Dissipation and Compact Attractors

Journal of Dynamics and Differential Equations, Vol. 18, No.
3, July 2006 ( 2006)

DOI: 10.1007/s10884-006-9021-6
Dissipation and Compact Attractors

Jack K. Hale
Received December 27, 2005
We present an approach to the study of the qualitative theory of innite

dimensional dynamical systems. In nite dimensions, most of the success
has been with the discussion of dynamics on sets which are invariant and
compact. In the innite dimensional case, the appropriate setting is to con-
sider the dynamics on the maximal compact invariant set. In dissipative sys-
tems, this corresponds to the compact global attractor. Most of the time is
devoted to necessary and sufcient conditons for the existence of the compact
global attractor. Several important applications are given as well as important
results on the qualitative properties of the ow on the attractor.
1. INTRODUCTION
The purpose of these notes is to discuss an approach to the study of a
qualitative theory of dynamical systems in innite dimensions and to indi-
cate by examples that the theory is applicable in many important situ-
ations. The ultimate objective is to understand well the family of orbits
dened by the dynamical system and then determine the sensitivity of the
dynamics to perturbations of the dynamical system. The dynamical system
is stable if it does not change under perturbations and it is a bifurcation
point in the family of dynamical systems if it does change.
In nite dimensions, a precise theory began to evolve with Andronov
and Pontrjagin (1937) on the characterization of the stable planar vector
elds transversal to the boundary of a disk D. Two vector elds f and
g are equivalent if there is a homeomorphism h : D D which maps the
orbits of the ow for f onto those of g and preserves the sense of direc-
tion in time. Peixoto (1959) gave a characterization for such stable ows
School of Mathematics, Georgia Institute of Technology, Atlanta, GA 30332-0160, USA.

E-mail: [email protected]
485
1040-7294/06/0700-0485/0 2006 Springer Science+Business Media, Inc.
486 Hale
on smooth 2-manifolds. The work of Smale (1967) was a great stimulus

for the investigation of generic and stable properties of dynamical systems
on compact manifolds in nite dimension greater than 2 (see Palis and de
Melo (1982)). Simultaneously, further important developments were occur-
ring, especially in the Soviet Union (see, e.g., Silnikov (1968, 1970)).
If the space is locally compact but unbounded (e.g., R2 ), the theory
is still incomplete. Special topologies are needed to take care of the ow
at innity. In some situations, it is possible to avoid these difculties by
making an assumption that innity is unstable; that is, there is some type
of dissipative mechanism which forces each solution to leave a neighbor-
hood of innity. The idea for doing this goes back to Levinson (1944) in
his study of the effect of an -periodic forcing function on the van der Pol
equation. The dynamics of such a system is determined by the Poincare
map T taking an initial value at time zero to the solution at time . Lev-
inson was particularly interested in determining if there must be a periodic
solution of the same period as the forcing. Such solutions correspond to
xed points of the map T . To avoid the difculties of the behavior of the
map T at innity, Levinson said that the system is point dissipative if there
is a bounded set B R2 with the property that, for any x R2 , there is
an integer n0 such that T n x B for nn0 (innity is unstable). He proved
under this condition that there was a xed point of T n for some n, but he
did not prove that n could be taken to be 1. Pliss (1966) show that there
is a xed point of T . His proof also implied the existence of a compact
global attractor; that is, a compact set A R2 which is stable in the sense
of Lyapunov and attracts any bounded set under iterates of the map T .
The work of Levinson and Pliss were an inspiration to attempt to use
concepts of dissipativeness and additional properties on the dynamical sys-
tem to imply the existence of compact global attractors when the space
is innite dimensional and not locally compact. In this way, the essential
dynamics is reduced to a compact set. However, in general, it will not be
a manifold.
These notes give an introduction to this basic theory and illustrate
the theory with a few simple examples. The notes should serve only as
a beginning to the extensive subject of dynamics in innite dimensions.
Much more information is contained in Hale (1988), Hale and Verduyn-
Lunel (1993), Raugel (2002), Hale, et al. (2002) Hale and Raugel (2006),
Henry (1981), (2005). Other detailed works on this subject, especially for
partial differential equations, are Babin and Vishik (1992), Temam (1997),
Sell and You (2002). All of these works greatly inuenced the presentation
below, especially the unpublished notes of Hale and Raugel (2006).
Dissipation and Compact Attractors 487
2. MAXIMAL COMPACT INVARIANT SETS AND COMPACT

ATTRACTORS
Let X be a complete metric space with metric X . If T: X X is a
continuous map, the discrete dynamical system on X dened by T is the set
of iterates {T n , n = 0, 1, 2, . . . }. A continuous dynamical system on X con-
sists of a map T : R+ X X, such that the family of mappings T (t): X
X dened by T (t)x = T (t, x), t 0, satises the following properties:
(i) T (t)x is continuous for t 0, x X;
(ii) T (0) = I , the identity;
(iii) T (t + s) = T (t)T (s) for t, s 0;
Since the theory below applies both to discrete and continuous dynam-
ical systems, it is convenient to let G denote either R or the set Z =
{0, 1, 2, . . . }, G+ = {t G : t 0}, G = {t G : t0}.
For any point x X, let + (x) = {T (t)x, t G+ } be the positive orbit
through x. A negative orbit (x) through x is a sequence {y(t), t G },
y(0) = x, such that, for any st G , y(t) = T (t s)y(s). Let (x) be
the set of all negative orbits through x. If (x) is nonempty, then it may
contain more than one orbit since we do not assume that the map T (t) is
one-to-one. A global orbit (x) through x is a set (x) + (x) and we let
(x) be the set of all global orbits through x. In the same way, we dene
(B), (B), (B), (B), and (B) for any subset B of X.
The -limit set (B) and the -limit set (B) of a set B in X are
dened by

(B) = tG+ Cl + (T (t)B),

(B) = tG Cl (T (t)B).
A set J in X is said to be invariant if T (t)J = J , t G+ . If J is invariant,

then there is at least one global orbit through each point of J .
A set J in X attracts a subset K of X if X (T (t)K, J ) 0 as t
G+ .
If T is a dynamical system on X, J is a compact set for which
Cl + (J ) is compact, then (J ) is a compact invariant set which attracts
J . In a later section, we will identify an important class of dynamical sys-
tems (called asymptotically smooth) which occur frequently in the applica-
tions and have the property that the positive orbit + (B) being bounded
implies that (B) is a compact invariant set which attracts B.
Of primary interest in the theory of dynamical systems is the set
A = A(T ) = { bounded global orbits of T }.

488 Hale
If (B) is bounded and invariant, then all bounded global orbits through
elements of B must belong to A.
The set A(T ) is the maximal compact invariant set if it is compact,
invariant and maximal with respect to this property. The set A(T ) is the
compact global attractor if
(i) A(T ) is compact;
(ii) A(T ) is invariant;
(iii) A attracts bounded sets of X.
3. DEFINITIONS OF STABILITY AND BIFURCATION

For innite dimensional systems for which bounded sets do not have
compact closure, the concept of equivalence of two dynamical systems as
dened for the nite dimensional case does not lead to interesting results.
The set A(T ) dened above contains all of the dynamics of the dynam-
ical system T if we ignore the transient behavior. Of course, the tran-
sient behavior is very important, but it always requires special treatment
depending upon particular cases.
Denition 3.1.
(i) The dynamical system T is topologically equivalent to the dynam-
ical system S, T S, if there is a homeomorphism h: A(T )
A(S) preserving the orbit structure and sense of direction in
time.
(ii) Suppose that X, are complete metric spaces, T : R+
X X is continuous, and, for any , T : (t, x) R+ X
T (, t, x) X is a dynamical system on X. The dynamical sys-
tem T is A-stable if there is a neighborhood U of such
that T T for each U . T is a bifurcation point if T is not
A-stable.
Important dynamical systems which are stable belong to the class of
MorseSmale systems, which we now dene. If T is a dynamical system on
X, the nonwandering set T of T is the set of all x X such that, given a
neighborhood V of x and any t G+ there is a t G+ , t > t0 , such that
T (t)V V = . An m-periodic point of a discrete dynamical system is a
set {x, T x, . . . , T m1 x}, x X, with T m x = x and T p x = x for 1pm 1
for m > 1. A xed point of T is a 1-periodic point. A point x X is an
equilibrium of a continuous dynamical system T if T (t)x = x for all t R.
A periodic orbit of a continuous dynamical system T is an invariant
closed curve not containing an equilibrium. Let M(T ) be the set of peri-
odic points of a discrete dynamical system and the set of equilibria and
periodic orbits of a continuous T . The dynamical system is MorseSmale

if
(i) (T ) = M(T ) is nite;
(ii) each element of M(T ) is hyperbolic;
(iii) the stable and unstable manifolds of elements of M(T ) are
transversal.
Suppose that we are working in a parametrized family of discrete dynam-
ical systems T as dened above and suppose that
(i) for each , there exists the compact global attractor A(T );
(ii) A(T ) is upper semicontinuous in ;
(iii) for each , T and the derivative Dx T x are one-to-one.
If such a family is also MorseSmale, then Oliva (1982) has proved that
the system is A-stable.
Oliva (see Hale, et al. (2002)) also has shown that the same result is
true for a general class of continuous dynamical systems which includes
quasilinear parabolic equations on a bounded domain and retarded func-
tional differential equations with a nite delay. The hypotheses on this
class is too restrictive to apply to neutral functional differential equations
and hyperbolic equations.
We turn now to presenting conditions on the dynamical system which
will imply that there exists the compact global attractor. It is instruc-
tive to begin with the case where X is unbounded and locally compact;
namely, Rn .
4. DISSIPATIVE DYNAMICAL SYSTEMS IN Rn

We begin with the denition of point dissipative introduced by
Levinson (1944) in nite dimensional space which we note is meaningful
on a metric space X.
Denition 4.1. If X is a metric space, a dynamical system T on X is
point dissipative if there is a bounded set B X such that, for any x X,
there is a t0 (x) G+ such that T (t)x B if t t0 (x), t G+ .
Consider the ordinary differential equation
y = f (t, y), (4.1)
where f : R Rn Rn satises enough conditions to imply that, for any

