1
R. Then, as a consequence of the maximum principle, the zero number
has the following properties.
Lemma 3.1. (See [21, Lemma 3.2], [2].) Let J R be an open interval and v be a nontrivial
classical solution of the linear parabolic equation
v
t
=v
xx
+b(t, x)v +d(t, x)v
x
, t J, x S
1
, (3.7)
where b, b
x
, b
t
and d are bounded on any compact subset of J S
1
, then the zero number of v(t )
has the following properties:
(i) z(v(t )) is nite for any t J,
(ii) z(v(t )) is nonincreasing in t on J,
(iii) z(v(t )) drops strictly at t =t
0
if and only if
S
1
x v(t
0
)(x) R
has a multiple zero.
Observe that the assertions of this lemma hold for the zero number of the difference of two
different solutions for a scalar semilinear parabolic equation.
Lemma 3.2. (See [21, Lemma 3.4].) If u
1
and u
2
are two different X
=T (, 0) =D
u
0
S()a is called a period map (cf. [14, Denition 7.2.1])
and the function w(t ) =T (t, 0) satises the linear nonautonomous equation
w
t
=w
xx
+b(t, x)w +d(t, x)w
x
, t >0, x S
1
, (4.1)
with
b(t, x) =f
y
_
x, p(t )(x), p
x
(t )(x)
_
, d(t, x) =f
z
_
x, p(t )(x), p
x
(t )(x)
_
.
Since T
, the
spectrum (T
) of T
= D
u
0
S()a
does not depend on the choice of the periodic point a , but the eigenfunctions do depend
on a. Observe also that 1 is always a characteristic multiplier with the corresponding eigenfunc-
tion p
t
(0) X
1
(cf. [14, Lemma 8.2.2]). If 1 is a simple eigenvalue of T
=
1
2i
_
(I T
)
1
d, {s, c, u},
where
s
=
_
(T
): || <1
_
,
c
=
_
(T
): || = 1
_
,
u
=
_
(T
): || >1
_
,
respectively. Note that dimP
u
X
, called the Morse index i(), is nite and equals the total
algebraic multiplicity of multipliers outside the closed unit ball. Similarly dimP
c
X
is nite and
equals the total algebraic multiplicity of multipliers on the unit circle.
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 701
Observe that X
= P
u
X
P
c
X
P
s
X
and P
and
(T
|
P
X
) =
.
Furthermore, the eigenvectors of T
belong to X
1
, so P
X
1
, {c, u}. Moreover, T
maps
bijectively P
u
X
onto P
u
X
and P
c
X
onto P
c
X
.
Assume that is a hyperbolic periodic orbit. Consequently, we have P
c
X
= span{p
t
(0)}.
We consider the Poincar map P
a
for the semiow {S(t )} corresponding to the cross section
a + P
u
X
+ P
s
X
(see [14, Section 8.4], [30, Section 4.1]). Then the spectrum of the tangent
map to P
a
at a is equal to (T
: S(t )u
0
U, t 0
_
is a C
1
submanifold of X
with codimW
s
loc
() =i(), whereas the local unstable manifold of
in U dened by
W
u
loc
() =
_
u
0
X
:
{u
s
}
s0
S(t )u
s
=u
t s
, 0 t s and u
s
U, s 0
_
is a C
1
submanifold of X
with dimW
u
loc
() =i() +1.
Moreover, W
s
loc
() is brated by local strong stable manifolds at each a
W
s
loc
() =
_
a
W
ss
loc
(a)
and W
u
loc
() by local strong unstable manifolds at each a
W
u
loc
() =
_
a
W
su
loc
(a),
where, for sufciently small >0, we have the following characterizations with certain ,
>0
W
ss
loc
(a) =
_
u
0
X
:
_
_
S(t )u
0
S(t )a
_
_
X
= 0
_
,
W
su
loc
(a) =
_
u
0
X
:
{u
t
}
t 0
_
_
u
t
S(t )
1
a
_
_
X
< for t 0,
S(r)u
s
=u
rs
for 0 r s and lim
t
e
t
_
_
u
t
S(t )
1
a
_
_
X
= 0
_
.
From [30, Section 15.2] (see also [4]) it follows that for each a , W
ss
loc
(a) is a C
1
submanifold
of X
tangent at a to P
s
X
and W
su
loc
(a) is a C
1
submanifold of X
tangent at a to P
u
X
.
702 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
5. Local stable manifold of a hyperbolic periodic orbit
In this section we consider a hyperbolic parabolic orbit with period > 0 and show that
for any u
0
W
s
loc
() \ there exists a such that u(t ; u
0
) p(t ; a) tends exponentially to
0 as t and
z(u
0
a) i() +1 +
1 +(1)
i()
2
. (5.1)
Choose u
0
W
s
loc
() \ and let a be such that u
0
W
ss
loc
(a). We consider the corre-
sponding solutions u(t ) = S(t )u
0
and p(t ) = S(t )a of (2.3). Let v(t ) = u(t ) p(t ), t 0, and
note that v satises the nonautonomous linear equation
1
v
t
=v
xx
+
b(t, x)v +
d(t, x)v
x
, t >0, x S
1
, (5.2)
where
b(t, x) =
1
_
0
f
y
_
x, u(t )(x) +(1 )p(t )(x), u
x
(t )(x) +(1 )p
x
(t )(x)
_
d,
d(t, x) =
1
_
0
f
z
_
x, u(t )(x) +(1 )p(t )(x), u
x
(t )(x) +(1 )p
x
(t )(x)
_
d.
