Professional Documents
Culture Documents
2-Luís Barreira, Claudia Valls - Dynamical Systems by Example-Springer (2019)
2-Luís Barreira, Claudia Valls - Dynamical Systems by Example-Springer (2019)
Luís Barreira
Claudia Valls
Dynamical
Systems by
Example
Problem Books in Mathematics
Series Editor
Peter Winkler
Department of Mathematics
Dartmouth College
Hanover, NH
USA
More information about this series at http://www.springer.com/series/714
Luís Barreira Claudia Valls
•
Dynamical Systems
by Example
123
Luís Barreira Claudia Valls
Instituto Superior Técnico Instituto Superior Técnico
Universidade de Lisboa Universidade de Lisboa
Lisbon, Portugal Lisbon, Portugal
This Springer imprint is published by the registered company Springer Nature Switzerland AG.
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This book is a large collection of problems, all with detailed solutions, on the core
of the theory of dynamical systems. Besides the basic theory, the topics include
topological dynamics, low-dimensional dynamics, hyperbolic dynamics, symbolic
dynamics, and basic ergodic theory.
As in any other area of mathematics, it is important while learning dynamical
systems to solve selected problems (always after a careful study of the material!), in
particular to get a first working knowledge of the topics as well as of some less
direct difficulties. It goes without saying that this helped substantially generations of
mathematicians in getting a solid understanding of the theory and of its applica-
tions. Nevertheless, it is difficult to find large collections of problems on less basic
subjects, at least other than on quite specific topics. Moreover, these problems often
lack detailed solutions or even any solutions, while it would certainly be quite
helpful for a student to have the possibility to study detailed solutions, especially if
studying independently.
Certainly, it is not a good practice to study detailed solutions of exercises
without first trying hard to solve them. Indeed, solving exercises is an important
step toward learning a particular subject and ultimately toward becoming an
independent mathematician, while perhaps also developing a personal style. On the
other hand, it would be quite welcome to have available comprehensive sources of
problems worked out in detail. In more advanced topics, these solutions can have
the role of providing detailed arguments for comparison or even alternative views,
possibly alerting to more direct approaches or to less obvious connections to other
topics (and mathematics is full of such connections).
Our text is a contribution to fill this gap on selected topics of the theory of
dynamical systems (as detailed below). It can be used as a companion to a textbook
for a one-semester or two-semester course on dynamical systems at the advanced
undergraduate or beginning graduate levels, or for independent study of those
topics. Other than some basic pre-requisites from linear algebra, differential and
integral calculus, complex analysis and topology, in each chapter we recall all
notions and results (without proofs) that are necessary for the problems in that
chapter, thus making the text self-contained.
v
vi Preface
The theory of dynamical systems is quite broad and active in terms of research.
Hence, it was necessary to make a careful selection of the material. In this aspect,
we followed closely our book [11], which gives an introduction to the theory of
dynamical systems and to which the present book can be an excellent companion.
The levels of the two texts are analogous, although the present book provides a
more comprehensive working knowledge of the topics as well as a much larger
variety of problems, also of various levels of difficulty (from simple to quite
elaborate) and with a carefully planned interdependence when appropriate, so that
the material is learned in successive steps.
We detail briefly the topics addressed in the book together with recommenda-
tions for further reading:
• Chapter I.1 considers the notion of a dynamical system, both for discrete and
continuous time, invariant sets, orbits, periodic points, rotations and expanding
maps of the circle, endomorphisms and automorphisms of the torus, autono-
mous differential equations and their flows, Poincaré sections, and Poincaré
maps (see [3, 10, 20, 22]).
• Chapter I.2 studies continuous maps of a topological space and their topological
properties, including the notions of a-limit set and of x-limit set, recurrent
points, nonwandering points, minimal sets, topological transitivity, topological
mixing, and topological conjugacies, as well as topological entropy and its
properties (see [16, 17, 48]).
• Chapter I.3 considers dynamical systems on low-dimensional spaces, including
homeomorphisms of the circle, their lifts, orientation-preserving homeomor-
phisms and the notion of rotation number, continuous maps on the interval, and
the Poincaré–Bendixson theory on the plane (see [2, 10, 22, 24]).
• Chapter I.4 studies hyperbolic dynamics, including the notion of a hyperbolic
set, the Smale horseshoe, invariant families of cones, topological conjugacies,
and invariant manifolds near a hyperbolic fixed point, as well as geodesic flows
and their hyperbolicity (see [28, 40, 41, 50, 54]).
• Chapter I.5 considers selected topics of symbolic dynamics, including one-sided
and two-sided shift maps, topological Markov chains, irreducible and transitive
matrices, topological transitivity, topological mixing, topological entropy, and
zeta functions (see [16, 33, 34]).
• Chapter I.6 studies selected basic topics of ergodic theory, including Poincaré’s
recurrence theorem, Birkhoff’s ergodic theorem, and the notion of metric
entropy (see [8, 23, 35, 46, 56]).
The book is conveniently separated into two parts. Part I (Chaps. I.1–I.6) recalls
briefly in each chapter various notions and results that are needed for the problems
of that chapter, also with the purpose of fixing notation, after which the problems
are formulated without their solutions. In Part II (Chaps. II.1–II.6) the problems are
restated but now with their solutions. In this way each problem can be solved
without the risk of looking inadvertently at the solution. There are 240 problems
ranging from simple to quite elaborate (with 40 per chapter), all with detailed
Preface vii
ix
Part I
Theory and Problems
Chapter I.1
Basic Theory
In this chapter we consider the notion of a dynamical system, both for discrete and
continuous time. In particular, we consider invariant sets, orbits, semiorbits, peri-
odic points, rotations and expanding maps of the circle, endomorphisms and auto-
morphisms of the torus, as well as autonomous ordinary differential equations and
their flows. We also consider some basic constructions that determine new dynami-
cal systems, including suspension semiflows, Poincaré sections, and Poincaré maps.
We refer the reader to [3, 10, 20, 22] for additional topics.
First we recall a few basic notions, including those of invariant set, orbit, and periodic
point.
Definition 1.1 A map f : X Ñ X is called a dynamical system with discrete time.
We define recursively
f n “ f ˝ f n´1
for each n P N, with the convention that f 0 “ id. When f is invertible, we also define
f ´n “ p f ´1 qn for each n P N.
Given a map f : X Ñ X and a set A Ď X, we write
(
f ´1 A “ x P X : f pxq P A .
and (
γ pxq “ f n pxq : n P Z
are called, respectively, the negative semiorbit of x and the orbit of x.
Definition 1.4 Given a map f : X Ñ X and an integer q P N, a point x P X satisfying
f q pxq “ x
x„y ðñ x ´ y P Z.
The circle S1 is defined by S1 “ R{„. The elements of S1 are thus the equivalence
classes
rxs “ tx ` m : m P Zu for x P R.
Given α P R, the rotation Rα : S1 Ñ S1 is defined by
Rα rxs “ rx ` α s.
Em rxs “ rmxs.
x„y ðñ x ´ y P Zn .
The n-torus or simply the torus is defined by Tn “ Rn {„. The elements of Tn are
thus the equivalence classes
rxs “ tx ` y : y P Zn u for x P Rn .
I.1 Basic Theory 5
ϕt`s “ ϕt ˝ ϕs for t, s P R
ϕt´1 A “ A for t ě 0.
ϕt´1 A “ A for t P R.
and
γ pxq “ tϕt pxq : t P Ru
are called, respectively, the negative semiorbit of x and the orbit of x.
In particular, for the flow in Proposition 1.8 associated with an autonomous dif-
ferential equation v1 “ f pvq, if v0 is a critical point of the equation, then γ pv0 q “ tv0 u
is the orbit of v0 .
Poincaré sections give relations between the notions of a dynamical system for
discrete and continuous time.
Definition 1.11 A set X Ĺ Y is called a Poincaré section for a semiflow of maps
ϕt : Y Ñ Y , for t ě 0, if
(
τ pxq :“ inf t ą 0 : ϕt pxq P X P R`
for each x P X, with the convention that inf ∅ “ `8 (see Figure I.1.1). The number
τ pxq is called the first return time of x to X.
Given a Poincaré section X for a semiflow pϕt qtě0 , its Poincaré map f : X Ñ X
is defined by
f pxq “ ϕτ pxq pxq for x P X.
In particular, given a Poincaré section, one can construct a dynamical system with
discrete time (the Poincaré map) from the semiflow. One can often study properties
of the semiflow by studying properties of the Poincaré map and vice versa.
τ x
x f x
Problems
Problem 1.1 Determine all values of a P R for which the map f : R Ñ R defined by
f pxq “ ax4 ´ x has nonzero fixed points.
Problem 1.2 Determine all the periodic points of the map f : R Ñ R defined by
f pxq “ ex .
Problem 1.3 Show that if a continuous map f : R Ñ R has a periodic point with
period 2, then it has at least one fixed point.
Problem 1.4 Show that a continuous map f : ra, bs Ñ R with f pra, bsq Ě ra, bs has
at least one fixed point.
Problem 1.5 Show that a continuous map f : ra, bs Ñ ra, bs has at least one fixed
point.
Problem 1.6 Consider the continuous map f : r1, 5s Ñ r1, 5s with
such that f is linear on rn, n ` 1s for n “ 1, 2, 3, 4 (see Figure I.1.2). Show that:
1. f has periodic points with period 5;
2. f 3 has no fixed points in r1, 3s Y r4, 5s, but has a fixed point in r3, 4s;
3. f has no periodic points with period 3.
1 2 3 4 5
R “ tz P C : |z| “ 1u.