> 0 and any y0 Rn , there is a unique solution y(t, , y0 ) with initial
value y0 at and the solution exists for all t .
490 Hale
Suppose that there is an > 0 such that f (t + , y) = f (t, y) for all

t, y and that is the minimal period. Dene the Poincare map T : Rn Rn
by T y0 = y(, 0, y0 ). A xed point of T corresponds to an -periodic solu-
tion of (4.1) and is called a harmonic solution.
If T is point dissipative, Levinson (1944) proved that there is an inte-
ger k such that T k has a xed point; that is, there is a k-periodic solution
of (4.1) which is called a subharmonic solution. However, he was not able
to prove that k = 1.
In Pliss (1966), proved the following result.
Theorem 4.2. If the continuous mapping T : Rn Rn is point dissipa-
tive, then the compact global attractor A exists.
Using this fact and an asymptotic xed point theorem of Browder (1959),
Pliss (1966) proved that (4.1) has an harmonic solution from the following
general result.
Theorem 4.3. If a continuous map T : Rn Rn is point dissipative, then
T has a xed point.
Remark 4.4. Massera (1950) has given an example of a differential
equation in R2 with -periodic coefcients for which all solutions are
bounded and there is no -periodic solution; that is, no xed point of T .
Therefore, the dissipative hypothesis is necessary if the conclusion is to be
true for all continuous Poincare maps T if n 2. Of course, it is easy to
show that there is a xed point of T if n = 1.
If the function f in (4.1) does not depend upon t and we let T (t)y0 =
y(t, 0, y0 ), then T : R Rn is a dynamical system on Rn . The analog of
Theorems 4.2 and 4.3 holds for continuous dynamical systems on Rn .
Theorem 4.5. If T (t), t 0, is a continuous dynamical system on Rn
which is point dissipative, then the compact global attractor exists.
Furthermore, there must be an equilibrium; that is, there is a point x0
Rn such that T (t)x0 = x0 for all t 0.
The existence of an equilibrium is proved by noting that, for any >
0, the Poincare map T ( ) has a xed point y and then showing that there
is a sequence j 0 as j such that yj z as j and z is an
equilibrium of the dynamical system.
Remark 4.6. As a consequence of Theorem 4.5, it is always possible
to nd a neighborhood U of the attractor with smooth boundary such
that the ow is transversal to the boundary of U . As mentioned before,
this is the situation considered by Andronov and Pontrjagin (1937) in R2 .
Remark 4.7. In R2 , as a consequence of the PoincareBendixon the-

ory, the conclusion of Theorem 4.5 is true if one assumes only that all
solutions are bounded. The dissipative condition is not needed to obtain
the existence of an equilibrium. In R3 , Jones and Yorke (1960) constructed
an autonomous differential equation in R3 with the property that all solu-
tions are bounded and there does not exist an equilibrium. This shows
that some type of dissipation is necessary for the existence of an equilib-
rium.
5. COMPACT DYNAMICAL SYSTEMS

The general theory of topological dynamics deals with dynamical
systems on a complete metric space X. The results in most of this theory
are based on only the continuity of the map T (t), t G+ . It is natural to
discuss additional properties that may be obtained if we assume some type
of dissipative property for the dynamical system.
The strong conclusions in Theorems 4.2 and 4.5 assuming only point
dissipativeness is a consequence of the fact that each closed bounded set
in Rn is a compact set. Therefore, the theorem can be restated as saying
that a dynamical system being point dissipative in Rn implies that there is
a maximal compact invariant set which attracts compact sets. It also can
be considered a consequence of the fact that T (t), t G+ , t > 0, is a com-
pact map. Motivated by specic applications, we shall discuss in the next
sections generalizations which exploit only the continuity of T (t)x as well
as ones for which additional properties of T (t) are assumed.
Before doing so, we mention a fundamental result of Billotti and
LaSalle (1971) (see also Gerstein and Krasnoselski (1968), Gerstein (1970))
for the situation where there is an r > 0 for which T (t) is compact for
t r.
Theorem 5.1. If the dynamical system T is point dissipative and there

is an r > 0 such that T (r) is compact, then the compact global attractor
exists.
Motivated by applications to NFDE, the following asymptotic xed

point theorem (Theorem 5.5) was proved independently and with differ-
ent proofs by Nussbaum (1972) and Hale and Lopes (1973)). To state the
result, we need some additional concepts.
Denition 5.2. If B is a bounded set in a Banach space X, the

Kuratowski measure of noncompactness (B) of B is dened as
(B) = inf {d : B has a nite cover of diameter < d}.

492 Hale
The term in the denition is used because (B) = 0 if and only if Cl B

is compact.
Denition 5.3. If X is a Banach space, then a continuous map T :

X X is -condensing if (T B) < (B) for any bounded set B for which
T B is bounded and (B) > 0. The map T is -contracting if there is
a constant k [0, 1) such that (T B)k(B) for any bounded set B for
which T B is bounded.
Denition 5.4. A dynamical system T on X is compact dissipative if

there is a bounded set B in X such that, for any compact set K in X, there
is a t0 = t0 (K, B) G+ such that T (t)K B for t t0 .
The following result is due to Nussbaum (1972) and Hale and Lopes
(1973).
Theorem 5.5. If X is a Banach space and T : X X is a continuous

map which is -condensing and compact dissipative, then there is a xed
point of T .
If T is a dynamical system on X for which there is an r > 0 such

that T (r) is compact, then one can prove that T being point dissipative is
equivalent to T being compact dissipative. If, in addition, T is a discrete
dynamical system, then Theorem 5.5 implies that T r has a xed point. If
T is a continuous dynamical system such that T (t) is compact for each t >
0, then one can nd a xed point x of T ( ) for any > 0. Using a com-
pactness argument (see Hale and Lopes (1973)), one can then show that
there a sequence of j 0 as j and a point x0 such that xj x0 as
j and x0 is an equilibrium of T . We state the nal result as
Theorem 5.6. If the discrete dynamical system on X dened by a com-

pact map T : X X is point dissipative, then there is a xed point of T .
If a continuous dynamical system T (t), t 0, on X has the property
that T (t) is compact for each t > 0, then there is an equilibrium of the
dynamical system.
This result has direct applications to certain types of evolutionary

equations
t u = Au + f (t, u), (5.1)
Suppose that the solution u(t, , u0 ) through u0 at is unique and dened

for all t . If we dene T (t, )u0 = u(t, , u0 ) and suppose that T (t, ) is
compact for each t > s, then we obtain from Theorem 5.6 the existence of
a xed point of T (, 0) if f has minimal period > 0 in t and an equi-

librium of the corresponding continuous dynamical system if f is indepen-
dent of t. These remarks are valid for a large class of quasilinear parabolic
equations.
Let r 0 be a constant, C = C([r, 0], Rn ) and, for any R and
any continuous function x : [ r, ) Rn , > , let xt C, t [, ), be
dened by xt ( ) = x(t + ), [r, 0]. A retarded functional differential
equation (RFDE) is
x(t) = f (t, xt ). (5.2)
For given R, a function x(t, , ), t r is a solution of (5.2)

with initial value at t = if x(t, , ) is continuous, x(t, , ) = (t)
for t [ r, ], x(t) has a continuous right hand derivative for t and
satises (6.2) for t .
Suppose that f : R C Rn satises enough conditions to ensure
that, for any C and any R, there is a unique solution x(t, , ),
x0 (, , ) = , which exists and is unique for all t .
Dene T (t, ) = xt (, , ). If f is a bounded map, then it is not dif-
cult to prove that T ( + r, ) is compact for each R. If f (t, ) is -peri-
odic in t, then the Poincare map T (, 0) is compact if r and Theorem 5.6
implies that there is an -periodic solution of (5.2) if T (, 0) is point dis-
sipative. In 1960s, many papers were devoted to the existence of -periodic
solutions of (5.2) making use of this compactness property and, therefore,
it was assumed always that the period was as large as the delay.
Hale and Lopes (1973) observed that, for any > 0, the Poincare map
was always an -contraction. Some iterate of the Poincare map is compact
and, therefore, point dissipative is equivalent to compact dissipative. As a
consequence, Theorem 5.5 implies the existence of an -periodic solution
without the restriction that r.
If f in (5.2) is independent of t, Hale and Lopes (1973) also showed
that the corresponding dynamical system has an equilibrium.
6. LINEAR POINT DISSIPATIVE SYSTEMS

It is instructive to rst discuss linear dynamical systems which are
point dissipative. For a bounded linear map T on a Banach space X,
we let r( (T )) (respectively re ( (T ))) denote the radius of the spectrum
(respectively essential spectrum) of T .
The following result is easily proved.
Theorem 6.1. If X is a Banach space, T (t): X X, t G+ , is a linear

dynamical system on X which is point dissipative, then r( (T (1))1 and
494 Hale
(i) {0} attracts compact sets and is the maximal compact invariant
set;
(ii) + (B) is bounded if B is bounded and {0} is stable;
(iii) {0} is the compact global attractor if and only if r( (T (1))) < 1;
(iv) If re ( (T (1))) = 1, then {0} is not a compact global attractor.
Example 6.2. (A Parabolic Equation) Consider the linear parabolic

equation
t u u = 0, x (6.1)
with u = 0 on the boundary of the smooth bounded domain . The

space for the initial data is X = H01 (). If T (t), X, is the solution of
(6.1) through at t = 0, then it is well known that T (t) 0 as t
and, therefore, T is point dissipative. Also, T (1) is a compact operator
for t > 0 which implies that re ( (T (1)) = 0. This fact and point dissipative-
ness imply that r( (T (1))) < 1. By Theorem 6.1, the origin is the compact
global attractor.
Example 6.3. (A simple map on a Hilbert space) Let X be a real,

separable Hilbert space with a complete orthonormal basis {ej , j 1} and
let {j , j 1} be a sequence of real numbers satisfying 0 < j < 1, j 1.
Dene an operator T on the basis vectors as T ej = j ej , j 1. Extend the
denition of T to all of X by the relation
T x = j 1 j x j ej , x = j 1 xj ej .
It is easy to see that T n x 0 as n for every x X. If we assume that

there is a < 1 such that j < for all j , then the origin is the compact
global attractor of the linear dynamical system dened by T . If j 1 as
j , then r( (T )) = re ( (T )) = 1. Therefore {0} is the maximal compact
invariant set, is stable, attracts compact sets, but is not a compact global
attractor. Furthermore, + (B) is bounded if B is bounded.
Example 6.4. If the linear dynamical system T (t), t 0, is generated

by a RFDE with the vector eld f () linear in , then we have observed
that T (r) is compact which implies that, for each t > 0, re ( (T (t)) = 0. If T
is point dissipative then | (T (1))| < 1 and the origin is the compact global
attractor.
Example 6.5. (A Neutral DifferentialDifference Equation) Suppose

that 0a1 and consider the equation
d
[x(t) ax(t 1)] + cx(t) = 0 (6.2)
dt
with initial data in the space C C([1, 0], R). A function x(t, ), t 1
is a solution of (5.2) with initial at t = 0 if x(t, ) is continuous, x(t, ) =
(t) for t [1, 0], x(t) ax(t 1) has a continuous right hand derivative
for t 0 and satises (6.2) for t 0.
Equation (6.2) is a RFDE if a = 0 and is called a neutral differential

difference equation if a = 0.
As noted earlier, if x(t, ) is the solution of this equation with ini-
tial data at t = 0, we let T (t): C C, t 0, be dened by (T (t))( ) =
x(t + , ). It is known that (T (1)) = Cl{e : (1 ae ) + c = 0} (Henry
(1987)).
One can show that re ( (T (1))) = a, each solution of the equation
(1 ae ) + c = 0 has real part < 0. It also can be shown that every solu-
tion approaches zero as t (point dissipativeness).
If a = 0 (the retarded case) or the neutral case with a < 1, Theorem 6.1
implies that the origin is the compact global attractor.
For a = 1, the values e accumulate on the unit circle and therefore,
r( (T (1))) = re ( (T (1))) = 1. As a consequence, {0} is the maximal com-
pact invariant set, is stable and is not a compact global attractor. Further-
more, + (B) is bounded if B is bounded.
Example 6.6. (Linearly damped wave equation) Consider the linearly
damped linear wave equation
t2 u + (x)t u u = 0, x , (6.3)
with Dirichlet boundary conditions u = 0 on the boundary of the