We also have with some >0
lim
t
e
t
_
_
v(t )
_
_
X
= 0. (5.3)
We consider the sequence v(n) = u(n) a, n N. Note that u(n) W
ss
loc
(a) for all n N.
Changing the norms to the equivalent ones, if necessary, but keeping the notation, we observe
that
W
ss
loc
(a) =
_
u =a +h
_
P
s
(u a)
_
+P
s
(u a): u B
X
(a, )
_
, (5.4)
where h : B
P
s
X
(0, ) B
P
u
X
P
c
X
(0, ) is a C
1
function such that h(0) = 0 and h
(0) = 0.
Let
= max
_
|
j
|: |
j
| <1
_
.
Then {
j
: |
j
| = } is a spectral set for T
by P. If i() = 2N 1, then
2N1
= 1 and
2N
form a spectral set and thus PX
is the
one-dimensional space spanned by the eigenfunction corresponding to
2N
, so PX
E
N
()
and
z() = 2N =i() +1 for PX
\ {0}. (5.5)
1
v(t ) = 0 for all t 0, since u
0
/ .
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 703
If i() = 2N, then
2N
= 1 and PX
is either E
N+1
() or the one-dimensional space spanned
by the eigenfunction corresponding to
2N+1
. In both cases we have PX
E
N+1
() and
z() = 2N +2 =i() +2 for PX
\ {0}. (5.6)
It can be shown that for each a the set
W
f s
loc
(a) =
_
u
0
X
:
_
_
S(t )u
0
S(t )a
_
_
X
= 0
_
for a certain = ( ) > is a C
1
submanifold of X
, tangent at a to (P
s
P)X
. We call
W
f s
loc
(a) the local fast stable manifold at a .
We are going to show that if u
0
W
ss
loc
(a) \ W
f s
loc
(a), then there exists a sequence t
k
such that the normalized vectors u(t
k
; u
0
) p(t
k
; a) tend to some PX
\ {0}. Consequently,
the zero number estimates for elements from PX
-coordinate.
Lemma 5.1. For u
0
W
ss
loc
(a) \ W
f s
loc
(a) we have
(P
s
P)v(n)
X
Pv(n)
X
0 as n . (5.7)
Proof. Note that
W
f s
loc
(a) =
_
u =a +g
_
(P
s
P)(u a)
_
+(P
s
P)(u a), u B
X
(a, )
_
,
where g : B
(P
s
P)X
(0, ) B
(P
u
+P+P
c
)X
(0, ) is C
1
and g(0) = 0, g
(0) = 0, is a subset
of W
ss
loc
(a). Taking into account (5.4) and setting y = P(u a) and z = (P
s
P)(u a) for
u W
ss
loc
(a), we see that
W
f s
loc
(a) =
_
u =a +h(y +z) +y +z W
ss
loc
(a): y =Pg(z), z B
(P
s
P)X
(0, )
_
.
This means that in the coordinates (y, z) for W
ss
loc
(a) the manifold W
f s
loc
(a) is a graph of the
function y =Pg(z).
Consider rst the behavior of the sequence {v(n)} for u
0
W
ss
loc
(a). Denote by
T (t, s) :
X
0 as n (5.8)
704 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
in the operator norm of L(X
, X
, satises
v
t
=v
xx
+
b(t, x)v +
d(t, x)v
x
, t
_
n, (n +1)
_
, x S
1
, v(n) =.
We change the variables v
n
(s) =v(s +n), s [0, ]. Then v
n
satises
_
v
n
s
= v
n
xx
+
b(s +n, x) v
n
+
d(s +n, x) v
n
x
, s (0, ], x S
1
,
v
n
(0) =.
(5.9)
Moreover, for w(s) =T (s, 0), s [0, ], from (4.1) we have
w
s
=w
xx
+b(s, x)w +d(s, x)w
x
, s (0, ], x S
1
, w(0) =.
Dene z
n
(s) = v
n
(s) w(s), s [0, ], and note that it satises
z
n
s
=z
n
xx
+b(s, x)z
n
+d(s, x)z
n
x
+
_
C
X
, 0 t , X
,
_
_
G(t, )
_
_
X
C
(t )
X
0 , 0 <t , X
0
,
and
z
n
(s) =
s
_
0
G(s, )h
n
() d, 0 s ,
where h
n
() =(
b( +n) b()) v
n
+(
d( +n) d()) v
n
x
. Thus we get
_
_
z
n
(s)
_
_
X
C
s
_
0
1
(s )
_
_
h
n
()
_
_
X
0
d. (5.10)
Moreover, from the variation of constants formula and (5.9) we obtain
_
_
v
n
(s)
_
_
X
c
X
+c
s
_
0
1
(s )
_
_
b() v
n
+d() v
n
x
+
_
b( +n) b()
_
v
n
+
_
d( +n) d()
_
v
n
x
_
_
X
0
d.
Note that by the regularity of f we have
b(, x)
M,
d(, x)
M, [0, ], x S
1
,
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 705
and
b( +n, x) b(, x)
M,
d( +n, x) d(, x)
M,
for any [0, ], x S
1
and n N. Therefore
_
_
v
n
(s)
_
_
X
c
X
+2Mc c
s
_
0
1
(s )
_
_
v
n
()
_
_
X
d, s [0, ].