E2 x
1 2 1 x
E22 x
1 4 1 2 3 4 1 x
using the Smith normal form (which says that A “ PDQ, for some matrices P, D
and Q with integer entries such that |det P| “ |det Q| “ 1 and D is diagonal).
Problem 1.25 Let TA : Tn Ñ Tn be an automorphism of the torus induced by a
matrix A whose spectrum contains no root of 1. Show that the number of q-periodic
points of TA is equal to |detpAq ´ Idq|.
Problem 1.26 Let TA : T2 Ñ T2 be the automorphism of the torus induced by the
matrix ˆ ˙
21
A“ .
11
Show that the number of q-periodic points of TA is equal to trpAq q ´ 2.
Problem 1.27 Let f : S1 Ñ S1 be a C1 map with nonvanishing derivative. Show that
there exists q P N such that
Show that:
1. if U is an open neighborhood of Λ , then f ´n A Ď U for any sufficiently large n;
2. the set Λ is f -invariant.
I.1 Basic Theory 11
E3 x
1 x
Fig. I.1.5 The set marked in gray is r0, 1{3s Y r2{3, 1s.
Problem 1.30 Find the largest E3 -invariant set A contained in J “ r0, 1{3s Y r2{3, 1s
(see Figure I.1.5).
Problem 1.31 Consider the differential equation
#
x1 “ 6y5 ,
y1 “ ´4x3
on R2 has unique solutions that are defined for all t P R. Let ϕt : T2 Ñ T2 be the
corresponding flow (see Proposition 1.8). Each solution
of the equation with pxp0q, yp0qq “ p0, zq crosses infinitely often the line x “ 0 (since
f is uniformly bounded from below). The first intersection into the future occurs at
the time (
Tz “ inf t ą 0 : xptq “ 1 .
Show that the map h : S1 Ñ S1 defined by
hpzq “ ypTz , zq
1, h z
0, z ϕt 0, z
1 x
ÿ
n
B ϕ pvq
pLϕ qpvq “ fi pvq
i“1
Bvi
for each C2 function ϕ : Rn Ñ R. Show that each periodic orbit of the equation:
1. has at least two points in the set
(
Aϕ “ v P Rn : pLϕ qpvq “ 0 ;
In this chapter we consider the class of topological dynamical systems, that is, the
class of continuous maps on a topological space. In particular, we consider the no-
tions of α -limit set and of ω -limit set, as well as various notions related to topo-
logical recurrence, including those of recurrent point, nonwandering point, and min-
imal set. We also consider the notions of topological transitivity, topological mixing,
and topological conjugacy, as well as topological entropy. We refer the reader to
[16, 17, 48] for additional topics.
There are also corresponding versions of Propositions 2.4 and 2.5 for continuous
time. In particular, given a semiflow Φ on a topological space X, for each x P X we
have y P ω pxq if and only if there exists a sequence tk Õ `8 in R` such that
Moreover, given a flow Φ on X, for each x P X we have y P α pxq if and only if there
exists a sequence tk Œ ´8 in R´ such that
dn ppi , p j q ě ε whenever i ‰ j.
H ˝ f “ g˝H on X,
Problems
written in polar coordinates. Compute the α -limit set and the ω -limit set of each
point in R2 for the flow determined by the differential equation.
I.2 Topological Dynamics 19
Problem 2.8 Compute the α -limit set of each point px, yq P R2 with |x| ă 1 for the
flow determined by the differential equation
#
x1 “ px2 ´ 1qpy ´ xq,
y1 “ x.
for every x P X.
Problem 2.12 Let f : X Ñ X be a continuous map. Show that Rp f q (the set of re-
current points) is forward f -invariant.
Problem 2.13 Given a map f : X Ñ X, show that Rp f q Ď NW p f q.
Problem 2.14 A nonempty closed forward invariant set without nonempty closed
forward invariant proper subsets is said to be minimal. Show that any two distinct
minimal sets for a map f must have empty intersection.
Problem 2.15 Give an example of an ω -limit set that is a minimal set.
Problem 2.16 Given a map f : X Ñ X on a topological space and a nonempty closed
forward f -invariant set M Ď X, show that the following properties are equivalent:
1. M is a minimal set;
2. M “ γ ` pxq for all x P M;
3. M is the ω -limit set of each of its points.
Problem 2.17 For a continuous map f : X Ñ X show that f is topologically transi-
tive if and only if for any nonempty open sets U,V Ĺ X there exists n P N such that
f n pUq XV ‰ ∅.
Ť
Problem 2.18 Let f : X Ñ X be a continuous map. Ť8 Show that if 8 n“1 f pUq is
n
dense for any nonempty open set U Ď X, then n“1 f ´nU is also dense for any
nonempty open set U Ď X.
Problem 2.19 Given a continuous map f : X Ñ X, show that if f is topologically
transitive, then any closed forward f -invariant proper subset of X has empty interior.
20 I.2 Topological Dynamics
tRm
α pxq : m P Zu “ S
1
for every x P S1 .
Problem 2.26 Let f : X Ñ X and g : Y Ñ Y be topologically conjugate maps (see
Definition 2.11). Show that for each q P N the number of q-periodic points of f is
equal to the number of q-periodic points of g.
Problem 2.27 Consider continuous maps f , g : R Ñ R that are topologically conju-
gate via a homeomorphism h : R Ñ R satisfying h ˝ f “ g ˝ h. Show that if p is an
attracting fixed point of f (which means that there exists an open neighborhood U
of p such that for x P U we have f n pxq Ñ p when n Ñ 8) if and only if q “ hppq is
an attracting fixed point of g.
Problem 2.28 Show that if two maps are topologically conjugate and one of them
is topologically mixing, then the other is also topologically mixing.
Problem 2.29 Given a topologically mixing map f : X Ñ X, show that the map
f ˆ f : X ˆ X Ñ X ˆ X defined by
hd p f q “ hd 1 p f q
and denoting by NpUn q the smallest cardinality of the finite subcovers of Un , the
limit
1
lim log NpUn q
nÑ8 n
exists;
2. if U is a finite open cover of X with Lebesgue number δ (this is a number such
that any open ball of radius δ is contained in some element of U), then
We start by recalling some notions and results associated to the study of dynamical
systems on low-dimensional spaces.
Definition 3.1 Consider a homeomorphism f : S1 Ñ S1 and let π : R Ñ S1 be the
map defined by π pxq “ rxs. A continuous function F : R Ñ R is called a lift of f if
f ˝π “ π ˝F on R.
F n pxq ´ x
ρ pFq “ lim PR
nÑ8 n
exists and is independent of x. Moreover, if G is another lift of f , then
ρ pGq ´ ρ pFq P Z.
1 ă 2 ă 22 ă 23 ă ¨ ¨ ¨ ă 2m ă ¨ ¨ ¨
¨¨¨
ă ¨ ¨ ¨ ă 2m p2n ` 1q ă ¨ ¨ ¨ ă 2m 7 ă 2m 5 ă 2m 3 ă ¨ ¨ ¨
¨¨¨
ă ¨ ¨ ¨ ă 2p2n ` 1q ă ¨ ¨ ¨ ă 2 ¨ 7 ă 2 ¨ 5 ă 2 ¨ 3 ă ¨ ¨ ¨
ă ¨ ¨ ¨ ă 2n ` 1 ă ¨ ¨ ¨ ă 7 ă 5 ă 3.
I.3 Low-Dimensional Dynamics 25
Problems
Problem 3.1 Show that the composition Rα ˝ Rβ of two rotations of the circle
Rα , Rβ : S1 Ñ S1 is also a rotation of the circle.
Problem 3.2 Show that given n ˆ n matrices A and B with entries in Z, the composi-
tion of the endomorphisms of the torus TA , TB : Tn Ñ Tn is again an endomorphism
of the torus and determine a matrix C such that TA ˝ TB “ TC .
Problem 3.3 Consider the homeomorphism of the circle defined by
„ j
1 1
f prxsq “ x ` ` sinp2π xq
2 4π
(see Figure I.3.1). Show that t0, 1{2u is a periodic orbit of f and that ρ p f q “ 1{2.
1 2
0 1 2 1
Problem 3.4 Given a P p0, 1{p2π qq, consider the map f : S1 Ñ S1 defined by
(see Figure I.3.2). Show that each orbit starting in p0, 1{2q is monotonous.
1 2
0 1 2 1
1 2
0 1 2 1
Fig. I.3.3 Homeomorphism of the circle with the single periodic orbit t0, 1{2u.
0 1
2.
r{s ă ρ p f q ă p{q ă 1
for all x P R;
I.3 Low-Dimensional Dynamics 29
(see Figure I.3.8). Show that f has period points with all periods.
1 2 3
and let
V px, yq “ x2 ` y2 and V9 px, yq “ ∇V px, yq ¨ f px, yq.
Show that:
I.3 Low-Dimensional Dynamics 33
Bpϕ f q Bpϕ gq
`
Bx By
has the same sign almost everywhere on a simply connected open set U Ď R2 , then
the equation has no periodic orbits contained in U.
Problem 3.40 Show that the differential equation
#
x1 “ y,
y1 “ ´x ´ y ` x2 ` y2
In this chapter we consider various notions and results of hyperbolic dynamics. Be-
sides the notion of a hyperbolic set, we consider the Smale horseshoe and some of its
modifications, the characterization of a hyperbolic set in terms of invariant families
of cones, as well as the construction of topological conjugacies near a hyperbolic
fixed point. Moreover, we consider briefly geodesic flows on the upper half plane
and their hyperbolicity. We refer the reader to [28, 40, 41, 50, 54] for additional
topics.
We start by recalling a few notions and results that are used in the chapter.