smooth bounded domain . The space for the initial data is X = H01 ()
L2 () and we assume that the function (x) is smooth and nonnegative.
For any (, ) X, let T (t)(, ) be the solution of (6.3) with initial data
(, ) at t = 0. If (x) > 0 at some x , then each solution approaches
zero as t (Iwasaki (1969), Dafermos (1978)). Theorem 6.1 implies
that {0} is the maximal compact invariant set for (6.3), it is stable, attracts
compact sets and + (B) is bounded if B is bounded. The point {0} is
the compact global attractor if and only if every light ray of the equation
t2 u u = 0 intersects the support of (Bardos et al. (1991)). In particu-
lar, if (x) > 0 for all x or even if the support of contains the bound-
ary of , then {0} is the compact global attractor.
Remark 6.7. From Theorem 6.1, we see that point dissipative is

equivalent to compact dissipative if the dynamical system is linear.
This is not true for general dynamical systems. In fact, Cholewa and
Hale (2000) have given a discrete dynamical system on a complete metric
496 Hale
space X which is point dissipative but not compact dissipative. More spe-
cically, they dene a map T : X X which has the property that there is
a compact invariant set J which attracts points of X, but there is a point
x0 J for which, for any > 0 and each integer 1, + (T (B(x0 , )) is
unbounded, where B(x0 , ) is the -neighborhood of x0 . The map con-
structed by Cholewa and Hale (2000) also is asymptotically smooth. A
map T : X X is asymptotically smooth if, for any bounded set B X for
which T n B B, n 0, there is a compact set J which attracts B. Using
the following important result, we conclude that the constructed map T is
not compact dissipative.
Theorem 6.8. If T : X X is an asymptotically smooth map which is

compact dissipative, then there is compact invariant set which attracts com-
pact sets of X.
If T is asymptotically smooth and attracts compact sets, then, for any
compact set K, there is a neighborhood U of K such that T attracts U .
It certainly would be interesting to characterize those nonlinear

dynamical systems for which point dissipative and compact dissipative are
equivalent. As remarked earlier, compact dissipative played an important
role in the xed point theorem. In the following sections, we will see that
compact dissipative also plays an important role.
7. COMPACT SETS ATTRACTING COMPACT SETS

In Theorem 6.1 for linear dynamical systems, we have seen that the
existence of the compact set {0} attracting points is equivalent to {0}
attracting compact sets. In Remark 6.7, we have noted that there are
dynamical systems for which a compact invariant set may attract each
point of X, but may not attract each compact set of X. It becomes nat-
ural to understand the implications of assuming that a dynamical system
has the property that there is a compact set which attracts each compact
set of X. To state the next result, we need the following concept due to
LaSalle (1976).
Denition 7.1. Let T (t), t G+ , be a dynamical system on X. A

closed invariant set A X is said to be invariantly connected if it cannot
be represented as the union of two nonempty, disjoint, closed positively
invariant sets.
Denition 7.2. An invariant set J of the dynamical system T is said

to be Lyapunov stable for T if, for any neighborhood V of J , there is a
neighborhood W of J such that T (t)W V for t G+ .
Parts (i)(iii) of the following theorem are due to Hale et al. (1972),
part (iv) to LaSalle (1976) and Massatt (1983).
Theorem 7.3. If T (t): X X, t G+ , is a dynamical system on X

such that there is a compact set K which attracts compact sets of X and
A = tG+ T (t)K, then A is independent of K,
(i) A is the maximal compact invariant set;
(ii) A is Lyapunov stable and attracts compact sets;
(iii) for any compact set J in X, there is a neighborhood U of J such
that + (U ) is bounded;
(iv) A is invariantly connected if X is connected, is connected if X is
a Banach space, and always connected for continuous dynamical
systems.
Remark 7.4. It might be expected that the maximal compact invari-

ant set always is connected if X is connected. However, this may not be
the case. Gobbino and Sardela (1997) have given an example showing that
there exists a connected metric space X and a continuous map T : X X
such that the compact global attractor exists and is not connected.
Theorem 6.1 for linear systems shows that there are dynamical sys-
tems for which the hypotheses of Theorem 7.3 are satised and the
maximal compact invariant set A is not a local attractor. To obtain a
local attractor, the dynamical system T not only must be continuous, but
must satisfy some additional structural conditions. In the next section, we
describe a large class of such systems.
8. THE MAXIMAL COMPACT INVARIANT SET

AS AN ATTRACTOR
In Theorem 7.3, we have exhibited a general class of dynamical sys-
tem for which the maximal compact invariant set A is stable. Therefore,
there is a neighborhood U of A such that + (U ) is bounded. If we knew
that + (U ) had a compact -limit set, then (U ) A and A is a local
attractor. This remark allows us to identify an interesting class of dynam-
ical systems which occur often in dissipative dynamical systems.
The following concept is due to Hale et al. (1972).
Denition 8.1. A dynamical system T (t), t G+ , on X is said to be

asymptotically smooth if, for any bounded set B X for which T (t)B B
for t G+ , there is a compact set J X such that J attracts B.
498 Hale
It is important to remark that the concept of asymptotically smooth

says nothing about the dynamics dened by a dynamical system. There
may not be any bounded orbits for the system and it can be asymp-
totically smooth. Any ordinary differential equation in Rn for which the
solutions are dened for all t R+ denes a dynamical system which is
asymptotically smooth. This is a consequence of the compactness of the
closed ball in Rn . The differential equation may not have any bounded
solution as, for example, y = 1. If an ordinary differential equation has a
nonempty compact invariant set, then every solution can be unbounded
except for those solutions which have initial data in the maximal compact
invariant; for example, y = y.
A linear system T (t), t G+ , is asymptotically smooth if and only if
r(e (T (1))) < 1.
If there is an r > 0 such that T (r) is compact, then T is asymptoti-
cally smooth.
An important more general class of dynamical systems are those for
which
T (t) = S(t) + U (t), (8.1a)
where U (t) is compact for t G+ . Such dynamical systems are asymptot-

ically smooth if and only if there exist k > 0, > 0 such that
|S(t)x| ket |x|, x X, t G+ . (8.1b)
The illustrations given below will be of this type. However, asymptot-

ically smooth dynamical systems are much more general (see, e.g., Hale
(1988), Raugel (2002)).
It is possible to prove the following equivalent characterization of
asymptotically smooth dynamical systems.
Proposition 8.2. The dynamical system T (t), t G+ , on X is asymp-

totically smooth if and only if, for any bounded set B X for which there is
a t0 such that + (T (t0 )B) is bounded, any set of the form T (tj )xj , xj B,
tj as j , tj t0 , j 1, is relatively compact.
Ladyzenskaya (1987) introduced the concept of asymptotically com-

pact dynamical systems as those which satisfy the characterization in
Proposition 8.2. Ball (1997) and Sell and You (2002) have also used the
term asymptotically compact for dynamical systems, but their denitions
imply the more restrictive condition that the positive orbits of bounded
sets are bounded. We will consistently use the term asymptotically smooth
to avoid confusion.
From Theorem 7.3, we can now obtain the following result.
Theorem 8.3. If T (t) : X X, t G+ , is an asymptotically smooth

dynamical system on X such that there is a compact set K which attracts
compact sets of X and A = tG+ T (t)K, then
(i) A is the maximal compact invariant set;
(ii) A is a local attractor and, for any compact set J in X, there is
a neighborhood U of J such that A attracts U .
From Theorem 8.3, it is not difcult to prove the following results.
Corollary 8.4. If T (t), t G+ is asymptotically smooth, then the fol-
lowing statements are equivalent:
(i) There is a compact set which attracts compact sets;
(ii) There is a compact set which attracts a neighborhood of each
compact set.
Corollary 8.5. If T (t), t G+ is asymptotically smooth, then there
exists a compact set which attracts compact sets if and only if there is a
compact invariant set which is a local attractor and attracts each point of
X.
Remark 8.6. Sell and You (2002) dened a global attractor of a
dynamical system as a set A which is a local attractor and attracts each
point of X. This denition is not the same as the one that we have used
above. If T is asymptotically smooth, from Corollary 8.5, the existence of
a compact global attractor in the sense of Sell and You is the same as the
existence of a compact set which attracts compact sets.
Another interesting consequence of the above results is the following.
Theorem 8.7. If
(i) T (t), t G+ , is asymptotically smooth;
(ii) T (t), t G+ , is point dissipative;
(iii) for any compact set K X, there is a t0 = t0 (K) such that
+ (T (t0 )K) is bounded,
then
(1) there exists the maximal compact set A;
(2) A is a local attractor;
(3) for each compact set K X, there is a neighborhood U of K such
that A attracts U .
From Theorem 8.7, we see that, if T is asymptotically smooth and
point dissipative, then it is compact dissipative if and only if + (K) is
500 Hale
bounded for each compact set K X. Can on describe a large class of

dynamical systems which occur in the applications which have the prop-
erty that + (K) is bounded for each compact set K X?
9. COMPACT GLOBAL ATTRACTORS

The following result is the fundamental theorem on the existence of a
compact global attractor.
Theorem 9.1. A dynamical system T (t), t G+ , has a compact global
attractor A if and only if
(i) T (t), t G+ , is asymptotically smooth;
(ii) T (t), t G+ , is point dissipative;
(iii) for any bounded set B X, there is a t0 = t0 (B) such that
+ T (t0 )B) is bounded.
The necessity in Theorem 9.1 under the assumption that t0 (B) = 0 for
all B (that is, + (B) is bounded if B is bounded) was proved by Hale
et al. (1972) taking into account renements by Cooperman (1978) and
Massatt (1980). The sufciency is shown by means of a example of Cho-
lewa and Hale (2000) for which the compact global attractor exists and yet
there is a bounded set B X for which + (B) is unbounded. On the other
hand, the existence of the compact global attractor implies that (iii) must
be satised.
One can obtain easily necessary and sufcient conditions for the exis-
tence of compact global attractors in another way.
Denition 9.2. A bounded set B in X is an absorbing set for T (t):
X X, t G+ , if, for any bounded set U X, there exists t0 (B, U ) such
that T (t)U B if t t0 (B, U ). The existence of an absorbing set is equiv-
alent to bounded dissipative or uniform ultimate boundedness, a concept
familiar in stability theory of ordinary differential equations. If A is the
compact global attractor for the dynamical system T , then any bounded
neighborhood of A is an absorbing set.
Theorem 9.3. A dynamical system T on X has a compact global
attractor if and only if
(i) T is asymptotically smooth,
(ii) there is an absorbing set for T .
In the applications, the verication of the existence of an absorbing
set often involves the construction of a Lyapunov function which is strictly
decreasing along orbits with large initial data. Such a function yields an
explicit bound for the compact global attractor. In many applications to

PDE, the Lyapunov function is a modied energy function.
If it is assumed that a dynamical system has a compact global attrac-
tor, then it is possible to prove that there exists a Lyapunov function
which is strictly decreasing along orbits with large initial data. However,
this is only an existence theorem and does not lead to any bounds on the
attractor.
On the other hand, there are important situations where one can use
Theorem 9.1 to show that the compact global attractor exists, whereas the
use of Theorem 9.3 for existence may be extremely difcult or perhaps
impossible. This is a familiar situation in ordinary differential equations
(even in dimension 2).
10. GRADIENT SYSTEMS

A stationary point of a dynamical system T is a xed point of T if
the dynamical system is discrete and is an equilibrium if T is a continuous
dynamical system. Let E(T ) be the set of stationary points of a dynamical
system T .
Denition 10.1. A dynamical system T is a gradient dynamical sys-

tem if there is a continuous function V : X R (called a Lyapunov function)
with the property that
(i) V (T (t)x)V (x) t [0, ) R+ , x X;
(ii) V (T (t)x) = V (x) for t G+ implies x E(T ).
If F C 1 (Rn , R), then the equation
x = F (x)
denes a gradient dynamical system with F (x) being a Lyapunov function.