From a Volterra type inequality we obtain for L =L(c, c, M, ) >0
_
_
v
n
(s)
_
_
X
L
X
, s [0, ]. (5.11)
Fix >0 and let n
0
N be such that for n n
0
we have for any [0, ], x S
1
b( +n, x) b(, x)
<
(1 )
C cL
1
,
d( +n, x) d(, x)
<
(1 )
C cL
1
.
Hence from (5.10) and (5.11) we get
_
_
z
n
()
_
_
X
(1 )
L
1
_
0
1
( )
_
_
v
n
()
_
_
X
d
X
.
This shows that U
n
0 in L(X
, X
).
Note that by denition v(n) =u(n) a with u(n) W
ss
loc
(a) and
v
_
(n +1)
_
=
T
_
(n +1), n
_
v(n).
Rewriting this equation in coordinates (y, z), we obtain with y
n
= Pv(n) and z
n
= (P
s
P)v(n)
_
y
n+1
=P
T
_
(n +1), n
__
h(y
n
+z
n
) +y
n
+z
n
_
,
z
n+1
=(P
s
P)
T
_
(n +1), n
__
h(y
n
+z
n
) +y
n
+z
n
_
.
Using the denition of U
n
in (5.8) and the properties of projections, our system can be written as
_
y
n+1
=PT
Py
n
+PU
n
_
h(y
n
+z
n
) +y
n
+z
n
_
,
z
n+1
=(P
s
P)T
(P
s
P)z
n
+(P
s
P)U
n
_
h(y
n
+z
n
) +y
n
+z
n
_
.
(5.12)
Finally, we make the change of coordinates y = y Pg(z), z = z, use the fact that in the new
coordinates W
f s
loc
(a) is described by the equation y = 0 and get
_
y
n+1
=PT
P y
n
+G
n
( y
n
, z
n
),
z
n+1
=(P
s
P)T
(P
s
P) z
n
+H
n
( y
n
, z
n
),
(5.13)
where
706 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
G
n
( y
n
, z
n
) =PU
n
_
h
_
y
n
+Pg( z
n
) + z
n
_
+ y
n
h
_
Pg( z
n
) + z
n
__
,
H
n
( y
n
, z
n
) =(P
s
P)U
n
_
h
_
y
n
+Pg( z
n
) + z
n
_
+ y
n
+Pg( z
n
) + z
n
_
.
Considering the spectra (PT
P) and ((P
s
P)T
(P
s
P)), we see that there exist
0 < r < and 0 < <
r
2
and norms equivalent to the original ones in the spaces PX
and
(P
s
P)X
such that
(P
s
P)T
(P
s
P) z
(P
s
P)X
(r +)||| z|||
(P
s
P)X
, z (P
s
P)X
, (5.14)
PT
P y
PX
( )||| y|||
PX
, y PX
. (5.15)
Indeed, choose 0 <r < so that
lim
n
_
(P
s
P)T
(P
s
P)
_
n
1
n
L((P
s
P)X
,(P
s
P)X
)
<r.
Then we set
||| z|||
(P
s
P)X
=
n=0
r
n
_
_
_
(P
s
P)T
(P
s
P)
_
n
z
_
_
(P
s
P)X
, z (P
s
P)X
.
The estimate (5.14) follows easily. Since
lim
n
_
_
(PT
P)
n
_
_
1
n
L(PX
,PX
)
=
1
,
we can dene
||| y|||
PX
=
n=0
( )
n
_
_
(PT
P)
n
y
_
_
PX
, y PX
.
Then the estimate (5.15) is straightforward.
Choose now (
r+
, 1). Then there exists
0
=
0
(r, , , ) > 0 such that for every
0 < <
0
we have
r ++
<.
Observe that (5.8) and the properties of h and g as well as the fact that
||| y
n
|||
PX
+ ||| z
n
|||
(P
s
P)X
0 as n
imply that for every 0 < <
0
there exists n
0
=n
0
() N such that for any n n
0
we have
G
n
( y
n
, z
n
)
PX
||| y
n
|||
PX
(5.16)
and
H
n
( y
n
, z
n
)
(P
s
P)X
_
||| y
n
|||
PX
+ ||| z
n
|||
(P
s
P)X
_
. (5.17)
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 707
Assume now that u
0
W
ss
loc
(a) \ W
f s
loc
(a). Then y
n
= 0 for all n N. Applying (5.14), (5.15),
(5.16) and (5.17) to (5.13), we estimate for n n
0
||| z
n+1
|||
(P
s
P)X
||| y
n+1
|||
PX
r ++
||| z
n
|||
(P
s
P)X
||| y
n
|||
PX
+
. (5.18)
Choose (0, 1 ). Therefore, from (5.18) we infer that
if
||| z
n
|||
(P
s
P)X
||| y
n
|||
PX
( )
, then
||| z
n+1
|||
(P
s
P)X
||| y
n+1
|||
PX
( +)
||| z
n
|||
(P
s
P)X
||| y
n
|||
PX
. (5.19)
Fix any >0. Choose 0 < <
0
such that
( )
<. Suppose that for every n n
0
=n
0
()
we have ||| z
n
|||
(P
s
P)X
||| y
n
|||
PX
. Then by (5.19) we get
( )
<
||| z
n
|||
(P
s
P)X
||| y
n
|||
PX
( +)
nn
0
||| z
n
0
|||
(P
s
P)X
||| y
n
0
|||
PX
, n n
0
,
which is a contradiction. Therefore, there exists n
1
n
0
such that ||| z
n
1
|||
(P
s
P)X
<||| y
n
1
|||
PX
.