Definition 4.1 Let f : M Ñ M be a C1 diffeomorphism. A compact f -invariant set
Λ Ď M is called a hyperbolic set for f if there exist λ P p0, 1q, c ą 0 and a splitting
Tx M “ E s pxq ‘ E u pxq
dx f n v ď cλ n v;
dx f ´n v ď cλ n v.
The spaces E s pxq and E u pxq are called, respectively, the stable and unstable spaces
at x.
© Springer Nature Switzerland AG 2019 35
L. Barreira and C. Valls, Dynamical Systems by Example, Problem Books
in Mathematics, https://doi.org/10.1007/978-3-030-15915-3 4
36 I.4 Hyperbolic Dynamics
dpv, Eq “ min v ´ w : w P E .
(
and
dpE u pxm q, E u pxqq Ñ 0 when m Ñ 8.
The following definition introduces the Smale horseshoe.
Definition 4.3 Consider the horizontal strips
The set č č
Λ“ f n pQq “ f n pH1 Y H2 q
nPZ nPZ
H2
H1 V1 V2
Tx M “ F s pxq ‘ F u pxq
and an inner product x¨, ¨y1 “ x¨, ¨y1x on Tx M such that the dimensions dim F s pxq and
dim F u pxq are independent of x. We denote by ¨1 the norm induced by the inner
product x¨, ¨y1 . Given γ P p0, 1q and x P Λ , we define cones by
and
Cu pxq “ pv, wq P F s pxq ‘ F u pxq : v1 ă γ w1 Y t0u.
(
h ˝ f “ dx f ˝ h on U
E s (x) x
E u (x)
and
V u pxq “ y P Bpx, ε q : f n pyq ´ f n pxq ă ε for n ă 0 .
(
I.4 Hyperbolic Dynamics 39
V s (x)
V u (x)
Theorem 4.8 (Stable and unstable manifolds) Let Λ be a hyperbolic set for a
C1 diffeomorphism f : R p Ñ R p . For any sufficiently small ε ą 0, the following
properties hold:
1. for each x P Λ the sets V s pxq and V u pxq are manifolds of class C1 satisfying
and
f ´n pyq ´ f ´n pxq ď Cρ n y ´ x for y P V u pxq,
for all x P Λ and n P N.
We also consider the notion of a hyperbolic set for a flow.
Definition 4.9 Let Φ be a C1 flow on a manifold M. A compact Φ -invariant set
Λ Ď M is called a hyperbolic set for Φ if there exist λ P p0, 1q, c ą 0 and a splitting
dx ϕt v ď cλ t v;
The spaces E s pxq and E u pxq are called, respectively, the stable and unstable spaces
at x.
i T
z v
i
H “ z P C : Im z ą 0
(
SH “ pz, vq P H ˆ C : |v|z “ 1 ,
(
with the norm |v|z given by |v|z “ |v|{ Im z. Given ps, vq P SH, one can show that
there exists a unique Möbius transformation
az ` b
T pzq “ with ad ´ bc “ 1
cz ` d
such that
I.4 Hyperbolic Dynamics 41
T pwq “ α w ` x, with α “ Im z;
Problems
Problem 4.1 Let Λ be a hyperbolic set with finitely many points. Show that Λ is
composed of periodic points.
Problem 4.2 Show that the second and third conditions in Definition 4.1 can be
replaced, respectively, by
1
dx f ´n v ě λ n v for x P Λ , v P E s pxq, n P N
c
and
1
dx f n v ě λ ´n v for x P Λ , v P E u pxq, n P N.
c
Problem 4.3 Show that if Λ is a hyperbolic set for a diffeomorphism f , then for all
sufficiently large k P N the set Λ is a hyperbolic set for f k with constant c “ 1.
Problem 4.4 Let Λ be a hyperbolic set for a diffeomorphism f : R p Ñ R p . Show
that Λ ˆ Λ is a hyperbolic set for the diffeomorphism g : R p ˆ R p Ñ R p ˆ R p de-
fined by
gpx, yq “ p f pxq, f ´1 pyqq.
42 I.4 Hyperbolic Dynamics
Problem 4.5 Let A be a p ˆ p matrix with real entries and without eigenvalues of
modulus 1. Show that there exist a splitting R p “ E s ‘ E u and constants λ P p0, 1q
and c ą 0 such that
and
A´n v ď cλ n v for v P E u , n P N.
Problem 4.6 For the Smale horseshoe Λ (see Definition 4.3), show that for any
two-sided sequence ω “ p¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q in t1, 2uZ the set
f ´nVin
č
Λω “
nPZ
ϕ1 ă ψ1 ă ϕ2 ă ψ2
and
ϕ1 ă ψ1 ă ϕ2 ă ψ2
(see Figure I.4.5). Moreover, let f be a C1 diffeomorphism on an open neighborhood
of the square r0, 1s2 such that f pHi q “ Vi for i “ 1, 2 and
and some C1 functions g, h defined on compact subsets of R. Show that if sup |g1 | ă 1
and inf |h1 | ą 1, then
I.4 Hyperbolic Dynamics 43
č
Λ“ f n pH1 Y H2 q
nPZ
H̄2
Problem 4.10 Show that if a compact connected manifold M is a hyperbolic set for
a diffeomorphism f : M Ñ M, then all stable spaces have the same dimension.
Problem 4.11 Let f : R Ñ R be a C1 map. Show that if there exist λ P p0, 1q and
c ą 0 such that |p f n q1 pxq| ď cλ n for all x P R and n P N, then
1
lim sup log | f n pIq| ă 0
nÑ8 n
Determine whether the set X of all bounded functions f : R Ñ R with the property
that f p0q ` f p1q “ 0 is a complete metric space with the distance d.
Problem 4.14 Show that the set of all C1 functions f : ra, bs Ñ R is not a complete
metric space with the distance
A “ S1 ˆ px, yq P R2 : x2 ` y2 ď 1
(
Problem 4.24 Show that the vector Xpxq “ dtd ϕt pxqˇt“0 in Definition 4.9 is neither
ˇ
for any p, q ą m.
Problem 4.27 Find the stable and unstable invariant manifolds at the origin for the
differential equation #
x1 “ x,
y1 “ ´y ` x2 .
Problem 4.28 Find the stable and unstable invariant manifolds at the origin for the
differential equation #
x1 “ xpa ´ yq,
y1 “ ´y ` x2 ´ 2y2 ,
for each given a ą 0.
Problem 4.29 Given a, b ą 0, show that the map h : R Ñ R defined by
$
&x
’ b{a if x ą 0,
hpxq “ 0 if x “ 0,
´|x| b{a if x ă 0.
’
%
h ˝ ϕt “ ψt ˝ h for t P R.
Problem 4.30 Let A and B be n ˆ n matrices. Show that if there exists a topological
conjugacy between the flows determined by the differential equations
v1 “ Av and v1 “ Bv
that is a diffeomorphism, then there exists also a topological conjugacy between the
flows that is a linear map.
Problem 4.31 Let A and B be n ˆ n matrices. Show that if there exists a topologi-
cal conjugacy between the flows determined by the differential equations in Prob-
lem 4.30 that is a linear map, then the matrices A and B are conjugate, that is, there
exists an invertible n ˆ n matrix C such that A “ C´1 BC.
46 I.4 Hyperbolic Dynamics
Problem 4.32 Let ϕt and ψt be the flows determined, respectively, by the linear
equations with matrices
´1 0 ´1 1
ˆ ˙ ˆ ˙
A“ and B “ .
0 ´1 0 ´1
Verify that the map h : R2 Ñ R2 defined by hp0, 0q “ 0 and
d
x2 ` y2 x2 ` y2
ˆ ˙
y
hpx, yq “ x ´ log , y
x2 ` xy ` 3y2 {2 2 2
x2
ˆ ˙
hpx, yq “ x, y `
3
is a topological conjugacy between the flows determined by the equation and its
linearization at the origin.
Problem 4.34 For the differential equation
#
x1 “ x,
y1 “ ´y ` xn ,
xn
ˆ ˙
hpx, yq “ x, y ´
n`1
is a topological conjugacy between the flows determined by the equation and its
linearization at the origin.
Problem 4.35 Consider the Möbius transformation
az ` b
TA pzq “
cz ` d
associated with the matrix
ˆ ˙
ab
A“ , with det A “ ad ´ bc “ 1,
cd
I.4 Hyperbolic Dynamics 47
and let
T1,c pzq “ T´ 1 c ¯ pzq “ z ` c for c P R,
01
T2,c pzq “ T´ c 0
¯ pzq “ c2 z for c P Rzt0u,
0 1{c
1
T3,c pzq “ T´ 0 ´1 ¯ pzq “ ´ for c P R.
1 c z`c
Show that #
T1,ab ˝ T2,a if c “ 0,
TA “
T1,a{c ˝ T2,1{c ˝ T3,d{c if c ‰ 0.
Problem 4.36 Show that any straight line or circle is determined by an equation of
the form
Azz ` Bz ` Bz `C “ 0 with A,C P R, B P C,
respectively, with A “ 0 and with A ‰ 0 and AC ă BB.
Problem 4.37 Show that the map T pzq “ 2z takes straight lines into straight lines
and circles into circles.
Problem 4.38 Show that the map T pzq “ ´1{pz ` 1q takes straight lines and circles
into straight lines or circles.
Problem 4.39 Show that the components of the geodesic flow ϕt : SH Ñ SH given
by
ϕt pz, vq “ pγ ptq, γ 1 ptqq
(see Definition 4.10), satisfy
2 Re γ 1 ptq Im γ 1 ptq
pRe γ 1 ptqq1 “
Im γ ptq
and
pIm γ 1 ptqq2 ´ pRe γ 1 ptqq2
pIm γ ptqq1 “ .