In fact, the derivative F (x) along the solution through x at t = 0 is given
by
V (x) = |F (x)|2 0, x Rn .
If F (T (t)x) = F (x) for t 0, then F (x) = 0 and x is an equilibrium.

Other examples of gradient systems will be given later.
Theorem 10.2. If T is a gradient dynamical system, then

(i) if + (x
0 ) is relatively compact, then (x0 ) E(T );
(ii) if (x0 ) is relatively compact, then (x0 ) E(T ).
502 Hale
Corollary 10.3. If the dynamical system T is asymptotically smooth

and gradient, then + (x0 ) bounded implies that (x0 ) E(T ).
Suppose that the dynamical system T is a gradient system which has

the compact global attractor A(T ). Let W u (E(T )) be the unstable set of
E(T ); that is, the set of x X such that T (t)x is dened for t G+ and
{T (t)x : t G+ } is bounded. Then A(T ) = W u (E(T )). A point x in E(T )
is hyperbolic if the dynamical system generated by the linearization about
x has no spectrum on the unit circle for t > 0, t G+ . If each element
of E(T ) is hyperbolic, then there are only a nite number of elements of
E(T ) and A(T ) = xE(T ) W u (x), where W u (x) is the unstable manifold of
x. In this case, we can state the following result.
Theorem 10.4. If a dynamical system is gradient, possesses a compact

global attractor with each element of E(T ) hyperbolic with transversal inter-
section of stable and unstable manifolds, then T is MorseSmale.
11. RETARDED AND NEUTRAL FUNCTIONAL DIFFERENTIAL

EQUATIONS
In an earlier section, we have mentioned RFDE and neutral func-
tional differential equations (NFDE). We now discuss these equations in
the context of the results above on maximal compact invariant sets and
compact global attractors. In spite of some repetition, we repeat the nota-
tion and some of the denitions.
Fix r > 0 and let C = C([r, 0], Rn ). For a given x C([r, ), Rn ), let
xt C be dened by xt ( ) = x(t + ), [r, 0].
Let f : C Rn be a C 1 function. A RFDE is
x(t) = f (xt ). (11.1)
A function x(t, ), C, t [r, ), > 0, is a solution of (10.1) with ini-

tial data at t = 0 if x0 (, ) = , x(t, ) has a continuous right hand deriv-
ative on [0, ) and satises (11.1). If we dene
Tf (t) = xt (, ), (11.2)
then T (t) is a dynamical system on C if solutions are dened for all t 0.

Furthermore, Tf (r) is compact since f is a bounded map. Hale and Lopes
(1973) have shown that Tf (t) can be written as a sum as in (8.1a), (8.1b)
and, thus, is asymptotically smooth. Since Tf (r) is compact, it is possible to
prove that (i) and (ii) in Theorem 8.7 (respectively, Theorem 9.1) imply that
condition (iii) in Theorem 8.7 (respectively, Theorem 9.1) is automatically
satised and the existence of the maximal compact invariant set (respectively,
compact global attractor) satisfying (2), (3) in Theorem 8.7.
We now dene an important class of NFDE introduced by Hale
and Meyer (1967) which includes RFDE. Let D: C Rn be a bounded
continuous linear operator which is atomic at zero. A NFDE is
d
Dxt = f (xt ). (11.3)
dt
The NFDE is a RFDE if and only if D = (0) for C.
A function x(t, ), C, t [r, ), > 0, is a solution of (11.3) with
initial data at t = 0 if x0 (, ) = , Dxt (, ) has a continuous right hand
derivative on [0, ) and satises (11.3). If we dene TD,f (t) = xt (, ) and
all solutions are dened for t 0, then TD,f (t), t 0, is a dynamical sys-
tem However, it may not be compact for any t > 0. For example, if D =
(0) A(r), det A = 0.
Since f is a compact map from C to Rn , the condition for TD,F to be
asymptotically smooth should be determined from properties of D. This is
actually the case.
Let CD = { C : D = 0} and consider the difference equation
Dxt = 0 (11.4)
with initial value in CD . One can show that (11.4) denes a dynamical sys-
tem TD,0 (t) on CD . We say that D is exponentially stable if k > 0, > 0,
such that
TD,0 (t)L(CD ,CD ) ket , t 0.
The following result is due to Cruz and Hale (1970).
Theorem 11.1. The dynamical system TD,f on C is asymptotically

smooth if and only if TD,0 is exponentially stable on CD .
The proof consists of rst showing that there is a nite dimensional

map : C CD such that D = I , the identity. If we write
TD,f (t) = TD,0 (t)(I D) + U (t), (11.5)
then
t
DU (t)x0 = D + f (TD,f ( )) d. (11.6)
0
Using the fact that D is exponentially stable together with some nontriv-
ial estimates, we see that TD,f (t) has the form (11.5), (11.6) with U (t)
504 Hale
compact for t 0. The operator TD,f (t) satises (8.1a) and the conclusion
of the theorem is valid.
It is interesting to ask if the set of MorseSmale systems is dense in
the set of RFDE and NFDE. For RFDE, Oliva (1969) has shown that
there is a residual set of RFDE with f C 1 such that each equilibrium
is hyperbolic. Mallet-Paret (1977) has proved the same thing for periodic
orbits. It is not known if this is true for the transversal intersection of sta-
ble and unstable equilibria and periodic orbits. As a consequence, it is not
known if being Morse-Smale is generic in the class of RFDE for which
there exists a compact global attractor.
The same type of proof as in Oliva (1969) and Mallet-Paret (1977)
should apply to NFDE with an exponentially stable D operator.
Some examples of MorseSmale systems for FDE are contained in
Hale et al. (2002).
12. REALIZATION ON CENTER MANIFOLDS OF FDE

The meaning of realization comes from the following observation.
Since the ow for a RFDE or NFDE evolves in an innite dimensional
space, for any given integers N and n, it is perhaps to be expected that,
for any system of ordinary differential equations of dimension N, there is
an RFDE of dimension n and an invariant set J of the RFDE such that
the ow for ODE is equivalent to the ow of on J . Of course, if n N,
this is the case. If n < N , we will observe below that there are ows dened
by an N-dimensional ODE which cannot be realized by an RFDE in Rn .
We verify this observation by considering the restrictions imposed on the
ODE which determines the ow on a center manifold.
We discuss only the retarded case and remark that the same results
are true for the neutral case for an exponentially stable D. Furthermore,
we consider only a scalar equation and refer to Hale, et al. (2002) for the
general case and complete proofs.
Consider the scalar RFDE
x(t) = Lxt , (12.1)
where L: C R is a bounded linear operator. Let TL (t), t 0, be the semi-

group generated by (12.1). The innitesimal generator AL of TL (t) is given
by
D(AL ) = { C : d d(0)
d C, d = L}, (12.2)
AL ( ) = d()
d , [, 0].
The spectrum of AL is only point spectrum (also called the eigenvalues of

(12.1)) is given by the solutions of the characteristic equation
det () = 0, () = L(e ). (12.3)
The point spectrum of TL (t) consists of the set of the elements et , where
is a solution of (12.3).
We note rst that, for any integer m, there is a linear operator L such
that there are exactly m eigenvalues (counting multiplicity) on the imagi-
nary axis and, therefore, m elements of the point spectrum of TL (t) with
modulus one.
If is an eigenvalue of (12.3), then e , [r, 0], is an eigenfunction
of AL . If has multiplicity p, then the generalized eigenspace of of AL
is spanned by the functions (e , e , . . . , p1 e ), [r, 0]. The gen-
eralized eigenspace of et of TL (t) is given by the span of the functions
(e(t+) , (t + )e(t+) , . . . , (t + )p1 e(t+l) ), [r, 0]. This is due to the
fact that (12.1) is a scalar equation.
Suppose that there are exactly m eigenvalues j , 1j m, on the
imaginary axis, counting multiplicity. Let = (1 , . . . , m ) be a basis for
the corresponding generalized eigenspaces of AL . The elements of span
a linear subspace P of dimension m which is invariant under the semi-
group TL (t), t 0. Also, there is an m m matrix B such that TL (t) =
exp (Bt); that is, the ow on the manifold P for (11.1) is equivalent to
the linear ordinary differential equation on Rm :
y = By. (12.4)
There also is a linear subspace Q of C which is invariant under TL (t)

and C = P Q. The spectrum of TL (t) restricted to Q is all of the spec-
trum of TL (t) except the elements corresponding to the eigenvalues on the
imaginary axis. Let : C P be a continuous linear projection such that
C = P , (I )C = Q.
Along with (12.1), consider the perturbed linear system
x(t) = Lxt + f (xt ), (12.5)
where f : C R is C p , p 1.
If f is sufciently small, we can apply the local theory of invariant
manifolds near the origin to determine a center manifold CM(f ) of (12.5)
given by
CM(f ) = { C : = y + h(f, y)}, (12.6)

506 Hale
where y Rm varies in a neighborhood of zero and h(f, y) Q, h(0, y) = 0,

is C r in y. The ow of (12.5) on CM(f ) is given by xt = y(t) + h(f, y(t)),
where y(t) is a solution of the ordinary differential equation
y = By + bf (y + h(f, y)) By + bY (f, y), (12.7)
where b is an m-vector determined by the projection .
To obtain other properties of (12.7), we use the concept of complete
controllability. The pair (B, b) is said to be completely controllable if the
matrix [b, Bb, . . . , B m1 b] is nonsingular. It can be shown that (B, b) is
completely controllable.
Since (B, b) is completely controllable, it is known that there is a non-
singular matrix C such that

0 1 0 0 0
0 0 1 0 ..