Hence from (5.18) it follows that for every n n
1
we have
||| z
n
|||
(P
s
P)X
||| y
n
|||
PX
<.
Since >0 was chosen arbitrarily, this shows that
z
n
y
n
Pg(z
n
)
X
0 as n . (5.20)
Since g(0) = 0 and g
y
n
z
n
y
n
Pg(z
n
)
X
_
1 +
Pg(z
n
)
X
y
n
_
,
i.e. we have
z
n
y
n
_
1
Pg(z
n
)
X
y
n
Pg(z
n
)
X
z
n
y
n
Pg(z
n
)
X
.
This shows that
z
n
y
n
0 as n ,
or in other words (5.7), which completes the proof.
708 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
We compute
v(n)
v(n)
X
=
h(P
s
v(n))
v(n)
X
+
P
s
v(n)
v(n)
X
. (5.21)
Observe that
2
lim
n
h(P
s
v(n))
v(n)
X
= lim
n
h(P
s
v(n))
P
s
v(n)
X
P
s
v(n)
X
v(n)
X
= 0, (5.22)
since v(n) 0 as n and h(0) = 0, h
(0) = 0.
Let u
0
W
ss
loc
(a) \ W
f s
loc
(a) and note that for n large enough
P
s
v(n)
v(n)
X
=
(P
s
P)v(n)
Pv(n)
X
Pv(n)
X
v(n)
X
+
Pv(n)
v(n)
X
. (5.23)
Since PX
= (5.24)
and, by (5.21), (5.22), (5.7) and (5.23), we obtain
lim
k
v(t
n
k
)
v(t
n
k
)
X
= lim
k
S(t
n
k
)u
0
a
S(t
n
k
)u
0
a
X
=. (5.25)
Since Lemma 3.2 applies to (5.2), we have for u
0
W
ss
loc
(a) \ W
f s
loc
(a) and k large enough
z(u
0
a) =z
_
v(0)
_
z
_
v(t
n
k
)
_
=z(). (5.26)
Let now u
0
W
f s
loc
(a) \{a}. Since Lemma 3.2 applies to (5.2) and the zero number is bounded
from below, there exists n N large enough so that v(n) = u(n) a has only simple zeros.
Note that u(n) W
f s
loc
(a) and choose u W
ss
loc
(a) \ W
f s
loc
(a) such that
z
_
v(n)
_
=z
_
u(n) a
_
=z( u a).
Therefore, by the above considerations there exists PX
= 0
and, for 2N =z(p
t
(0; a)),
z(u
0
a)
_
i() +1 = 2N if i() = 2N 1,
i() +2 = 2N +2 if i() = 2N.
(5.27)
6. Global unstable manifold of a hyperbolic periodic orbit
Following [14, Theorem 6.1.9], we prove a general result concerning the extension of sub-
manifolds.
Lemma 6.1. Let S(t ) : X
) with
D
v
h(v) L(E, X
, t 0,
be injective, belonging to C
1
(X
, X
) and let D
w
S(t )(w) L(X
, X
with dimension m.
Proof. Dene f
t
: V X
by
f
t
(v) =S(t )h(v), v V.
Since S(t )|
A
is a homeomorphism of A onto A, we infer that S(t )|
is a homeomorphism of
onto S(t ) (both equipped with the induced topology from X
). Thus f
t
is a homeomorphism
of V onto S(t ), f
t
C
1
(V, X
_
is injective.
Moreover, (D
v
f
t
(v))E is an m-dimensional closed linear subspace of X
, so it has a closed
complement in X
,
it follows from [30, Proposition B.4.3] that S(t ) is a C
1
submanifold of X
with dimen-
sion m.
In our problem we dene the global unstable manifold of a hyperbolic periodic orbit by
W
u
() =
_
t 0
S(t )W
u
loc
().
Using Lemma 6.1 we infer that this invariant subset of the global attractor A is the union of
C
1
submanifolds of X
. Moreover, we have
W
u
() =
_
t 0
_
a
S(t )W
su
loc
(a),
where again by Lemma 6.1 each S(t )W
su
loc
(a) is a C
1
submanifold of X
.
710 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
Below we examine the unstable manifold and, for simplicity, we keep the same notation as
in the stable manifold case. The aim here is to show that for any u
0
W
u
() \ there exists
a such that u(t ; u
0
) p(t ; a) tends exponentially to 0 as t and
z(u
0
a) i() 1 +
1 +(1)
i()
2
.
The crucial observation is, similarly to the case of W
s
loc
(), the existence of a sequence
t
k
such that the normalized vectors u(t
k
; u
0
) p(t
k
; a) tend to some P
u
X
\ {0}.
However, the proof is now easier, since W
u
loc
() is a nite-dimensional submanifold of X
.
Choose u
0
W
u
() \ and note that then there exist 0 and a such that
u
0
S()W
su
loc
( a). Let u
0
W
su
loc
( a) and a satisfy u
0
=S() u
0
and a =S() a. We consider
the corresponding backward solutions u(t ) =u(t ; u
0
) and p(t ) = p(t ; a) =p(t ; a) of (2.3).