Im γ ptq
Problem 4.40 For the geodesic flow ϕt : SH Ñ SH, show that the quantities
|γ 1 ptq| Re γ 1 ptq
H1 “ and H2 “
Im γ ptq pIm γ ptqq2
are independent of t.
Chapter I.5
Symbolic Dynamics
Proposition 5.2 We have hpσ |Σk` q “ log k with respect to any distance dβ .
We also consider the particular class of topological Markov chains.
The restriction σ |ΣA` : ΣA` Ñ ΣA` of σ |Σk` to ΣA` is called the one-sided topological
Markov chain with transition matrix A.
We denote by ρ pAq the spectral radius of a square matrix A.
Theorem 5.4 We have hpσ |ΣA` q “ log ρ pAq with respect to any distance dβ .
Now we consider corresponding two-sided versions of the former notions.
Definition 5.5 Given an integer k ą 1, consider the set Σk “ t1, . . . , kuZ of two-sided
sequences
ω “ p¨ ¨ ¨ i´1 pω qi0 pω qi1 pω q ¨ ¨ ¨ q.
The two-sided shift map σ : Σk Ñ Σk is defined by
ÿ8
an zn
ζ pzq “ exp
n“1
n
Problems
Problem 5.1 Show that for each ω P Σk` and r ą 0 the open ball
(
Bdβ pω , rq “ ω 1 P Σk` : dβ pω 1 , ω q ă r
is a cylinder, that is, there exist integers i1 , . . . , in P t1, . . . , ku with n “ nprq such that
(
Bdβ pω , rq “ Ci1 ¨¨¨in :“ ω P Σk` : i j pω q “ i j for j “ 1, . . . , n .
Problem 5.2 Show that any cylinder Ci1 ¨¨¨in (see Problem 5.1) is simultaneously
open and closed.
Problem 5.3 Given β , β 1 ą 1, show that the distances dβ and dβ 1 generate the same
topology on Σk` .
Problem 5.4 Show that one can define a distance on Σk` by
ÿ8
1
dpω , ω 1 q “ |i pω q ´ in pω 1 q|
n n
n“1
2
Problem 5.6 Show that the distances d and d in Problems 5.4 and 5.5 are not equiv-
alent.
Problem 5.7 Show that pΣk` , dq is a compact metric space.
Problem 5.8 Show that pΣk` , dq is a complete metric space.
Problem 5.9 Show that the shift map σ |Σk is topologically transitive.
Problem 5.10 For a topological Markov chain σ |ΣA` or σ |ΣA with transition matrix
A “ pai j q, show that:
1. the number of blocks pi1 ¨ ¨ ¨ in q of length n with ai p i p`1 “ 1 for p “ 1, . . . , n ´ 1
such that i1 “ i and in “ j is equal to the pi, jq entry of the matrix An ;
2. the number of n-periodic points is trpAn q.
Problem 5.11 Given a k ˆ k matrix A with entries ai j in t0, 1u, show that the fol-
lowing properties are equivalent:
1. there exists n P N such that ai1 i2 ¨ ¨ ¨ ain´1 in “ 0 for all i1 , . . . , in P t1, . . . , ku;
2. An “ 0 for some n P N;
3. ΣA “ ∅.
Problem 5.12 For a topological Markov chain σ |ΣA` , compute the number of peri-
odic orbits with period p prime in terms of the transition matrix A.
Problem 5.13 For a topological Markov chain σ |ΣA` , compute the number of pe-
riodic orbits with period p “ p1 p2 , with p1 , p2 prime, in terms of the transition
matrix A.
2 3
1 3
4 2
Problem 5.16 Show that the topological Markov chain σ |ΣA` with transition matrix
¨ ˛
1010
˚1 1 1 0‹
A“˚ ‹
˝1 1 0 1‚
0111
is dense in Σ2 .
Problem 5.23 Compute the zeta function of the shift map σ |Σk .
Problem 5.24 Compute the zeta function of the topological Markov chain σ |ΣA`
with transition matrix ¨ ˛
1 01
A “ ˝1 1 1‚.
1 11
Problem 5.25 Compute the zeta function of the automorphism of the torus T2 in-
duced by the matrix ˆ ˙
51
A“ .
92
Ť
Problem 5.26 Given a nonempty finite set F in ně2 t1, . . . , kun (that is, a set of
blocks with finite length), consider the set F1 Ĺ Σk of all sequences in Σk containing
no blocks in F. Determine two sets F1 and F2 with k “ 2 such that F11 “ F21 ‰ ∅
with F1 ‰ F2 .
Problem 5.27 Show that F11 X F21 “ pF1 Y F2 q1 .
Ť
Problem 5.28 Given a nonempty finite set of blocks F in ně2 t1, . . . , ku
n, show
that F1 Ĺ Σk is compact and σ -invariant.
I.5 Symbolic Dynamics 55
Ť
Problem 5.29 Given a nonempty finite set of blocks F in ně2 t1, . . . , kun , show
that if F contains only blocks of length 2, then F1 “ ΣA for some k ˆ k matrix A with
entries in t0, 1u.
Problem 5.30 Show that for k “ 2 and F “ tp22qu, the matrix A in Problem 5.29 is
ˆ ˙
11
A“ .
10
Problem 5.31 Compute the number of n-periodic points of the restriction σ |F1 of
σ : Σ2 Ñ Σ2 to F1 for the set F “ tp22qu.
f (x)
p1 1 x
(see Figure I.5.3 for an example). Let i0 ¨ ¨ ¨ in´1 be the base-2 representation of j
and let p1 , . . . , p2n “ p0 be the preimages of the unique fixed point p of f under f n
(see Figure I.5.4). Show that if f : S1 Ñ S1 is an expanding map with deg f “ 2 (see
Problem 1.2.7), then the intervals
satisfy:
56 I.5 Symbolic Dynamics
1. |Ii0 ¨¨¨in´1 | ď λ ´n ;
2. Ii0 ...in´1 in Ĺ Ii0 ...in´1 ; and
3. f m pIi0 ¨¨¨in q “ Iim ¨¨¨in for each m ď n.
f (x)
p1
In this chapter we consider some basic topics of ergodic theory. This includes the
notion of an invariant measure, Poincaré’s recurrence theorem and Birkhoff’s er-
godic theorem. We also consider briefly the notion of metric entropy of an invariant
probability measure. The prerequisites from measure theory and integration theory
are briefly recalled in the initial section. We refer the reader to [8, 23, 35, 46, 56] for
additional topics.
for any pairwise disjoint sets Bn P A, for n P N. Then the triple pX, A, μ q is called a
measure space.
One can show that there exists a unique measure λ : Bn Ñ r0, `8s on Rn such
that ˜ ¸
źn ź
n
λ pai , bi q “ pbi ´ ai q
i“1 i“1
The statement can be rephrased by saying that almost every point in A (that is,
every point in AzB for some set B of measure zero) returns infinitely often to A.
The second result is instead of quantitative nature.
Theorem 6.10 (Birkhoff’s ergodic theorem) Let f : X Ñ X be an A-measurable
map preserving a probability measure μ on X. Given a μ -integrable function
ϕ : X Ñ R, the limit
1ÿ
n´1
ϕ f pxq “ lim ϕ p f k pxqq
nÑ8 n
k“0
with the convention that 0 log 0 “ 0. Moreover, for each n P N we consider the new
partition ξn formed by the sets
C1 X f ´1C2 X ¨ ¨ ¨ X f ´pn´1qCn ,
with C1 , . . . ,Cn P ξ .
60 I.6 Ergodic Theory
hμ p f q “ sup hμ p f , ξ pnq q
nPN
Problems
2. if A, B P B and ` ˘
μ pAzBq Y pBzAq “ 0,
then μ pAq “ μ pBq.
Problem 6.3 Show that the Borel σ -algebra B on R is generated by the set of all
intervals:
I.6 Ergodic Theory 61
1. ra, bs with a ă b;
2. pa, bs with a ă b.
Problem 6.4 Show that the Borel σ -algebra B on R is generated by the set of all
intervals:
1. pa, `8q with a P R;
2. p´8, bq with b P R;
3. ra, `8q with a P R;
4. p´8, bs with b P R.
Problem 6.5 Show that the Borel σ -algebra on R coincides with the σ -algebra gen-
erated by the open sets in R.
Problem 6.6 Given a set C generating the Borel σ -algebra B on R, show that a
function ϕ : X Ñ R is A-measurable if and only if ϕ ´1 B P A for every B P C.
Problem 6.7 Show that the square of an A-measurable function ϕ : X Ñ R is also
an A-measurable function.
Problem 6.8 Show that any increasing function ϕ : R Ñ R is B-measurable.
Problem 6.9 Given A-measurable functions ϕn : X Ñ R, for n P N, show that
supnPN ϕn and infnPN ϕn (here assumed to be finite everywhere) are also A-mea-
surable functions.
Problem 6.10 Given A-measurable functions ϕn : X Ñ R, for n P N, show that
for all x P X. Show that given ε ą 0, there exists a set B P A with μ pBq ă ε such that
ϕn Ñ ϕ when n Ñ 8
uniformly on XzB.
Problem 6.12 Let ϕ : X Ñ R be an A-measurable function and let μ be a measure
on X. Show that for each t, c ą 0 we have
ż
` (˘ 1
μ x P X : |ϕ pxq| ě t ď c |ϕ |c d μ .
t X
62 I.6 Ergodic Theory
Problem 6.13 Show that the measure μ on S1 obtained from the Lebesgue mea-
sure λ on r0, 1s is invariant under the expanding map Em : S1 Ñ S1 .