CBC 1 = . .. .. .. , Cb = . . (12.8)
.. . . . 0
am am1 am2 a1 1
Therefore, the change of variables y = C 1 z, in (12.7) implies that z =

(z1 , . . . , zm ) is a solution of
z = CBC 1 z + CbY (f, C 1 z). (12.9)
(j 1)
From (12.8), zj = zj +1 , j = 1, 2, . . . , m 1, zj = j = 2, 3, . . . , m,
z1 ,
(k)
where z1 is the kth derivative of z1 . Therefore, Eq. (12.9) is equivalent
to an mth-order scalar ODE,
+ + am z1 = Y (f, C 1 z).
(m) (m1)
z 1 + a1 z 1 (12.10)
Note that Y (f, C 1 z) depends only upon z1 for 0km 1.

(k)
The fact that the ow on a center manifold satises (12.10) was rst
observed by Hale (1985a). This already yields some interesting information
about RFDE. Even though our scalar RFDE is an evolutionary equa-
tion in the innite dimensional space C, the ow on a center manifold of
dimension m is restricted to the form (12.10). In particular, if m > 1, there
are systems of ODE of dimension m which cannot be realized as the ow
on a center manifold of dimension m. For example, for m = 2, the ow on
a center manifold is given by a second order equation (12.10). Therefore,
in the (z1 , z1 )-plane, all stationary points of the ow on the center mani-
fold are on the z1 axis. There is an ODE in the plane which will not be
equivalent to the ow on the center manifold.
Knowing that the ow on a center manifold has the above restric-
tions, we arrive at another realization problem. For a given small
g C p (Rm , R), can one nd an f C p (C, R) such that Y (f, C 1 z) = g(z)?

This can be accomplished with only using functions which depend only
upon m delays; that is, by differential difference equations (Hale (1985a)).
We state this precisely as
Theorem 12.1. If Y (f, y) is dened as in (12.7) and C as in (12.8),

then there exist constants 0 < 1 < < m , depending only upon the basis ,
such that, if, for any F C r (Rm , R), we dene fF C r (C, R) by the relation
fF () = F ((1 ), (2 ), . . . , (m )), (12.11)
then there are positive constants r , r such that, for any g C r (Rm , R),
gr < r , there exists an F C r (Rm , R) with F r < r such that
Y (fF , C 1 z) = g(z), (12.12)
that is, the ow on a center manifold can be realized by the differential

difference equation
x(t) = F (x(t 1 ), x(t 2 ), . . . , x(t m )). (12.13)
In Hale (1985a), it was shown that the jet of every vector eld could
be realized (it was falsely claimed that ever vector eld could be real-
ized). Faria and Magalhaes (1995a) extended the result to n dimensions
using their theory of normal forms (1995b, c). Rybakowski (1994) used
the Nash-Moser iteration technique to realize all vector elds in C m+15
by functions f C m if m 17. Theorem 12.1 was proved by Prizzi (2000)
using some of the ideas of Polacik and Rybakowski (1995) in their work
on realization of vector elds on center manifolds for parabolic equations.
For specic applications, see the references above as well as Hale
et al. (2002).
Remark 12.2. The ow on the center manifold of either an RFDE

or a NFDE can have a compact global attractor AL+f . If the remaining
eigenvalues of the linear equation (12.1) have real parts 0, then AL+f
will be a local attractor for (12.5). For m large, this makes it possible to
obtain stable chaotic behavior near the origin that is induced by the small
perturbation F .
13. A VISCOELASTIC MODEL

This section is devoted to a model in viscoelasticity which we present
in some detail because it illustrates well some of the ideas that we have
been discussing above. It is a gradient system for which one can prove the
508 Hale
existence of the compact global attractor and describe completely the ow

on the attractor. It will make use of a Lyapunov function, but it does not
seem possible to have such a function which is strictly decreasing along
solutions with large initial data. Therefore, the use of absorbing sets to
prove the existence of the attractor does not seem to be useful.
Suppose g C 1 (R, R), r > 0, a C 2 ([0, r], R)
a(0) = 1, a() = 0, a(s) > 0 for s [0, ],
a (s) 0, a (s) 0, s [0, ], (13.1)
x
G(x) = 0 g(u)du as |x|
and consider the equation
0
x(t) = a( )g(x(t + ))d. (13.2)

This is a RFDE with delay . If C = C([, 0], R), then a solution x(t, )
through at zero is dened on some interval [, ), > 0. If = for
each such solution, then (13.2) denes a dynamical on C, which we denote
by
(Ta,g (t))( ) = x(t + , ), [, 0].
Theorem 13.1. The dened system Ta,g is well dened on C. Also, if

there is an s0 [0, ] such that a (s0 ) > 0, then Ta,g is a gradient system on
C.
If + () is bounded, then () is a singleton.
If, in addition, the zeros of g belong to a bounded set, then the compact
global attractor A(a, g) exists.
Furthermore, if g has a nite number a1 < a2 < < a2k+1 of zeros with
each being simple, then each equilibrium is hyperbolic with the odd numbered
ones being hyperbolically stable, the even numbered ones being hyperbolic
saddles with one dimensional unstable manifolds and
A(a, g) = 0j k W u (a2j +1 ) = Cl 1k W u (a2j ). (13.3)
Finally, if the stable and unstable manifolds of the aj are transversal, then
Aa,g is a smooth one dimensional manifold and Ta,g is MorseSmale.
Under the assumption that a (s > 0 for all s and the additional
assumption that g has only one zero which is simple, Levin and Nohel
(1964) proved that every solution must approach the zero of g. Hale
(1965) developed a general theory of stability for RFDE based on ideas
from dynamical systems and obtained the general result stated for simple
zeros of g. The other statements in the theorem came much later after the
concept of attractors was introduced. We will indicate the proof below and
more details can be found in Hale et al. (2002).
To discuss the ow on the attractor in more detail, we let Gk be the
family of functions g which have 2k + 1 zeros a1 < a2 < < a2k+1 with
each being simple. We say that a2j is connected to {ap , aq } if, for each
x W u (a2j \ {a2j }, (x) = aq and (x) = ap .
If k = 0 (one zero) then the compact global attractor is a singleton.
If k = 1 (three zeros), then a2 is connected to {a1 , a3 } for every g G3 .
The ow on A(a, g) is equivalent to the ow of the ODE y = g(y) on
its compact global attractor.
If the attractor is a smooth manifold, then one would expect that of
the zeros of g on Aa,g is the same as the order of these zeros on the real
line; that is, the ow on the attractor is equivalent to the ow of the ODE
y = g(y). However, this may not be the case as the following result of Hale
and Rybakowski (1982) shows.
Theorem 13.2. Suppose that the conditions of Theorem 13.1 are satis-
ed. In the family G3 (that is, the set of g with ve zeros a1 < a2 < < a5 ,
each being simple), each of the following ows on the attractor can be real-
ized:
(i) a2 is connected to {a1 , a3 }, a4 is connected to {a3 , a5 };
(ii) a2 is connected to {a1 , a4 }, a4 is connected to {a3 , a5 };
(iii) a2 is connected to {a1 , a5 }, a4 is connected to {a3 , a5 };
(iv) a2 is connected to {a1 , a3 }, a4 is connected to {a2 , a5 };
(v) a2 is connected to {a1 , a3 }, a4 is connected to {a1 , a5 }.
Cases (i), (iii), and (v) are MorseSmale and cases (iii) and (v) do not
preserve the ordering of the zeros of g on the real line. Cases (ii) and (iv) are
not MorseSmale since the stable and unstable manifolds of a2 and a4 do
not intersect transversally. The nontransversal cases (ii) and (iv) are proved
to exist by showing that one can obtain the MorseSmale case through a
homotopy of the functions in G3 . However, it is not known if the Morse
Smale classes are generic in G3 ; that is, if the system is not MorseSmale,
can one perturb g a small amount and obtain transversality?
13.1. Brief Idea of the Proof of Theorem 13.1

If
0 0
1
V () = G((0)) a ( ) [g((s))ds]2 d, (13.4)
2
510 Hale
then
0
1
V () = a () [g((s))ds]2 (13.5)
2
0
1 0
a ( ) [g((s))ds]2 d 0.
2
These relations imply that each solution of (13.1) and (13.2) is

dened and bounded for t 0 and Ta,g is a dynamical system on C
C(, 0], R). Since T (t) is compact for t , () is compact and invari-
ant and must belong to the set

0
S = C : g(( ))d = 0 if a () = 0, (13.6)

0 (s) > 0 .
s g(( ))d = 0 if a
Next observe that any solution of (13.1) and (13.2) dened and bounded
on R must satisfy
0
x(t) + g(x(t)) = a () g((s))ds (13.7)

0 0
+ a ( ) g((s))ds d 0.

Since () S, a solution x(t) of (13.1) and (13.2) corresponding to an

orbit in () must satisfy
x + g(x) = 0. (13.8)
There is an interval Is0 containing s0 such that a (s) = 0 for s Is0 . A
solution x(t) of (13.1)
0 and (13.2) corresponding to an orbit in () sat-
ises (13.8) and s g(x(t + ))d = 0, for t R, s Is0 , it follows that
x(t) = x(t s) for s Is0 . As a consequence, x(t) is a constant. Since x(t)
is bounded, it follows that x(t) is a constant. This proves that Ta,g is a
gradient dynamical system.
If the set of zeros of G is bounded, then Ta,g is point dissipative and
Theorem 5.1 implies the existence of the compact global attractor.
If a is a zero of g, then one can show that the eigenvalues of the lin-
ear variational equation about a are simple and there is only one eigen-
value with real part 0. If this eigenvalue is positive, then the equilibrium
is hyperbolic with a one dimensional unstable manifold. If the eigenvalue
is zero, it is simple and one can use a result of Hale and Raugel (1992) to
show that a bounded positive orbit converges to a singleton.
For the proof of Theorem 13.2, the reader should consult Hale and
Rybakowski (1982).
The case where a is a linear function (a (s) = 0 for all s) is, in general,
not a gradient system. In fact, using an argument similar to the above, one
can show that any periodic solution of the second order equation (13.8)
is a periodic solution of (13.1) and (13.2). On the other hand, no periodic
solution can be hyperbolic, because a small perturbation in a leads to a
gradient system. There can be no generic Hopf bifurcation. To obtain a
better geometric theory for (13.1) and (13.2), one should allow a to belong
to a class of functions which is not concave. An indication of why this
might be possible can be found in Hale et al. (2002) where the Hopf bifur-
cation is discussed.
14. A 1D SCALAR PARABOLIC EQUATION

In this section, we give several results dealing with the qualitative
behavior of a scalar 1D quasi-linear parabolic equation. It is possible to
show that it is always a gradient system, the transversality property holds
and a complete description of the ow on the compact global attractor is
obtained in a constructive way from a permutation operator on the equi-
libria. Many other results are know about such equations; for example,
the effects of boundary conditions on the ow, characterization of the sta-
ble equilibrium and the effects of nonlocal terms in the equation (see, e.g.,
Hale 1997).
Consider the boundary value problem
ut = (b2 ux )x +f (x, u, ux ) in = (0, 1), (14.1)
ux + u|x=0 = 0 = ux + u|x=1 ,
where
b C([0, 1], R), b > 0, f C 2 (R, R), (14.2)
, , , constants, 2 + 2 = 1 = 2 + 2 .
Assume that there is an s > 0 such that (14.1) and (14.2) denes a local
dynamical system T (t) on H s (0, 1) and that, if B is a bounded set in
H s (0, 1), then the closure of T (t)B is compact if t > 0. If f is indepen-
dent of ux , we can take s = 1. If the growth rate of f in ux is less than
cubic, then we can take s = 2.
Assume that the solutions of
(b2 y ) = f (x, y, y ), y(x0 ) = y0 , y (x0 ) = y0 , (14.3)
are dened for 0x1, 0x0 1, y0 R, y0 R.
512 Hale
Theorem 14.1. (Gradient Structure). Suppose that (14.3) is satised.