Let v(t ) =u(t ) p(t ), t 0, and note that v satises the nonautonomous linear equation
3
v
t
=v
xx
+
b(t, x)v +
d(t, x)v
x
, t <0, x S
1
, (6.1)
where
b(t, x) =
1
_
0
f
y
_
x, u(t )(x) +(1 ) p(t )(x), u
x
(t )(x) +(1 ) p
x
(t )(x)
_
d,
d(t, x) =
1
_
0
f
z
_
x, u(t )(x) +(1 ) p(t )(x), u
x
(t )(x) +(1 ) p
x
(t )(x)
_
d.
We also have
lim
t
e
t
_
_
v(t )
_
_
X
= 0. (6.2)
We consider the sequence v(n) = u(n) a, n N. Note that u(n) W
su
loc
( a) for all
n N. Changing the norms to the equivalent ones, if necessary, but keeping the notation, we see
that
W
su
loc
( a) =
_
u = a +h
_
P
u
(u a)
_
+P
u
(u a): u B
X
( a, )
_
, (6.3)
where h : B
P
u
X
(0, ) B
P
s
X
P
c
X
(0, ) is a C
1
function such that h(0) = 0 and h
(0) = 0.
We compute
v(n)
v(n)
X
=
h(P
u
v(n))
v(n)
X
+
P
u
v(n)
v(n)
X
. (6.4)
3
v(t ) = 0 for all t 0, since u
0
/ .
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 711
Again observe that
4
lim
n
h(P
u
v(n))
v(n)
X
= lim
n
h(P
u
v(n))
P
u
v(n)
X
P
u
v(n)
X
v(n)
X
= 0. (6.5)
Since
P
u
v(n)
v(n)
X
, n N,
is a bounded sequence in a nite-dimensional subspace of X
= (6.6)
and by (6.4) and (6.5), we obtain
lim
k
v(t
n
k
)
v(t
n
k
)
X
=. (6.7)
If i() = 2N 1, then
2N1
= 1, 2N =z( p
t
(0; a)) =z(p
t
(0; a)) and
z() 2N 2 =i() 1 for P
u
X
\ {0}, (6.8)
whereas if i() = 2N, then
2N
= 1, 2N =z( p
t
(0; a)) =z(p
t
(0; a)) and
z() 2N =i() for P
u
X
\ {0}. (6.9)
Using Lemma 3.2, we have for u
0
W
u
() \ and k large enough
z(u
0
a) z( u
0
a) =z
_
v(0)
_
z
_
v(t
n
k
)
_
=z(). (6.10)
Summarizing the above considerations we obtain
Theorem 6.2. For any u
0
W
u
() \ there exist a and
t
_
_
u(t ; u
0
) p(t ; a)
_
_
X
= 0
and, for 2N =z(p
t
(0; a)),
z(u
0
a)
_
i() 1 = 2N 2 if i() = 2N 1,
i() = 2N if i() = 2N.
(6.11)
4
P
u
v(n) = 0 for any n N, because otherwise v(n) =h(P
u
v(n)) +P
u
v(n) would be 0.
712 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
7. Exclusion of a homoclinic connection for a hyperbolic periodic orbit
In this section we will consider two (not necessarily distinct) hyperbolic periodic orbits
and
+
with periods
>0 and
+
>0, respectively. We also assume that there exists a point
u
0
_
W
u
(
) W
s
loc
(
+
)
_
\ (
+
).
Note that if
=
+
, then u
0
is a homoclinic point and the corresponding orbit is a homoclinic
connection for the periodic orbit.
Consequently, to u
0
/
+
there corresponds an X
solution u(; u
0
) of (2.3), which is
dened for all t R, its orbit is bounded in X
(; a
(t ; a
) = 0 (7.1)
in X
N so that 2N
=z(p
t
(0; a
)).
In order to combine the estimates (5.27) and (6.11) we observe in the following two lemmas
that there exists some neighborhood of {p
+
(t ; a
+
) p
(t ; a
): t R} in X
consisting of a
nite number of balls such that in a sufciently bigger neighborhood the zero number of functions
is constant.
Lemma 7.1. If
=
+
and a
=a
+
, then
z
_
p
+
(t ; a
+
) p
(t ; a
)
_
= const, t R.
Therefore there exists a nite cover
n
i=1
B
X
(p
+
(t
i
; a
+
) p
(t
i
; a
),
i
) of the set
{p
+
(t ; a
+
) p
(t ; a
): t R} in X
n
i=1
B
X
(p
+
(t
i
; a
+
) p
(t
i
; a
), 2
i
).
Proof. Suppose that v(t ) =p
+
(t ; a
+
) p
(t ; a
(t ; a
): t R} in X
(as
a continuous image of a compact interval), since to each point v of this set there corresponds a
ball B
X
(v, 2
v
) in which the zero number is constant and we can choose a nite subcover from
v
B
X
(v,
v
).
Lemma 7.2. If
=
+
, then we have
z
_
p
+
(t ; a
+
) p
(s; a
)
_
= const, s, t R.
Therefore there exists a nite cover
n
i=1
B
X
(p
+
(t
i
; a
+
) p
(s
i
; a
),
i
) of the set
{p
+
(t ; a
+
) p
(s; a
): s, t R} in X
n
i=1
B
X
(p
+
(t
i
; a
+
) p
(s
i
; a
), 2
i
).