ř Given finitely many or countably many intervals pai , bi q Ĺ r0, 1s, for
Problem 6.14
i P I, with iPI pbi ´ ai q “ 1, consider a map f : r0, 1s Ñ r0, 1s such that
x ´ ai
f pxq “
bi ´ ai
for each x P pai , bi q and i P I (see Figure I.6.1 for an example). Show that f preserves
the Lebesgue measure λ on r0, 1s.
(see Figure I.6.3). Show that f preserves the measure μ on r0, 1s defined by
ż
dx
μ pAq “ .
A 1`x
Problem 6.17 Show that the map g : r0, 1s Ñ r0, 1s given by gpxq “ 4xp1 ´ xq pre-
serves the measure μ on r0, 1s defined by
ż
dx
μ pAq “ a .
A xp1 ´ xq
px, yq ÞÑ px ` α , x ` yq
preserves the measure m on T2 obtained from the Lebesgue measure on r0, 1s2 .
Problem 6.19 Let f be the Gauss map in Problem 6.16. Show that if
1
x “ ra1 , a2 , . . .s “
a1 ` a2 `¨¨¨
1
64 I.6 Ergodic Theory
3. for m P N we have
1 pm
ra1 , a2 , . . . , am s :“ “ .
a1 ` 1
1
qm
a2 `
¨¨¨` a1m
ÿ
k ÿ
k
pi “ 1 and pi pi j “ p j
i“1 i“1
Moreover, let μ be the measure on S1 obtained from the Lebesgue measure on r0, 1s.
Show that:
1. if ϕ pxq “ e2π imx with m P Z, then
ż
1ÿ
n
lim ϕ pRα pxqq “
k
ϕ dμ for all x P S1 ;
nÑ8 n S1
k“1
66 I.6 Ergodic Theory
1 ÿ ` ´k ˘
n´1
lim μ f A X B “ μ pAqμ pBq
nÑ8 n
k“0
1 ÿ ` ´i ˘
n´1
lim μ f A X B ´ μ pAqμ pBq “ 0
nÑ8 n
i“0
Hμ pξ q ď Hμ pη q.
Problem 6.34 Let μ be a probability measure on X. Show that if ξ and η are parti-
tions of X and (
ξ _η “ C XD : C P ξ,D P η ,
then
Hμ pξ _ η q ď Hμ pξ q ` Hμ pη q.
Problem 6.35 Let f : X Ñ X be an A-measurable map preserving a probability
measure μ on X. Show that if ξ is a partition of X and
f ´1 ξ “ t f ´1C : C P ξ u,
then
Hμ p f ´1 ξ q “ Hμ pξ q.
Problem 6.36 Let f : X Ñ X be an A-measurable map preserving a probability
measure μ on X. Show that for each partition ξ of X we have
˜ ¸ ˜ ¸
1 ł
n´1
1 ł
n´1
hμ p f , ξ q “ inf Hμ f ´i ξ “ lim Hμ f ´i ξ .
nPN n nÑ8 n
i“0 i“0
then
Hμ pξ |ζ q ď Hμ pξ |η q.
Problem 6.38 Let μ be a probability measure on X. Show that if ξ , ζ and η are
partitions of X, then
Hμ pξ _ ζ |η q “ Hμ pξ |ζ _ η q ` Hμ pζ |η q.
Problem 1.1 Determine all values of a P R for which the map f : R Ñ R defined by
f pxq “ ax4 ´ x has nonzero fixed points.
Solution The fixed points of f are the solutions of the equation f pxq “ x, that is,
ax4 ´ x “ x ðñ xpax3 ´ 2q “ 0.
Problem 1.2 Determine all the periodic points of the map f : R Ñ R defined by
f pxq “ ex .
Solution First observe that since ex takes only positive values, there are no periodic
points in R´
0 . Now consider the function
gpxq “ ex ´ x.
This shows that f has no fixed points in R` . Finally, one can show by induction that
for x ą 0, which implies that f has no periodic points in R` . Summing up, f has no
periodic points in R.
Problem 1.3 Show that if a continuous map f : R Ñ R has a periodic point with
period 2, then it has at least one fixed point.
© Springer Nature Switzerland AG 2019 71
L. Barreira and C. Valls, Dynamical Systems by Example, Problem Books
in Mathematics, https://doi.org/10.1007/978-3-030-15915-3 7
72 II.1 Basic Theory
gpxq “ f pxq ´ x.
Then
gppq “ f ppq ´ p “ q ´ p
and
gpqq “ f pqq ´ q “ p ´ q.
Since p ‰ q, the numbers gppq and gpqq have different signs. Hence, by the inter-
mediate value theorem there exists c in the interval determined by p and q with
gpcq “ 0, which thus is a fixed point of f .
a c d b
Fig. II.1.1 A map f satisfying ra, bs Ď f pra, bsq.
Problem 1.4 Show that a continuous map f : ra, bs Ñ R with f pra, bsq Ě ra, bs has
at least one fixed point.
Solution Since f is continuous and f pra, bsq Ě ra, bs, there exist c, d P ra, bs such
that
f pcq “ a ď c and f pdq “ b ě d
(see Figure II.1.1). Since
it follows from the continuity of f that there exists α P ra, bs such that f pα q´ α “ 0,
which thus is a fixed point of f .
II.1 Basic Theory 73
Problem 1.5 Show that a continuous map f : ra, bs Ñ ra, bs has at least one fixed
point.
a b
Fig. II.1.2 A map f satisfying ra, bs Ě f pra, bsq.
such that f is linear on rn, n ` 1s for n “ 1, 2, 3, 4 (see Figure I.1.2). Show that:
1. f has periodic points with period 5;
2. f 3 has no fixed points in r1, 3s Y r4, 5s, but has a fixed point in r3, 4s;
3. f has no periodic points with period 3.
t1, 3, 4, 2, 5u (II.1.1)
1 2 3 4 5
2. Note that 1, 2, 3, 4 and 5 are not 3-periodic points, since they belong to the
periodic orbit in (II.1.1). Moreover,
f 3 pr1, 2sq “ r2, 3s, f 3 pr2, 3sq “ r3, 5s and f 3 pr4, 5sq “ r1, 4s.
Hence, f 3 has no fixed points in the intervals r1, 2s, r2, 3s or r4, 5s. On the other
hand, since
f 3 pr3, 4sq “ r1, 5s Ľ r3, 4s,
it follows from Problem 1.4 that f 3 has a fixed point x in r3, 4s. Alternatively, one
could look at the graph of the map f 3 (see Figure II.1.3).
3. We need to verify that the point x in item 2 is a fixed point of f and not a periodic
point with period 3. Observe that f pxq P r2, 4s. If f pxq P r2, 3s, then
which is impossible because f 3 pxq “ x P r3, 4s. Thus, we must have f pxq P r3, 4s.
Moreover, f 2 pxq P r3, 4s. Again, if f 2 pxq P r2, 3s, then f 3 pxq P r4, 5s, which is im-
possible. Therefore, the orbit of x is contained in r3, 4s. But on this interval we have
f pxq “ 10 ´ 2x and so 10{3 is the only fixed point. Finally,
f 3 pxq “ 30 ´ 8x on r3, 4s
and so 10{3 is also the only fixed point of f 3 . This shows that f has no periodic
points with period 3.
Problem 1.9 Show that if x is a periodic point of a map f with period 2n, then x is
a periodic point of f 2 with period n.
Since ( (
f i pxq : i “ 1, . . . , 2n ´ 1 Ľ f 2i pxq : i “ 1, . . . , n ´ 1 ,
this yields the desired result.
Problem 1.10 Show that if x is a periodic point of f 2 with period n even, then x is a
periodic point of f with period 2n.
Solution Assume that x is a periodic point of f 2 with period n even. Then f i pxq ‰ x
for all even i with 1 ă i ă 2n. We claim that f i pxq ‰ x for all odd i dividing 2n
(these are the possible periods of x with respect to f ). Note that any such i satisfies
i ă n. If f i pxq “ x, then f 2i pxq “ x. But 2i is even and less than 2n, which gives a
contradiction. Therefore, f i pxq ‰ x for all odd i dividing 2n and so x has period 2n
with respect to f .
76 II.1 Basic Theory
Problem 1.11 Show that if x is a periodic point of f 2 with period n odd, then x is a
periodic point of f with period n or 2n.
Solution Assume that x is a periodic point of f 2 with period n odd. Then f i pxq ‰ x
for all even i with 1 ă i ă 2n. To show that x has period n or 2n with respect to
f , it suffices to verify that f i pxq ‰ x for all odd i ‰ n dividing 2n. Note that any
such i satisfies i ă n. This allows one to proceed as in Problem 1.10 to reach a
contradiction. Hence, x has period n or 2n with respect to f .
Problem 1.12 Given positive integers m and n, show that if x is a periodic point
of f with period m, then it is a periodic point of f n with period m{pm, nq (recall that
pm, nq denotes the greatest common divisor of m and n).
Solution Let p be the period of x with respect to f n . Then m divides np since
x “ p f n q p pxq “ f np pxq
and so m{pm, nq divides np{pm, nq. Since the integers m{pm, nq and n{pm, nq are
coprime, we conclude that m{pm, nq divides p. On the other hand,
f pXq Ě f pPq “ P.
Since f is continuous and X is compact, the set f pXq is also compact and so
f pXq Ě P “ X
But since x and y are periodic points, denoting by n and m the periods, respectively,
of x and y we obtain
x “ f nm pxq “ f nm pyq “ y.