(i) System (14.1), (14.2) is gradient.
(ii) + () bounded implies that () is a singleton.
The result is due to Zelenyak (1968). He proved gradient by show-
ing the existence of a Lyapunov function (but did not explicitly con-
struct it). The convergence to a singleton requires a different argument (it
is obtainable from the simplicity of eigenvalues of the linear variational
equation if we use a general result of Hale and Raugel (1992) on con-
vergence of orbits to singletons). If f does not depend upon ux , Matano
(1978) proved the same result. In this case, the natural energy function is
a Lyapunov function.
Let E be the set of equilibria of (14.1) and (14.2).
Theorem 14.2. If (14.1) and (14.2) is point dissipative, then there

exists the compact global attractor A.
If, in addition, the equilibria are hyperbolic, then
A = E W u ().
Theorem 14.3. (Transversality). (Henry (1985), Angenent (1986)). If

, are hyperbolic equilibria of (14.1)(14.3), then W u () is transversal to
W s ().
The transversality property holds for every equation of the form

(14.1)(14.3), independently of the function f and the boundary condi-
tions. For dynamical systems generated by the set of all ordinary differen-
tial equations in a nite dimensional space, such a general transversality
result is valid only in dimension one.
There are other important classes of dynamical systems for which the
transversality property will hold. For example, the ordinary differential
equations obtained by the standard discretization of the spatial variable
in (14.1). There are much more general classes of monotone systems for
which this is true (see, e.g. Hale et al. (2002)).
From Theorems 14.114.3, we see that the systems (14.1)(14.3), is
Morse-Smale if it is point dissipative and there is only a nite number
of equilibria with each being hyperbolic. We can then obtain the follow-
ing result using the result mentioned above of Oliva (2002) on stability of
MorseSmale systems.
Theorem 14.4. Let T be the dynamical system dened by (14.1)

(14.3) and suppose that T is point dissipative and the set E(T ) of equilibria
is nite with each element being hyperbolic. If the compact global attractor
A(T ) is upper semicontinuous in the parameters (b, f, , , , ), then T is

A-stable with respect to perturbations of these parameters.
Assume the existence of the compact global attractor, A(T ), of the

dynamical system T of (14.1) and (14.2) and that the set E(T ) of equi-
libria is given by
E(T ) = {1 , 2 , . . . , k }
with each j hyperbolic. We have previously observed that
A = E W u ().
Given , E, let C(, ) denote the set of connecting orbits from to

; that is, the set of trajectories u(t) which are dened for t R with the
property that (u()) = and (u()) = . It is clear that
A = E (,E C(, )),
C(, ) = W u () W s ().
Suppose that
1 (0) < 2 (0) < < k (0)
and, following Fusco and Rocha (1991), dene the permutation T of the
integers 1, 2, . . . , k as
T (1) (1) < T (2) (1) < < T (k) (1).
This permutation completely determines the ow on the attractor and the

topological classication of the ow. The verication requires several non-
trivial but constructive steps. We only state the result and refer the reader
to the papers of Fiedler and Rocha for the detailed proofs.
Theorem 14.5. (Fiedler and Rocha (1996, 2000)).
(i) The permutation T determines in a constructive way which of the
sets C(, ) are nonempty.
(ii) If T1 and T2 are dynamical systems associated with (14.1) and
(14.2) for different values of the parameters, then T1 = T2
implies that T1 is equivalent to T2 .
514 Hale
The equation
t u = 2 uxx + u u3 in (0, 1) (14.4)
with homogeneous Neumann boundary conditions was rst considered
from a dynamical point of view by Chafee and Infante (1974). It was an
attempt to understand the ow dened by this equation that led to the
nal solution 20 years later. The ow is easily obtained by an applica-
tion of the constructive theory of Fiedler and Rocha using the permuta-
tion. For 1 (n, (n + 1) ), there are exactly 2n + 1 equilibria u0 = 0, u j ,

j = 1, 2, . . . , n 1, un = 1, each of which is hyperbolic. Also, if i des-
ignates index, then i(uo ) = n, i(u j ) = n j , j = 1, 2. . . . , n, the compact
global attractor An has dimension n, is the closure of W u (0) and u+ j +1 is
connected to u+ j (the same with the minus). More complete information
about the dimension of the connecting orbits is known (see Henry (1985)).
15. REMARKS ON AN N D PARABOLIC EQUATION

It is natural to investigate the extent to which the beautiful results for
the 1D parabolic equation hold for the spatial variable x in a bounded
smooth domain RN , N 2. Of course, the second derivatives are
replaced by the Laplacian and ux is replaced by u.
Under rather general conditions, one obtains a dynamical system T (t),
t 0, on H s () for some s > 0 and T (t) is compact for t > 0. Therefore, if
+ () is bounded, then the -limit set () is compact and invariant.
Hirsch (1988) have shown that there is an open dense set H of H s ()
such that, for any H for which + () is bounded, () consists of of
a single point which is an equilibrium. However, it is not true in general
that the system is gradient; that is, that () E(T ), the set of equilib-
ria. There may be more complicated compact invariant sets even chaos
(Polacik (1991, 2002), Polacik and Rybakowski (1995)). Due the fact that
the open dense set H exists with the above mentioned properties, such cha-
otic behavior cannot be stable (contrast this with Remark 12.2).
If f is independent of u, then the system is gradient and a positive
bounded orbit must have its -limit set in E(T ). We discuss more specic
results for this case.
Suppose that is a smooth bounded domain in RN and consider the
system
t u = u + f (x, u) in (15.1)
with the boundary condition u = 0 on . This equation can be consid-
ered in the space H01 ().
Even though the system is gradient, for a ball in R2 , there is a

function f (x, u) and a H01 () for which () is a closed curve which
consists entirely of equilibria (Polacik and Rybakowski (1996)). They prove
this by discussing the realization of vector elds on center manifolds. As
noted above, no such ow can be stable.
A result of Simon (1983) shows that () is a singleton if f is ana-
lytic in u.
Brunovsky and Polacik (1997) have shown that (15.1) being a Morse
Smale system is a generic property in the class of functions f = f (x, u)
depending on both x and u. Polacik (1999) has also exhibited an explicit
function f0 (u) for which (15.1) has equilibria with the stable and unsta-
ble manifold not transversal. Thus, the MorseSmale property does not
hold for f0 . He also shows that there is a C 1 -neighborhood U of f0 in
the class of function which are independent of the spatial variable such
that transversality of stable and unstable manifolds does not hold for any
f U . If one wishes to stay in this class of functions, then nontransversal-
ity must be broken by the variation of some other parameter. A natural
candidate is to perturb in a smooth way. The implications of this type
of perturbation has not been extensively investigated. A recent book of
Henry (2005) contains fundamental results on the calculus associated with
domain variation which should play an important role in such a study.
This book also contains many applications.
The domain in (15.1) could also be given by depending on a
parameter which degenerates to a domain 0 in a lower dimensional
space (the so called thin domains) and one must compare the ow on
with the ow on 0 . The status of such investigations up to 1995 are
contained in the lecture notes of Raugel (1995).
16. LINEARLY DAMPED HYPERBOLIC EQUATION

In a seminal paper, Babin and Vishik (1983) studied the existence of
attractors for hyperbolic PDE with constant linear damping. Their deni-
tion of attractor was not the one that we have given above. They assumed
it attracted special compact sets of the space (bounded sets in the domain
of the generator of the linear part) and not arbitrary bounded sets. Under
the same assumptions that they imposed, independent proofs of the exis-
tence of the compact global attractor as dened above were given by Hale
(1985) and Haraux (1985).
These ideas have been applied to many other types of PDE.
It perhaps is interesting to note that, Hale and Meyer (1967), in the
study of NFDE, encountered the same difculty as Babin and Vishik. It
was resolved by Hale et al. (1972) using asymptotically smooth maps.
516 Hale
Consider the equation
t2 u + (x)u u = f (u), x , (16.1)

u = 0, x ,
[0, ) is
where Rn is bounded, Lipschitzian and f is C 1 , :
2
C .
This equation can be written as a system
t U = CU + F(U ), (16.2)

u 0 I 0
U= , C= , F(U ) = . (16.3)
v D f (u)
Let X = H01 () L2 ()) and let eCt be the semigroup on X generated
by C. A mild solution U (t0 , U0 ) of (15.3) on an interval [0, ) with initial
value U0 X at t = 0 is a solution of
t
U (t) = eCt U0 + eC(ts) F(U (s))ds. (16.4)
0
Let T (t)U0 U (t, U0 ) be the local dynamical system on X. If solu-

tions are dened for all t 0, it is a dynamical system. From (16.4), we
obtain the following result.
Theorem 16.1. If
(i) F: H01 () L2 () is compact
(ii) k > 0, > 0 such that
eCt L(X,X) ket , t 0,
then T (t) is asymptotically smooth.
If we suppose that f satises the growth condition
|f (y1 ) f (y2 )|C1 (1 + |y1 | + |y2 | )|y1 y2 |, (16.5)

(n 2) < 2
then f : H01 () L2 () is compact.

then
If > 0 is a constant or even just bounded away from zero on ,
one can prove that the exponential bound on eCt is satised and T (t) is
asymptotically smooth.
We have noted in an earlier section that, if the support of is
contained in , then the exponential bound on eCt may not be satised.
We also remarked that the exponential estimate on eCt holds if . If

n = 1 and there is an x0 such that (x0 ) > 0, the same estimate also
holds and T (t) is asymptotically smooth.
To prove the existence of the compact global attractor, we rst show
that the dissipative condition
f (y)
lim sup < 1 , (16.6)
|y| y
where 1 is the rst eigenvalue of the Laplacian D with homogeneous