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 713
Proof. Fix R and suppose that v(t ) =p
+
(t ; a
+
) p
(t +; a
is rational, say
+
=
k
m
for some k, m N, then with
=m
+
=k
(t +k
+; a
) =p
+
(t ; a
+
) p
(t +; a
) =v(t ),
i.e. v is periodic with period . Then we have
z
_
v(t
0
)
_
=z
_
v(t
0
)
_
>z
_
v(t
0
)
_
,
which is a contradiction.
Consider now
+
+
<t
0
such
that v(t
1
) does not have multiple zeros and observe that
z
_
v(t
1
)
_
>z
_
v(t
0
)
_
.
To obtain a contradiction it sufces to nd t
2
>t
0
such that z(v(t
2
)) =z(v(t
1
)). Below we show
that
>0
t
2
>t
0
_
_
_
p
+
(t
2
; a
+
), p
(t
2
+; a
)
_
_
p
+
(t
1
; a
+
), p
(t
1
+; a
)
__
_
X
<,
which implies that v(t
1
) v(t
2
)
C
1
(S
1
)
< and thus z(v(t
1
)) = z(v(t
2
)) for sufciently small
> 0. Fix > 0. Since p
+
and p
t , s) R
2
we know that |(t, s) (
t , s)| <
implies
_
_
_
p
+
(t ; a
+
), p
(s +; a
)
_
_
p
+
(
t ; a
+
), p
( s +; a
)
__
_
X
<.
Choose n
0
N such that
n
0
< and set s
0
=
t
0
t
1
+
1. Denoting by [ ] the oor function,
consider the n
0
+1 real positive numbers
_
[s
0
] +1
_
, 2
_
[s
0
] +1
_
, . . . , (n
0
+1)
_
[s
0
] +1
_
> m
2
+
=l +r,
where l N{0} and 0 <|r| <
1
n
0
.
Hence the distance between the points (t
1
+ (m
1
m
2
)
+
, t
1
+ (m
1
m
2
)
+
) and (t
1
+
(m
1
m
2
)
+
, t
1
+l
) is less than
n
0
<. In conclusion, for t
2
=t
1
+(m
1
m
2
)
+
>t
0
we
have
714 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
_
_
_
p
+
(t
2
; a
+
), p
(t
2
+; a
)
_
_
p
+
(t
1
; a
+
), p
(t
1
+; a
)
__
_
X
=
_
_
_
p
+
(t
2
; a
+
), p
(t
2
+; a
)
_
_
p
+
_
t
1
+(m
1
m
2
)
+
; a
+
_
, p
(t
1
+l
+; a
)
__
_
X
<.
This ends the proof of the rst assertion.
The second claim follows from the compactness of {p
+
(t ; a
+
) p
(s; a
): s, t R} in X
(as a continuous image of a two-dimensional torus), since to each point v of this set there corre-
sponds a ball B
X
(v, 2
v
) in which the zero number is constant and we choose a nite subcover
from
v
B
X
(v,
v
).
The above lemmas allow us to combine the inequalities (5.27) and (6.11). As a particular
result, we deduce that there is no homoclinic connection for a hyperbolic periodic orbit.
Theorem 7.3. If u
0
(W
u
(
) W
s
loc
(
+
)) \ (
+
) and 2N
=z(p
t
(0; a
)), then
N
N
+
and i(
) i(
+
) +1, (7.2)
which excludes a homoclinic connection for a hyperbolic periodic orbit.
Moreover, if i(
+
) = 2N
+
, then we even have
N
N
+
+1. (7.3)
Proof. Let
0
> 0 be the minimum of
1
, . . . ,
n
from Lemmas 7.1, 7.2. Since (7.1) implies for
large t >0
_
_
u(t ; u
0
) p
+
(t ; a
+
)
_
_
X
<
0
and
_
_
u(t ; u
0
) p
(t ; a
)
_
_
X
<
0
,
we have
z(u
0
a
) z
_
u(t ; u
0
) p
+
(t ; a
+
) +p
+
(t ; a
+
) p
(t ; a
)
_
=z
_
p
+
(t ; a
+
) p
(t ; a
)
_
=z
_
p
(t ; a
) p
+
(t ; a
+
)
_
=z
_
u(t ; u
0
) p
(t ; a
) +p
(t ; a
) p
+
(t ; a
+
)
_
z(u
0
a
+
).
Here we have assumed that a
= a
+
, but if a
= a
+
we have z(u
0
a
) =z(u
0
a
+
) imme-
diately. By Theorem 5.2 we have
z(u
0
a
+
)
_
i(
+
) +1 = 2N
+
if i(
+
) = 2N
+
1,
i(
+
) +2 = 2N
+
+2 if i(
+
) = 2N
+
,
and by Theorem 6.2 we have
z(u
0
a
)
_
i(
) 1 = 2N
2 if i(
) = 2N
1,
i(
) = 2N
if i(
) = 2N
.
This leads straightforward to (7.2) and (7.3).
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 715
8. Transversal intersection of stable and unstable manifolds of hyperbolic periodic orbits
We are going to show the transversal intersection of stable and unstable manifolds of hyper-
bolic periodic orbits following the approach used for equilibria by M. Chen, X.-Y. Chen and
J.K. Hale in [6] (see also [25]). Assume that
u
0
S()W
u
loc
(
) W
s
loc
(
+
)
with some 0. Our purpose is to show that
T
u
0
S()W
u
loc
(
) +T
u
0
W
s
loc
(
+
) =X
.