This contradiction shows that f pxq ‰ f pyq and so f is one-to-one.
|x j ´ y j | “ | f n j m j px j q ´ f n j m j py j q| ď ε
x ´ y “ lim pxn ´ yn q “ 0,
nÑ8
Solution Consider the map F : R Ñ R defined by Fpxq “ 2x. Note that F n pIq “
p2n a, 2n bq is an open interval of length 2n pb ´ aq. Let n be the smallest positive
integer such that
2n pb ´ aq ě 1.
Then the projection of F n pIq on S1 is exactly S1 and so E2n pIq “ S1 .
In particular, there exist u, v P px ´ δ , x ` δ q such that E2n puq “ 0 and E2n pvq “ 1{2.
Since
1 n
“ E puq ´ E2n pvq
2 2
ď E2n puq ´ E2n pxq ` E2n pxq ´ E2n pvq,
we conclude that
n n
E puq ´ E n pxq ě 1 E pxq ´ E n pvq ě 1 .
or
2 2 2 2
4 4
Ť
Problem 1.20 Given x P S1 , show that the union 8 ´n 1
n“1 E2 x is dense in S .
R “ tz P C : |z| “ 1u.
mq x ´ x “ pmq ´ 1qx “ p1
and
mq y ´ y “ pmq ´ 1qy “ p2
for some p1 , p2 P Z. This implies that the q-periodic points of f are
´ p p2 ¯
1
px, yq “ , for p1 , p2 “ 1, 2, . . . , mq ´ 1.
mq ´ 1 mq ´ 1
To verify that the set of periodic points of f is dense in T2 , it suffices to note that
1{pmq ´ 1q tends to zero when q Ñ 8.
80 II.1 Basic Theory
pAq ´ Idqx “ y.
x “ pAq ´ Idq´1 y.
Since Aq ´ Id has only integer entries, the matrix pAq ´ Idq´1 has rational entries
and so x P Qn .
Now take rxs P Qn {Zn . One can always write x in the form
´p pn ¯
1
x“ ,..., (II.1.4)
r r
for some integers p1 , . . . , pn P t0, 1, . . . , r ´ 1u. Since A has only integer entries, for
each q P N we have ˆ 1 ˙
p1 p1
Aq x “ ,..., n `y
r r
for some p11 , . . . , p1n P t0, 1, . . . , r ´ 1u and y P Zn . But since the number of points as
in (II.1.4) is rn , there exist q1 , q2 P N with q1 ‰ q2 such that
Aq1 x ´ Aq2 x P Zn .
and so
TAq1 ´q2 rAq2 xs “ rAq2 xs.
In other words, TAq2 rxs “ rAq2 xs is a pq1 ´ q2 q-periodic point. But since TA is invert-
ible, rxs is also a periodic point of TA .
using the Smith normal form (which says that A “ PDQ, for some matrices P, D
and Q with integer entries such that |det P| “ |det Q| “ 1 and D is diagonal).
|det P| “ |det Q| “ 1,
II.1 Basic Theory 81
Dy “ x ` p with p “ pp1 , . . . , pn q P Zn .
Therefore, ˆ ˙
p1 pn
y “ D´1 x ` ,..., ,
k1 kn
which gives „ ˆ ˙j
p1 pn
rys “ D´1 x ` ,...,
k1 kn
with pi “ 1, . . . , |ki | for each i “ 1, . . . , n. Hence,
of A are not roots of 1. Hence, by Problem 1.25, the number of q-periodic points
of TA is equal to |detpAq ´ Idq|. Moreover,
ˆ q ˙
τ ´1 0
|detpAq ´ Idq| “ det
0 τ ´1
´q
f pXzAq Ď XzA.
Show that:
1. if U is an open neighborhood of Λ , then f ´n A Ď U for any sufficiently large n;
2. the set Λ is f -invariant.
II.1 Basic Theory 83
f 1A
Λ 2A
f
The open sets Vn “ XzBn satisfy Vn`1 Ě Vn for all n P N and together with U they
cover X. By compactness, there exists a finite subcover, say tU,V1 , . . . ,VN u, and so
tU,VN u is a cover of X. Hence,
f ´n A “ Bn Ď U for all n ě N.
2. We have
8
č
f ´1Λ “ f ´n A.
n“1
Problem 1.30 Find the largest E3 -invariant set A contained in J “ r0, 1{3s Y r2{3, 1s
(see Figure I.1.5).
On the other hand, E3´1 t0u ‰ t0u and so the set t0u is not E3 -invariant. Hence, ∅ is
the largest E3 -invariant set contained in J.
Solution First note that the vector field is of class C1 , which ensures that each
initial condition determines a unique solution. Each solution px, yq “ pxptq, yptqq of
the equation satisfies
This implies that each orbit remains for all time in a set of the form
(
Ca “ px, yq P R2 : y6 ` x4 “ a
for some a ě 0 (see Figure II.1.5). Since the sets Ca are compact, this ensures that
each solution is global and so
Ť that the equation determines a flow. Moreover, each
set Ca and so also the union aPI Ca are invariant under the flow determined by the
equation.
x2
z “ y´ .
1 ` 2a
This is indeed a coordinate change since the corresponding Jacobian matrix has
determinant ˆ ˙
1 0
det “ 1 ‰ 0.
´2x{p1 ` 2aq 1
We have
´ x 2 ¯1
z1 |z“0 “ y ´ | 2
1 ` 2a y“x {p1`2aq
´ 2x2 ¯
“ ´y ` x2 ´ 2y2 ´ pa ´ yq |y“x2 {p1`2aq
1 ` 2a
x 2 2x4 2x2 ´ x2 ¯
“´ ` x2 ´ ´ a´ “ 0.
1 ` 2a p1 ` 2aq 2 1 ` 2a 1 ` 2a
This implies that each solution starting in S2 remains in S2 for all time.
Solution One can easily verify that the solutions of the equation are
ˆ ˙1 ˆ ˙ˆ ˙
x 0 1 x
“ (II.1.5)
y ´1 0 y
(which has the phase portrait in Figure II.1.6). To obtain the corresponding flow, we
need to write each solution pxptq, yptqq in terms of the initial condition px0 , y0 q “
pxp0q, yp0qq. We have
yptq “ x1 ptq “ ´c1 sint ` c2 cost
for t P R and so
x0 “ c1 and y0 “ c2 .
Therefore,
xptq “ x0 cost ` y0 sint
and
yptq “ ´x0 sint ` y0 cost,
which gives the flow
ˆ ˙
x0 cost ` y0 sint
ϕt px0 , y0 q “
´x0 sint ` y0 cost
ˆ ˙ˆ ˙
cost sint x0
“ ,
´ sint cost y0
for t P R.
II.1 Basic Theory 87
Solution First note that the equation has the phase portrait in Figure II.1.7. We have
x1 x3 ` x
“ ą1
x3 x3
for x ą 0. Integrating we obtain
ˆ ˙
1 1 1
´ ´ ět
2 xptq2 xp0q2
Problem 1.35 Show that the identity map is a Poincaré map for the differential
equation x2 ` x “ 0 and compute the corresponding first return time.
Solution It was shown in Problem 1.33 that the equation generates the flow
for px0 , y0 q P R2 and t P R. The first return time of the point px0 , 0q to the half-line
R` ˆ t0u is 2π , because ϕ2π px0 , 0q “ px0 , 0q and
Solution The critical points of the equation are px, yq “ pnπ , 0q, for n P Z. Note that
in the neighborhoods of the points
for n P Z, the behavior is different: the former points are centers while the latter
are saddles (see Figure II.1.8). It follows readily from the phase portrait that the
Poincaré map is the identity map.
Now we compute the first return time or, equivalently, the periods of the periodic
orbits. The solutions of the equation with initial condition xp0q “ x0 P p0, π q and
yp0q “ 0 satisfy
d ´1 2 ¯
y ´ cos x “ 0.
dt 2
Since cos x “ 1 ´ 2 sin2 px{2q, we obtain
´ dx ¯2
“ 2pcos x ´ cos x0 q
dt ´ ´x ¯ ´ x ¯¯ (II.1.6)
0
“ 4 sin2 ´ sin2 .
2 2
Now let ´x¯ ´x ¯
0
z “ sin , k “ sin2 . (II.1.7)
2 2
?
Note that zp0q “ k P p0, 1q. By (II.1.7) we obtain
II.1 Basic Theory 89
dz 1 dx ´x¯
“ cos
dt 2 dt 2
and ´ dz ¯2 ´ x ¯´ dx ¯2
1
“
cos2
dt 4 2 dt
1´ ´ x ¯¯´ dx ¯2
“ 1 ´ sin2 (II.1.8)
4 2 dt
1 ´ dx ¯2
“ p1 ´ z q 2
.
4 dt
Using identities (II.1.7) and (II.1.8), it follows from (II.1.6) that
´ dz ¯2
“ p1 ´ z2 qpk ´ z2 q.
dt
?
Hence, writing w “ z{ k we obtain
´ dw ¯2
“ p1 ´ w2 qp1 ´ kw2 q.
dt
Moreover, wp0q “ 1 and w1 p0q “ 0. When a periodic orbit travels in the ? upper half
plane, it completes a closed curve in the variables pw, w1 q. Since w “ 1{ k for x “ π ,
the period is given by
ż 1{?k
ds
T “4 a
1 p1 ´ s2 qp1 ´ ks2 q
ż1 ż 1{?k
ds ds
“ ´4 a `4 a .
0 p1 ´ s2 qp1 ´ ks2 q 0 p1 ´ s2 qp1 ´ ks2 q
Let ż1
ds
Kpmq “ a
0 p1 ´ s qp1 ´ ms2 q
2
and ż sin ϕ
ds
Fpϕ , mq “ a
0 p1 ´ s2 qp1 ´ ms2 q
be, respectively, the complete and incomplete elliptical integrals of the first kind.