Dirichlet boundary conditions, implies that + (B) is bounded if B X is
bounded.
Dene the energy function
(x)
1 2 1 2
V (, ) = |(x)| + |(x)| + f (u)du dx. (16.7)
2 2 0
There is a dense set of initial data for which

d
V (u(t), v(t)) = (x)|t u(t, x)|2 dx0 (16.8)
dt
As a consequence of (16.8), each solution u(t) of (16.1) must satisfy the

inequality
t
V (u(t), t u(t)) = V (u(0), t u(0)) (x)|t u(t, x)|2 dx dt. (16.9)
0
Therefore, V is non increasing along solutions of the equation. The dis-

sipative condition (16.6) implies that V (, ) as |(, )| and,
therefore, + (B) is bounded if B is bounded.
To prove the existence of the compact global attractor for (16.1), we
need to show that T is point dissipative. The fact that the equilibrium set
E(T ) of T is bounded is proved as in the previous section for the para-
bolic equation.
If we could conclude from (16.9) that each bounded solution has
-limit set in E(T ), then we would have point dissipativeness and the
existence of the compact global attractor.
We remark rst that any solution v(t) of (16.1) which lies in the -
limit set of a solution of (16.1) is dened for all t R and V (v(t), t v(t)) =
V (v(0), t v(0)) for all t R.
If > 0 is a constant or if is a function of x which is bounded away
from zero, then (16.9) implies that any solution v(t) of (16.1) which lies in
the -limit set of a solution of (16.1) must have t v = 0; that is, v(t) is an
equilibrium and we have
518 Hale
Theorem 16.2. If > 0 or is a function of x which is positive on ,

then there is the compact global attractor for (16.1), (16.6), and (16.9).
The above proof for constant is due to Hale (1985). Haraux (1985)
gave a different proof but it was necessary to use some regularity results
and thus he had to assume more smoothness on the boundary.
One can also prove Theorem 16.2 by showing that there exists an
absorbing set in X. To do this, one makes use of a skewed energy func-
tion which is the same as the one in (16.7) except with some terms involv-
ing quadratic terms . An appropriate choice of these terms will yield an
absorbing set (see, e.g., Hale (1988), Temam (1997)).
Following Hale and Raugel (1993) and the same type of argument as
for the case when is a constant, if depends upon x, then, for all t R,
any solution v(t) of (16.1) which lies in the -limit set of a solution of
(16.1) must satisfy the equations
vtt v = f (v), in \
vt = 0, in (16.10)
v = 0, in ,
where is the support of .

It remains to show that any bounded complete orbit of (16.10) is an
equilibrium.
Let us now distinguish the cases n = 1 and n 2.
If n = 1, then f (v(t))vt belongs to Cb0 (R, L2 ()) and, using some reg-
ularity results in Hale and Scheurle (1985), we deduce that the function
V = (v, t v) satises Vt Cb0 (R, X). Hence, v Cb0 (R, H 2 () H01 ()). Since
vxx = f (v) + G (x)vt vtt .
Thus, f (v(t)) belongs to Cb2 (R, L2 ()), and, using again Hale and
Scheurle (1985), we conclude that (v, vt ) is in Cb2 (R, H01 () L2 ())
Cb1 (R, (H 2 () H01 ()) H01 ()). If w = vt , from (16.10), we deduce that
w is a globally dened bounded solution of the system
w Cb1 (R; L2 ()) Cb0 (R; H01 ()),

wtt w f (v(t, x))w = 0, in R , (16.11)
w = wt = 0, in R ,
If the only solution of (16.11) is w = 0, then the -limit set of any

solution of (16.1), (16.6), (16.9) belongs to the set E of equilibrium points.
It is shown in Hale and Raugel (1993) that the only solution of (16.11) is
w = 0. Thus, the compact global attractor exists. We state this as
Theorem 16.3. If n = 1 (i.e. is a bounded interval) and there is

an x0 such that (x0 ) > 0, then (16.1), (16.5), (16.6) has a compact global
attractor.
If n 2, Hale and Raugel (1993) show that f (v(t)) belongs to

Cb0 (R,Ln ()) and that w vt is a globally dened bounded weak solution
of the system
w Cb1 (R; H 1 ()) Cb0 (R; L2 ()),

wtt w f (v(t, x))w = 0, in R, (16.12)
w = wt = 0, in R ,
If the only solution of (16.12) is w = 0, then the -limit set of any solu-
tion of (16.1) belongs to the set E of equilibrium points. This leads us to
the unique continuation property (u.c.p.)
there is a time T > 0 such that if w is a weak L2 ( (0, T )) solution
of (16.13)
wtt w + b(x, t)w = 0, in (0, T ),

w = 0, in (0, T ),
b L ((0, T ); Ln ()),
then w 0 .
Ruiz (1992) has shown that the condition u.c.p. (16.13) holds if .
We, therefore, obtain the following result of Hale and Raugel (1993).
Theorem 16.4. If the support of contains the boundary of , then
there exists the compact global attractor for the system (16.1), (16.6), and
(16.9).
If R3 and = 2 in (16.5), then f : H01 () L2 () is a bounded
map but not compact. Therefore, the above proof of the existence of the
compact global will not work. When > 0 is a constant, Arrieta, et al.
(1992)) proved the existence of the compact global attractor under the
dissipative hypothesis (16.6).
For other results on the existence of the compact global attractor of
(16.1) (see Raugel (2002).
For Eq. (16.1) and > 0 constant, one can prove that, for any solu-
tion u(t) A(T ), t R is as smooth in t as the function f . Using this fact,
one can obtain spatial regularity of solutions on the attractor. See Hale
and Raugel (2003) for much more general results on regularity for the case
where is not constant, there is critical growth rate ((n 2) = 2), and
other equations.
520 Hale
REFERENCES
1. Andronov, A., and Pontrjagin, L.S. (1937). Systemes grossiers. Dokl. Akad. Nauk SSSR
14, 247251.
2. Angenent, S. (1986). The MorseSmale property for a semilinear parabolic equation.
J. Diff. Eq. 62, 427442.
3. Arrieta, J., Carvalho, A.N., and Hale, J.K. (1992). A damped hyperbolic equation with a
critical exponent. Comm. PDE 17, 841866.
4. Babin, A.V., and Vishik, M.I. (1992). Attractors of Evolution Equations, North-Holland,
Amsterdam.
5. Ball, J.M. (1997). Continuity properties of global attractors of generalized semiows and
the Navier-Stokes equations. J. Nonlinear Sci. 7, 475502.
6. Bardos, C. Lebeau, G., and Rauch, J. (1992). Sharp sufcient conditions for the observa-
tion, control and stabilization of waves from the boundary. SIAM J. Control Optimization
30(5), 10241065.
7. Billotti, J.E., and LaSalle, J.P. (1971). Periodic dissipative processes. Bull. Amer. Math. Soc.
(N.S.) 6, 10821089.
8. Browder, F.E. (1959). On a generalization of the Schauder xed point theorem. Duke
Math. J. 26, 291303.
9. Brunovsky, P., and Polacik, P. (1997). The Morse-Smale structure of a generic reaction
diffusion equation in higher space dimension. J. Diff. Eq. 135, 129181.
10. Chafee, N., and Infante, E.F. (1974). A bifurcation problem for a nolinear parabolic equa-
tion. J. Appl. Anal. 4, 1737.
11. Cholewa, J.W., and Hale, J.K. (2000). Some counterexamples in dissipative systems.
Dynamics of Continuous. Discrete Impulsive Syst. 7, 159176.
12. Cruz, M.A., and Hale, J.K. (1970). Stability of functional differential equations of neutral
type. J. Diff. Eq. 7, 334355.
13. Cooperman, G. (1978). -condensing maps and dissipative processes. Ph.D. Thesis, Brown
University, Providence RI.
14. Dafermos, C. (1978). Asymptotic behavior of solutions of evolutionary equations. In Non-
linear Evolution Equations, M.G. Crandall (ed.), 103123.
15. Faria, T., and Magalhaes, L. (1995a). Realization of ordinary differential equations by
retarded functional differential equations in neighborhoods of equilibrium points. Proc.
Roy. Soc. Edinburgh 125A, 759776.
16. Faria, T., and Magalhaes, L. (1995b). Normal forms for retarded functional differential
equations and applications to Bogdanov-Takens singularity. J. Diff. Eq. 122, 201224.
17. Faria, T., and Magalhaes, L. (1995c). Normal forms for retarded functional differential
equations with parameters and applications to Hopf bifurcations. J. Diff. Eq. 122 181
200.
18. Fiedler, B., and Rocha, C. (1996). Heteroclinic orbits of semilinear parabolic equations.
J. Diff. Eq. 156, 239281.
19. Fiedler, B., and Rocha, C. (2000). Orbit equivalence of global attractors of semilinear par-
abolic equations. Trans. Am. Math. Soc. 352, 257284.
20. Fusco, G., and Rocha, C. (1991). A permutation related to the dynamics of a scalar para-
bolic PDE. J. Diff. Eq. 91, 111137.
21. Gerstein, V.M. (1970). On the theory of dissipative differential equations in a Banach
space. Funk. Anal. I Prilzen 4, 99100.
22. Gerstein, V.M., and Krasnoselskii, M.A. (1968). Structure of the set of solutions of dissi-
pative equations. Dokl. Akad. Nauk. SSSR 183, 267269.
23. Gobbino, M., and Sardella, M. (1997). On the connectedness of attractors for dynamical
systems. J. Diff. Eq. 133, 114.
24. Hale, J.K. (1965). Sufcient conditions for stability and instability of autonomous func-
tional differential equations. J. Diff. Eq. 1, 452482.
25. Hale, J.K. (1985). Asymptotic behavior and dynamics in innite dimensions. In Res. Notes
in Math., Hale and Martinez-Amores (eds.), Pitman, London, Vol. 132, pp. 141.
26. Hale, J.K. (1985a). Flows on center manifolds for scalar functional differential equations.
Proc. Royal Soc. Edinburgh 101A, 193201.
27. Hale, J.K. (1988). Asymptotic Behavior of Dissipative Systems. American Mathematical
Society.
28. Hale, J.K., LaSalle, J.P., and Slemrod, M. (1972). Theory of a general class of dissipative
processes. J. Math. Ana. Appl. 39, 171191.
29. Hale, J.K., and Lopes, O. (1973). Fixed point theorems and dissipative processes. J. Diff.
Eq. 13, 391402.
30. Hale, J.K., Magalhaes, L., and Oliva, W.M. (2002). Dynamics in Innite Dimensions. Appl.
Math. Sci. Vol. 47. Second edition. Springer-Verlag, Berlin.
31. Hale, J.K., and Meyer, K.R. (1967). A class of functional equations of neutral type. Mem.
Amer. Math. Soc. 76, 165.
32. Hale, J.K., and Raugel, G. (1992). Convergence in gradient like systems. ZAMP 43, 63
124.
33. Hale, J.K., and Raugel, G. (1993). Attractors for dissipative evolutionary equations. In
Equadiff 91, International Conference on Differential Equations, Barcelona (1991), World
Scientic, Singapore, pp. 322.
34. Hale, J.K., and Raugel, G. (2003). Regularity, determining modes and Galerkin methods.
J. Math. Pures. Appl. 82(9), 10751136.
35. Hale, J.K., and Raugel, G. (2006). Innite Dimensional Dynamical Systems. In prepara-
tion.
36. Hale, J.K., and Scheurle, J. (1985). Smoothness of bounded solutions of nonlinear evolu-
tionary equations. J. Diff. Eq. 56, 142163.
37. Hale, J.K., and Verduyn-Lunel, S. (1993). Introduction to functional differential equa-
tions. Appl. Math. Sci. Vol. 99. Springer-Verlag, Berlin.
38. Haraux, A. (1985). Two remarks on hyperbolic dissipative problems. Nonlinear Partial
Differential Equations and their Applications. College de France Seminar, Vol. VII (Paris
1983-1985), Research Notes in Math. Pitman, London, Vol 122, pp. 161179.
39. Henry, D. (1981). Geometric Theory of Semilinear Parabolic Equations. Springer, Berlin.
40. Henry, D. (1985). Some innite dimensional MorseSmale systems dened by parabolic
differential equations. J. Diff. Eq. 59, 165205.
41. Henry, D. (1987). Topics in analysis. Pub. Mat. UAB 31, 2984.
42. Henry, D. (2005). Perturbation of the Boundary in Boundary Value Problems in Partial
Differential Equations. London Math. Soc. Lect. Notes Series Vol. 318. Cambridge Uni-
versity, Press, Cambridge.
43. Hirsch, M.W. (1988). Stability and convergence in strongly monotone dynamical systems.
J. Reine Angew. Math. 383, 153.
44. Iwasaki, N. (1969). Local decay of solutions for symmetric hyperbolic systems and coer-
cive boundary conditions in exterior domains. Pub. Res. Inst. Math. Sci. Kyoto Univ. 5,
193218.
45. Jones, G.S., and Yorke, J. (1969). The existence and nonexistence of critical points in
bounded ows. J. Diff. Eq. 6, 238247.
46. LaSalle, J.P. (1976). The stability of dynamical systems. CBMS Regional Conf. Ser.
SIAM 25.
522 Hale
47. Ladyzenskaya, O.A. (1987). On the determination of minimal global attractors for the
Navier Stokes equation and other differential operators. Russian Math. Surveys 42, 2773.
48. Levin, J.J., and Nohel, J.A. (1964). On a nonlinear delay equation. J. Math. Anal. Appl. 8,
3144.
49. Levinson, N. (1944). Transformation theory of nonlinear differential equations of the sec-
ond order. Ann. Math. 45(2), 724737.
50. Mallet-Paret, J. (1977). Generic periodic solutions of functional differential equations.
J. Diff. Eq. 25, 163183.
51. Massatt, P. (1980). Some properties of -condensing maps. Ann. Mat. Pura Appl. 125(4),
101115.
52. Massatt, P. (1983) Attractivity properties of -contractions. J. Diff. Eq. 48, 326333.
53. Massera, J.L. (1950). The existence of periodic solutions of systems of differential equa-
tions. Duke Math. J. 17, 457475.
54. Matano, H. (1978). Convergence of solutions of one-dimensional semilinear parabolic
equations. J. Math. Kyoto Univ. 18, 224243.
55. Nussbaum, R. (1972). Some asymptotic xed point theorems. Trans. Am. Math. Soc. 171,
349375.
56. Oliva, W.M. (1969). Functional differential equations on compact manifolds and an
approximation theorem. J. Diff. Eq. 5, 483496.
57. Oliva, W.M. (1982) Stability of Morse-Smale maps. D.Mat. Applicada-IME-Un. Sao
Paulo-Brasil 149.
58. Palis, J., and W. de Melo (1982). Geometric Theory of Dynamical Systems. Springer-Verlag,
New York.
59. Peixoto, M.M. (1959). On structural stability. Ann. Math. 69, 189222.
60. Pliss, V. (1966). Nonlocal Problems in the Theory of Oscillations. Academic Press, New
York.
61. Polacik, P. (1991). Complicated dynamics in scalar semilinear parabolic equations in
higher space dimensions. J. Diff. Eq. 89, 244271.
62. Polacik, P. (1999). Persistent stable connections in a class of reaction-diffusion equations.
J. Diff. Eq. 156, 182210.
63. Polacik, P. (2002). Parabolic equations: asymptotic behavior and dynamics on invari-
ant manifolds. Handbook of Dynamical Systems. North-Holland, Amsterdam, Vol. 2, pp.
835883.
64. Polacik, P., and Rybakowski, K. (1995). Embedding vector elds into Dirichlet BVPs.
Ann. Scuola Norm. Sup. Pisa 21, 737749.
65. Polacik, P., and Rybakowski, K. (1996). Nonconvergent bounded trajectories in semilinear
heat equations. J. Diff. Eq. 124, 472494.
66. Prizzi, M. (1998). Realizing vector elds without loss of derivatives. Ann. Scuola Norm.
Sup. Pisa 27, 289307.
67. Raugel, G. (1995). Dynamics of partial differential equations on thin domains. In Dynam-
ical Systems, R. Johnson (ed.) LNM 1609, Springer, Berlin, pp. 208315.
68. Raugel, G. (2002). Global attractors in partial differential equations. In Handbook of
Dynamical Systems. North-Holland, Amsterdam, Vol. 2, pp. 885982.
69. Ruiz, A. (1992). Unique continuation for weak solutions of the wave equation plus a
potential. J. Math. Pures Appl. 71, 455467.
70. Rybakowski, K.P. (1994). Realization of arbitrary vector elds on invariant manifolds of
delay equations. J. Diff. Eq. 114, 222231.
71. Sell, G.R., and You, Y. (2002). Dynamics of Evolutionary Equations. Springer-Verlag,
New York.
72. Silnikov, L.P. (1968). On the generation of periodic motion from trajectories doubly
asymptotic to an equilibrium of saddle type. Math. Sb. 6, 428438.
73. Silnikov, L.P. (1970). A contribution to the problem of the structure of an extended
neighborhood of a rough equilibrium state of a saddle-focus type. Math. Sb. 10, 91102.
74. Simon, L. (1983). Asymptotics for a class of nonlinear evolution equations. Ann. of Math.
118(2), 525571.
75. Smale, S. (1967). Differentiable dynamic systems. Bull. Amer. Math. Soc. 73, 747817.
76. Temam, R. (1997). Innite Dimensional Dynamical Systems in Mechanics and Physics, Sec-
ond Edition. Springer-Verlag, Berlin.
77. Zelenyak, T.J. (1968). Stabilization of solutions of boundary value problems for a second
order parabolic problem with one space variable. Diff. Eq. 4, 1722.