Without loss of generality we assume that u
0
/
+
. We consider the linearization of (1.1)
around u(; u
0
)
_
v
t
=v
xx
+
b(t, x)v +
d(t, x)v
x
, t >s, x S
1
,
v(s) =
(8.1)
and the corresponding evolution operators
T (t, s) L(X
, X
), t s. Therefore, we have
(; a
such that 2N
=z(p
t
(0; a
)) and
u(t ; u
0
) p
(t ; a
) 0 as t .
First we concentrate on the description of the tangent space to the local stable manifold. To this
end, let us consider the linearization of (1.1) around p
+
(; a
+
)
_
w
+
t
=w
+
xx
+b
+
(t, x)w
+
+d
+
(t, x)w
+
x
, t >s, x S
1
,
w
+
(s) =
and the corresponding evolution operators T
+
(t, s) L(X
, X
), t s. We denote by T
+
+
=
T
+
(
+
, 0) the period map and as in Section 4 we put its multipliers in a sequence {
j
}
j0
such
that they appear according to their algebraic multiplicity and are ordered by |
j
| |
j+1
|.
Since the coefcients of the equations converge to each other, following the part of the proof
of Lemma 5.1 we get U
n
=
T ((n + 1)
+
, n
+
) T
+
+
0 in L(X
, X
+
v(n
+
) +U
n
v(n
+
)
and the assumptions (B.1)(B.2) of [6] are satised.
For any X
(m, ) = limsup
n
_
_
v(n
+
; m
+
, )
_
_
1
n
X
= limsup
n
_
_
T (n
+
, m
+
)
_
_
1
n
X
.
Let be the set of all nonnegative numbers r such that
_
T
+
+
_
_
z C: |z| =r
_
=.
716 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
We also set r
j
= |
j
|, j 0. For any integer j 0 and m N we dene the spaces
F
+
j
(m) =
_
X
(m, ) r
j
_
.
From [6, Corollary B.3] it follows that for every X
=r.
Therefore, we have
X
=F
+
0
(m) F
+
1
(m) F
+
2
(m) .
Since it is well known that linear equations like (8.1) do not possess nontrivial solutions
decaying faster than any exponential (see [16]), as in [6, Theorem 3.1] we see that for all = 0
and m N we have
_
j=0
F
+
j
(m) = {0},
because
(m, 0) = 0, m N.
From the asymptotic behavior of the solution v(n
+
; m
+
, ) we are able to characterize
the zero number z() of the initial condition . Assume that
lim
n
_
_
v(n
+
; m
+
, )
_
_
1
n
X
=r
j
.
We have the following two cases. If j is odd and r
j
> r
j+1
, then F
+
j
(m) \ F
+
j+1
(m) and
by [6, Theorems B.2, B.4] there exists Ej+1
2
(
+
),
X
= 1 and a subsequence {t
n
k
}
{n
+
: n m} such that
v(t
n
k
; m
+
, )
v(t
n
k
; m
+
, )
X
.
Therefore, we have for large k N
z() z
_
v(t
n
k
; m
+
, )
_
=z
_
v(t
n
k
; m
+
, )
v(t
n
k
; m
+
, )
X
_
=z() =j +1.
If j is even, then F
+
j
(m) \ F
+
j+1
(m) and by [6, Theorems B.2, B.4] there exists
Ej
2
(
+
),
X
= 1 and a subsequence {t
n
k
} {n
+
: n m} such that
v(t
n
k
; m
+
, )
v(t
n
k
; m
+
, )
X
.
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 717
Therefore, we have for large k N
z() z
_
v(t
n
k
; m
+
, )
_
=z
_
v(t
n
k
; m
+
, )
v(t
n
k
; m
+
, )
X
_
=z() =j.
Fix m N, F
+
k
(m), = 0 and let j k be such that F
+
j
(m) \ F
+
j+1
(m). Suppose
that k is even. If j is even, then z() j k, whereas if j is odd, then z() j + 1 k.
Suppose now that k is odd. If j is even, then z() j k +1 due to the difference of parity of
k and j. Moreover, if j is odd, then z() j +1 k +1.
Since the local stable manifold W
s
loc
(
+
) of the hyperbolic periodic orbit
+
coincides
locally with the local center-stable manifold W
cs
loc
(a
+
) of a xed point a
+
+
for the map
S(
+
), it follows from [6, Theorem C.4] that
T
u
0
W
s
loc
(
+
) =
_
X
: limsup
n
_
_
v(n
+
; 0, )
_
_
1
n
X
1
_
.
If i(
+
) = 2N
+
, then r
2N
+ =
2N
+ = 1 and by [6, Theorem B.7] for sufciently large
m
0
Nwe see that T
u(m
0
+
;u
0
)
W
s
loc
(
+
) =F
+
2N
+
(m
0
) is isomorphic to cl
X
(span{p
+
t
(0; a
+
)}
E
N
+
+1
(
+
) ). Therefore, z() 2N
+
for T
u(m
0
+
;u
0
)
W
s
loc
(
+
) \ {0} and
codimT
u(m
0
+
;u
0
)
W
s
loc
(
+
) = 2N
+
.