Note that ˆ ˙
1
T “ ´4Kpkq ` 4F arcsin ? , k .
k
Problem 1.37 Show that there exists a homeomorphism h : R` Ñ R` mapping the
orbits ϕt pxq determined by the equation x1 “ ´x on R` (which has the phase portrait
in Figure I.1.6) onto the orbits ψt pxq determined by the equation x1 “ ´x2 on R`
(which has the phase portrait in Figure I.1.7), and a map τ : R` ˆ R` Ñ R` with
t ÞÑ τ px,tq increasing for each x such that
Solution First note that ϕt pxq “ xe´t . To find the second flow, we solve the initial
value problem
x1 “ ´x2 , xp0q “ x0 .
One can easily verify that
x0 x
xptq “ , that is, ψt pxq “ .
1 ` tx0 1 ` tx
Now we find h and τ such that
hpxq
hpxe´τ q “ .
1 ` thpxq
For example, taking hpxq “ x, the equation becomes
x
xe´τ “ .
1 ` tx
Solving it for τ , we obtain
τ px,tq “ logp1 ` txq
(note that 1 ` tx ě 0), which is an increasing function of t for each given x.
on R2 has unique solutions that are defined for all t P R. Let ϕt : T2 Ñ T2 be the
corresponding flow (see Proposition 1.8). Each solution
of the equation with pxp0q, yp0qq “ p0, zq crosses infinitely often the line x “ 0 (since
f is uniformly bounded from below). The first intersection into the future occurs at
the time (
Tz “ inf t ą 0 : xptq “ 1 .
Show that the map h : S1 Ñ S1 defined by
hpzq “ ypTz , zq
Solution Let pxpt, wq, ypt, wqq be the solution of the equation
Moreover, let (
Sw “ sup t ă 0 : xpt, wq “ 0
and consider the map h : S1 Ñ S1 defined by hpwq “ ypSw , wq. We have
h ˝ h “ id and h ˝ h “ id,
The corresponding solutions on the torus T2 (see Figure II.1.9) are periodic if and
only if pα t, β tq P Z2 for some t ‰ 0, which implies that
α {β “ pα tq{pβ tq P Q.
That is, the solutions are periodic if and only if α {β P Q. Clearly, a periodic solution
is not dense and so one cannot have α {β P Q.
On the other hand, when α {β R Q all orbits are dense. Indeed, for the times
t “ n{α with n P Z we have
This implies that the intersection of each orbit with a given horizontal line coincides
with an orbit of a rotation with irrational rotation number α . This readily implies
that when α {β R Q each orbit of the equation px1 , y1 q “ pα , β q is dense.
Problem 1.40 Consider the differential equation v1 “ f pvq on Rn for some map
f : Rn Ñ Rn of class C1 . Write f “ p f1 , . . . , fn q, v “ pv1 , . . . , vn q and let
ÿ
n
B ϕ pvq
pLϕ qpvq “ fi pvq
i“1
Bvi
for each C2 function ϕ : Rn Ñ R. Show that each periodic orbit of the equation:
1. has at least two points in the set
(
Aϕ “ v P Rn : pLϕ qpvq “ 0 ;
92 II.1 Basic Theory
0 1
Solution 1. Note that if vptq is a periodic solution of the equation, then there exist
t “ t1 and t “ t2 such that
(because any periodic orbit is a compact set). Since the points t1 and t2 are local
extrema of the function ϕ ˝ v, for t “ t1 and t “ t2 we have
d2 d
2
ϕ pvptqq “ pLϕ qpvptqq
dt dt
d ÿ
n
Bϕ
“ fi pvptqq pvptqq
dt i“1 Bvi
ÿ n ÿn ˆ ˙
B fi B ϕ B2 ϕ
“ fj ` fi f j |v“vptq
i“1 j“1
Bv j Bvi Bvi Bv j
and so
ÿ
n
BpLϕ q
pL2 ϕ qpvptqq “ f j pvptqq pvptqq
j“1
Bv j
ÿ n ÿn ˆ ˙
B fi B ϕ B2 ϕ
“ f j pvptqq ` fi |
i“ j j“1
Bvi Bvi Bvi Bv j v“vptq
d2
“ ϕ pvptqq.
dt 2
Therefore, vpt1 q P Bϕ and vpt2 q P Cϕ .
3. It follows from (II.1.9) that
Therefore, if pL2 ϕ qpvptqq ‰ 0, then ∇ϕ pvptqq ‰ 0 and so the periodic orbit deter-
mined by the solution vptq intersects the surface pLϕ qpvq “ 0 transversally.
Chapter II.2
Topological Dynamics
f pω pxqq Ď ω pxq.
Now we show that y “ f pzq for some z P ω pxq. Since X is compact, the sequence
f nk ´1 pxq has a convergent subsequence, say f mk pxq, with limit z P ω pxq. But since
f is continuous, we obtain
´ ¯
f pzq “ f lim f mk pxq
kÑ8
“ lim f mk `1 pxq “ y.
kÑ8
Solution Assume that γ ` pxq “ X. Given y P X, since X has no isolated points, there
exist sequences of positive integers nk Õ 8 and of positive numbers rk Œ 0 such
that
f nk pxq P Bpy, rk qzBpy, rk`1 q for all k P N
(see Figure II.2.1). Clearly, f nk pxq Ñ y when k Ñ 8 and so y P ω pxq. This shows
that ω pxq “ X, which gives a contradiction.
© Springer Nature Switzerland AG 2019 95
L. Barreira and C. Valls, Dynamical Systems by Example, Problem Books
in Mathematics, https://doi.org/10.1007/978-3-030-15915-3 8
96 II.2 Topological Dynamics
B(y, rk + 1 )
y B(y, rk )
f nk (x)
Solution Assume that ω pxq contains infinitely many points and that y P ω pxq is
an isolated point. Then there exists a sequence nk Õ 8 in N such that f nk pxq Ñ y
when k Ñ 8. But since y is isolated, one must have fnk pxq “ y for infinitely many
integers k. Hence, y must be periodic and so ω pxq “ γ pyq is finite. This contradiction
shows that ω pxq has no isolated points.
Problem 2.4 Let f : I Ñ I be a continuous map on the interval I “ r0, 1s. Given
x P I, show that for any a, b P ω pxq and any open neighborhood U of a, there exists
Ť
an increasing sequence of positive integers pki qiPN such that b P iPN f ki pUq.
Solution Since a, b P ω pxq, for any open neighborhood V of b there exist increasing
sequences ni ą mi , for i P N, and an integer N P N such that
with
f mi pxq P U and f ni pxq P V
for all i ě N. For each i P N, let zi “ f mi pxq and ki “ ni ´ mi . Clearly,
Thus,
AzB Ĺ f ´1V “ W,
with W an open set (by the continuity of f ). So
f pW q “ f pW q Ď V
Therefore, ´ ¯
f pyq “ f lim f ni pxq “ lim f ni `1 pxq P U
iÑ8 iÑ8
Problem 2.6 Compute the α -limit set and the ω -limit set of each point in R for the
flow determined by the differential equation x1 “ ´xp1 ` cos2 xq on R.
On the other hand, x1 ă 0 for x ą 0 and x1 ą 0 for x ă 0. Hence, ω pxq “ t0u and
α pxq “ ∅ for each x ‰ 0 (see Figure II.2.2).
written in polar coordinates. Compute the α -limit set and the ω -limit set of each
point in R2 for the flow determined by the differential equation.
Problem 2.8 Compute the α -limit set of each point px, yq P R2 with |x| ă 1 for the
flow determined by the differential equation
#
x1 “ px2 ´ 1qpy ´ xq,
y1 “ x.
II.2 Topological Dynamics 99
By sketching the phase portrait (see Figure II.2.4), we find that the α -limit set of
any point px, yq with |x| ă 1 is the origin.
x 1 x 1
z P f n pUq XU ‰ ∅,
for every x P X.
Solution We proceed by contradiction. If the property did not hold, then it would
exist a compact set K Ĺ X containing infinitely many points of the positive semiorbit
of x such that K X NW p f q “ ∅. This implies that the semiorbit has an accumulation
point y P K. Then y P ω pxq, but on the other hand y R NW p f q, which contradicts to
Problem 2.10.
Problem 2.12 Let f : X Ñ X be a continuous map. Show that Rp f q (the set of re-
current points) is forward f -invariant.
Problem 2.14 A nonempty closed forward invariant set without nonempty closed
forward invariant proper subsets is said to be minimal. Show that any two distinct
minimal sets for a map f must have empty intersection.
Solution Let M1 and M2 be minimal sets for the map f and assume that
A “ M1 X M2 ‰ ∅.
Solution (1 ñ 2). Assume that M is a minimal set and take x P M. Then γ ` pxq is a
nonempty closed forward f -invariant subset of M and so γ ` pxq “ M.
(2 ñ 3). Now assume that M “ γ ` pxq for all x P M. Then for each y P M we have
č č
ω pyq “ γ ` pxq “ M “ M.
xPγ ` pyq xPγ ` pyq
∅ ‰ f n p f ´nV XUq
Ď f n p f ´nV q X f n pUq
Ď V X f n pUq
and so V X f n pUq ‰ ∅.
102 II.2 Topological Dynamics
Now assume that for any nonempty open sets U,V Ĺ X there exists n P N such
that f n pUq X V ‰ ∅. Given x P f n pUq X V , there exists y P U such that x “ f n pyq.
Hence, y P U X f ´nV , which shows that U X f ´nV ‰ ∅ and so f is topologically
transitive.