Dissipation and Compact Attractors

Uploaded by

Copyright:

Available Formats

Dissipation and Compact Attractors

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dissipation and Compact Attractors

Uploaded by

Copyright:

Available Formats

Journal of Dynamics and Differential Equations, Vol. 18, No.

3, July 2006 ( 2006)

Dissipation and Compact Attractors

Received December 27, 2005

We present an approach to the study of the qualitative theory of innite

School of Mathematics, Georgia Institute of Technology, Atlanta, GA 30332-0160, USA.

on smooth 2-manifolds. The work of Smale (1967) was a great stimulus

2. MAXIMAL COMPACT INVARIANT SETS AND COMPACT

A set J in X is said to be invariant if T (t)J = J , t G+ . If J is invariant,

A = A(T ) = { bounded global orbits of T }.

3. DEFINITIONS OF STABILITY AND BIFURCATION

periodic orbits of a continuous T . The dynamical system is MorseSmale

4. DISSIPATIVE DYNAMICAL SYSTEMS IN Rn

y = f (t, y), (4.1)

where f : R Rn Rn satises enough conditions to imply that, for any

Suppose that there is an > 0 such that f (t + , y) = f (t, y) for all

Remark 4.7. In R2 , as a consequence of the PoincareBendixon the-

5. COMPACT DYNAMICAL SYSTEMS

Theorem 5.1. If the dynamical system T is point dissipative and there

Motivated by applications to NFDE, the following asymptotic xed

Denition 5.2. If B is a bounded set in a Banach space X, the

(B) = inf {d : B has a nite cover of diameter < d}.

The term in the denition is used because (B) = 0 if and only if Cl B

Denition 5.3. If X is a Banach space, then a continuous map T :

Denition 5.4. A dynamical system T on X is compact dissipative if

Theorem 5.5. If X is a Banach space and T : X X is a continuous

If T is a dynamical system on X for which there is an r > 0 such

Theorem 5.6. If the discrete dynamical system on X dened by a com-

This result has direct applications to certain types of evolutionary

t u = Au + f (t, u), (5.1)

Suppose that the solution u(t, , u0 ) through u0 at is unique and dened

a xed point of T (, 0) if f has minimal period > 0 in t and an equi-

x(t) = f (t, xt ). (5.2)

For given R, a function x(t, , ), t r is a solution of (5.2)

6. LINEAR POINT DISSIPATIVE SYSTEMS

Theorem 6.1. If X is a Banach space, T (t): X X, t G+ , is a linear

Example 6.2. (A Parabolic Equation) Consider the linear parabolic

with u = 0 on the boundary  of the smooth bounded domain . The

Example 6.3. (A simple map on a Hilbert space) Let X be a real,

It is easy to see that T n x 0 as n for every x X. If we assume that

Example 6.4. If the linear dynamical system T (t), t 0, is generated

Example 6.5. (A Neutral DifferentialDifference Equation) Suppose

Equation (6.2) is a RFDE if a = 0 and is called a neutral differential

with Dirichlet boundary conditions u = 0 on the boundary  of the

Remark 6.7. From Theorem 6.1, we see that point dissipative is

Theorem 6.8. If T : X X is an asymptotically smooth map which is

It certainly would be interesting to characterize those nonlinear

7. COMPACT SETS ATTRACTING COMPACT SETS

Denition 7.1. Let T (t), t G+ , be a dynamical system on X. A

Denition 7.2. An invariant set J of the dynamical system T is said

Theorem 7.3. If T (t): X X, t G+ , is a dynamical system on X

Remark 7.4. It might be expected that the maximal compact invari-

8. THE MAXIMAL COMPACT INVARIANT SET

Denition 8.1. A dynamical system T (t), t G+ , on X is said to be

It is important to remark that the concept of asymptotically smooth

T (t) = S(t) + U (t), (8.1a)

where U (t) is compact for t G+ . Such dynamical systems are asymptot-

|S(t)x| ket |x|, x X, t G+ . (8.1b)

The illustrations given below will be of this type. However, asymptot-

Proposition 8.2. The dynamical system T (t), t G+ , on X is asymp-

with u = 0 on the boundary of the smooth bounded domain . The

with Dirichlet boundary conditions u = 0 on the boundary of the

TD,0 (t)L(CD ,CD ) ket , t 0.

TD,f (t) = TD,0 (t)(I D) + U (t), (11.5)

det () = 0, () = L(e ). (12.3)

CM(f ) = { C : = y + h(f, y)}, (12.6)

Even though the system is gradient, for a ball in R2 , there is a

t2 u + (x)u u = f (u), x , (16.1)

eCt L(X,X) ket , t 0,