If i(
+
) = 2N
+
1, then r
2N
+
1
=
2N
+
1
= 1 and by [6, Theorem B.7] for suf-
ciently large m
0
N we see that T
u(m
0
+
;u
0
)
W
s
loc
(
+
) = F
+
2N
+
1
(m
0
) is isomorphic to
cl
X
(E
N
+(
+
) E
N
+
+1
(
+
) ). Therefore, z() 2N
+
for T
u(m
0
+
;u
0
)
W
s
loc
(
+
) \
{0} and codimT
u(m
0
+
;u
0
)
W
s
loc
(
+
) = 2N
+
1.
Since the adjoint operator of the evolution operator
T (m
0
+
, 0) is injective (thus
T (m
0
+
, 0) has dense range by [14, Theorem 7.3.3]) and T
u
0
W
s
loc
(
+
) is the preimage of
T
u(m
0
+
;u
0
)
W
s
loc
(
+
) under
T (m
0
+
, 0), we see that if i(
+
) = 2N
+
, then
codimT
u
0
W
s
loc
(
+
) = codimF
+
2N
+
(0) = 2N
+
and z() 2N
+
for T
u
0
W
s
loc
(
+
) \ {0}, whereas if i(
+
) = 2N
+
1, then
codimT
u
0
W
s
loc
(
+
) = codimF
+
2N
+
1
(0) = 2N
+
1
and z() 2N
+
for T
u
0
W
s
loc
(
+
) \ {0}.
In both cases of i(
+
) we now consider r
2N
+
+1
< 1 and the subspace F
+
2N
+
+1
(0) of
T
u
0
W
s
loc
(
+
), which is isomorphic to cl
X
(E
N
+
+1
(
+
) ). Then we have z() 2N
+
+2
for F
+
2N
+
+1
(0) \ {0} and codimF
+
2N
+
+1
(0) = 2N
+
+1.
From the above considerations the following result follows.
Lemma 8.1. We have
z(v) 2N
+
, v T
u
0
W
s
loc
(
+
) \ {0}.
Moreover, there exists a subspace W
+
of T
u
0
W
s
loc
(
+
) such that
z(v) 2N
+
+2, v W
+
\ {0},
and codimW
+
= 2N
+
+1.
718 R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721
Using Lemma 8.1 we nally prove the main result of the paper.
Theorem 8.2. The stable and unstable manifolds of two hyperbolic periodic orbits
for the
problem (1.1) have a transversal intersection
W
u
(
) W
s
loc
(
+
),
i.e. if u
0
S()W
u
loc
(
) W
s
loc
(
+
) with some 0, then
T
u
0
S()W
u
loc
(
) +T
u
0
W
s
loc
(
+
) =X
.
Proof. The proof is in the same vein as the proof of Lemma 8.1 and is based on the description
of the tangent space to the global unstable manifold.
First assume i(
+
) = 2N
+
1. Thus we have codimW
s
loc
(
+
) = 2N
+
1. From the
inequality N
N
+
(see Theorem 7.3) we see, in a similar way as in the local sta-
ble manifold case, that for sufciently large m
0
N there exists a subspace F
m
0
,N
+
of
T
u(m
0
;u
0
)
S()W
u
loc
(
) such that
dimF
m
0
,N
+
= dim
_
E
0
(
) E
N
+
1
(
)
_
= 2N
+
1,
z(v) 2N
+
2, v F
m
0
,N
+
\ {0}.
Now note that F
0,N
+
=
T (0, m
0
)F
m
0
,N
+
is a subspace of T
u
0
S()W
u
loc
(
) with
dimF
0,N
+
= dimF
m
0
,N
+
and z(v) 2N
+
2, v F
0,N
+
\ {0},
since the evolution operator
T (,) for the linearized equation around the solution u(; u
0
) is
injective and does not increase the zero number. Thus we have
dimF
0,N
+
= 2N
+
1 = codimW
s
loc
(
+
)
and F
0,N
+
T
u
0
W
s
loc
(
+
) = {0}, since z(v) 2N
+
2 for v F
0,N
+
\ {0} and by Lemma 8.1
we have z(v) 2N
+
for v T
u
0
W
s
loc
(
+
) \ {0}. This proves that
T
u
0
S()W
u
loc
(
) +T
u
0
W
s
loc
(
+
) =X
in this case.
Let now i(
+
) = 2N
+
. Then by Theorem 7.3 we have N
N
+
+ 1 and we see again that
for sufciently large m
0
N there exists a subspace F
m
0
,N
+
of T
u(m
0
;u
0
)
S()W
u
loc
(
) such
that
dimF
m
0
,N
+
= dim
_
E
0
(
) E
N
+(
)
_
= 2N
+
+1,
z(v) 2N
+
, v F
m
0
,N
+
\ {0}.
R. Czaja, C. Rocha / J. Differential Equations 245 (2008) 692721 719
As before, note that F
0,N
+
=T (0, m
0
)F
m
0
,N
+
is a subspace of T
u
0
S()W
u
loc
(
) with
dimF
0,N
+
= dimF
m
0
,N
+
and z(v) 2N
+
, v F
0,N
+
\ {0}.
Thus we get
dimF
0,N
+
= 2N
+
+1 = codimW
+
,
where W
+
is the subspace of T
u
0
W
s
loc
(
+
) given in Lemma 8.1. Moreover, we have
F
0,N
+
W
+
= {0},
since z(v) 2N
+
for v F
0,N
+
\ {0} and z(v) 2N
+
+ 2 for v W
+
\ {0}, by Lemma 8.1.
Therefore, we obtain F
0,N
+
W
+
=X
) +T
u
0
W
s
loc
(
+
) =X