Ť
Problem 2.18 Let f : X Ñ X be a continuous map. Ť8 Show that if 8
n“1 f pUq is
n
´n
dense for any nonempty open set U Ď X, then n“1 f U is also dense for any
nonempty open set U Ď X.
Solution It suffices to show that for any nonempty open sets U,V Ĺ X there exists
n P N such that U X f ´nV ‰ ∅. By the hypothesis, we have
8
ď
UX f n pV q ‰ ∅.
n“1
Hence, there exists x P U such that x P f n pV q for some n P N. Therefore, there exists
y P V such that x “ f n pyq and so y P V X f ´n pUq.
f ´n pXzCq XW ‰ ∅. (II.2.2)
Take x P f ´1 pXzCq. Then f pxq P XzC. Since C is forward f -invariant, this implies
that x R C (because f pxq P C whenever x P C). Hence, f ´1 pXzCq Ď XzC and so also
f ´n pXzCq Ď XzC,
Solution (1 ñ 2). Let pUn qnPN be a basis for the topology. For each n P N, we
consider the open set ď
Vn “ f ´iUn .
iPN
Therefore,
x P f ´mU X f ´nV “ f ´n p f ´pm´nqU XV q
and so the set f ´pm´nqU XV is nonempty. Thus, f is topologically transitive.
Solution Consider the set X “ t0, 1u with the discrete topology and define a map
f : X Ñ X by f p0q “ f p1q “ 0. Then the positive semiorbit of the point 1 is the
whole space, but f is not topologically transitive. Indeed, let U “ t0u and V “ t1u.
Then
f n pUq “ U and f n pUq XV “ ∅
for all n P N. Hence, it follows from Problem 2.17 that f is not topologically transi-
tive.
X “ t0u Y t1{n : n P Nu
with the distance induced from the usual distance on R and define a map f : X Ñ X
by
f p0q “ 0 and f p1{nq “ 1{pn ` 1q for n P N.
Note that f is continuous and that γ ` p1q is the only dense positive semiorbit. On
the other hand, f is not topologically transitive. For example, for the open sets U “
t1{2u and V “ t1u we have f ´nU XV “ ∅ for all n P N.
Problem 2.23 Show that the continuous map f : r0, 1s X Q Ñ r0, 1s X Q defined by
f pxq “ 1 ´ |2x ´ 1| has no dense positive semiorbits, but is topologically transitive.
Solution Each point x P r0, 1s X Q has a finite positive semiorbit. Indeed, if x “ q{p
with p, q P N, then f pxq “ q1 {p for some q1 P N. Therefore, the positive semiorbit
of x has at most cardinality p and so it is not dense.
104 II.2 Topological Dynamics
On the other hand, the graph of f m consists of 2m´1 “tents” (see Figure II.2.5)
and so given nonempty open intervals I, J Ĺ r0, 1s X Q, there exists m P N such that
f ´m I X J ‰ ∅.
Problem 2.24 Show that the map f in Problem 1.21 is topologically transitive.
Solution Let U,V Ĺ R be nonempty connected open sets. Take points e2π iθ P V and
e2π iϕ P U with
θ “ 0.θ1 θ2 ¨ ¨ ¨ and ϕ “ 0.ϕ1 ϕ2 ¨ ¨ ¨
written in base-3. For any sufficiently large n P N, the point
p “ expp2π ip0.θ1 θ2 ¨ ¨ ¨ θn ϕ1 ϕ2 ¨ ¨ ¨ qq
Problem 2.25 Let Rα : S1 Ñ S1 be a rotation of the circle with α P RzQ. Show that
tRm
α pxq : m P Zu “ S
1
for every x P S1 .
II.2 Topological Dynamics 105
Solution If Rm m2
α pxq “ Rα pxq for some integers m1 ą m2 , then
1
m1 α ´ m2 α “ m
H ˝ f q “ gq ˝ H on X
Now assume that p is an attracting fixed point of f . Then there exists an open
neighborhood U of p such that for x P U we have f n pxq Ñ p when n Ñ 8. Let V “
hpUq. Then V is an open neighborhood of q (because h is a homeomorphism). Take
y P V and let x “ h´1 pyq P U. Then f n pxq Ñ p when n Ñ 8. Since h is continuous,
we obtain
gn pyq “ gn phpxqq “ hp f n pxqq Ñ hppq “ q
when n Ñ 8. Hence, q is an attracting fixed point of g. The desired statement fol-
lows now from interchanging the roles of f and g.
Problem 2.28 Show that if two maps are topologically conjugate and one of them
is topologically mixing, then the other is also topologically mixing.
f ˝h “ h˝g
Problem 2.29 Given a topologically mixing map f : X Ñ X, show that the map
f ˆ f : X ˆ X Ñ X ˆ X defined by
Solution Let U1 ,U2 ,V1 ,V2 Ď X be nonempty open sets and define U “ U1 ˆU2 and
V “ V1 ˆV2 . Since f is topologically mixing, there exists m P N such that
for all n ě m. Since the sets of the form U ˆ V with U and V open sets in X form
a basis for the topology of X ˆ X, we conclude that the map f ˆ f is topologically
mixing.
Solution It follows from the definition of the distance dn in Definition 2.10 that
dn “ d for all n P N. Hence,
1
hp f q “ lim lim sup log Np1, ε q “ 0.
ε Ñ0 nÑ8 n
and
dn, f ´1 p f n´1 pxq, f n´1 pyqq
(
“ max dp f ´k p f n´1 pxqq, f ´k p f n´1 pyqqq : k “ 0, . . . , n ´ 1
“ dn, f px, yq.
Hence, if some points x1 , . . . , x p , with p “ N f pn, ε q, satisfy dn, f px j , x j q ě ε for i ‰ j,
then
dn, f ´1 p f n´1 pxi q, f n´1 px j qq ě ε for i ‰ j.
Therefore,
N f ´1 pn, ε q ě p “ N f pn, ε q.
Interchanging the roles of f and f ´1 we also obtain
N f pn, ε q ě N f ´1 pn, ε q
1
hp f q “ lim lim sup log N f pn, ε q
ε Ñ0 nÑ8 n
1
ě lim lim sup log N f |Y pn, ε q
ε Ñ0 nÑ8 n
“ hp f |Y q.
f n ˝ H “ H ˝ gn
Problem 2.35 Compute the topological entropy of the map f in Problem 1.21.
x j “ e2π ip j{3
n`k q
for j “ 0, . . . , 3n`k ´ 1.
II.2 Topological Dynamics 109
Then
dn px j , x j`1 q “ dp f n´1 px j q, f n´1 px j`1 qq
“ 3n´1 dpx j , x j`1 q
“ 3´pk`1q
for j “ 0, . . . , 3n`k ´ 1. Therefore,
and so
Npn, 3´pk`1q q ě 3n`k . (II.2.4)
On the other hand, given a set X Ĺ R with cardinality at least 3n`k ` 1,
there exist
x, y P X with x ‰ y such that dpx, yq ă 3´pn`kq . Then dn px, yq ă 3´pk`1q and so
which gives
1
hp f q “ lim lim sup log Npn, ε q
ε Ñ0 nÑ8 n
1
“ lim lim sup log Npn, 3´pk`1q q
kÑ8 nÑ8 n
n`k
“ lim lim sup log 3
kÑ8 nÑ8 n
“ log 3.
H ˝ f “ Em ˝ H on Xm
Problem 2.37 Show that if the distances d and d 1 generate both the topology of a
compact topological space X, then
hd p f q “ hd 1 p f q
Solution Let (
Aε “ px, yq P X ˆ X : dpx, yq ě ε .
Note that (
δ pε q :“ min d 1 px, yq : px, yq P Aε Œ 0
when ε Œ 0. If dn px, yq ě ε , then there exists an integer i P t0, . . . , n ´ 1u such that
p f i pxq, f i pyqq P Aε and so
Therefore,
Nd 1 pn, δ pε qq ě Nd pn, ε q
(making explicit the distance that is considered) and so
1
hd 1 p f q “ lim lim sup log Nd 1 pn, δ q
δ Ñ0 nÑ8 n
1
“ lim lim sup log Nd 1 pn, δ pε qq
ε Ñ0 nÑ8 n
1
ě lim lim sup log Nd pn, ε q
ε Ñ0 nÑ8 n
“ hd p f q.
Problem 2.38 Let f : X Ñ X be a continuous map on a compact metric space pX, dq.
Show that:
1. if U is a finite open cover of X, then letting
II.2 Topological Dynamics 111
# +
č
n´1
Un “ f ´kUk : U0 , . . . ,Un´1 P U
k“0
and denoting by NpUn q the smallest cardinality of the finite subcovers of Un , the
limit
1
lim log NpUn q
nÑ8 n
exists;
since q Ñ 8 when n Ñ 8 (for a fixed k). Since k is arbitrary, this implies that
cn ck cn
lim sup ď inf ď lim inf ,
nÑ8 n kPN k nÑ8 n
which establishes property (II.2.8). It follows readily from (II.2.6) and (II.2.8) that
the limit in item 1 exists.
2. Given n P N and ε ą 0, let Bn ppi , ε q be an open ball in the distance dn . It follows
from the definition of Mpn, ε q that
ďε q
Mpn,
Bn ppi , ε q “ X.
i“1
č
n´1
Bn ppi , ε q “ f ´k Bp f k ppi q, ε q
k“0
and so
ďε q n´1
Mpn, č
X“ f ´k Bp f k ppi q, ε q.
i“1 k“0
Now let δ be a Lebesgue number of the open cover U (in particular, any open ball
of radius δ {2 is contained in some element of the cover). Then
for some open set Uki P U. Therefore, for each i “ 1, . . . , Mpn, δ {2q we have
č
n´1 č
n´1
f ´k Bp f k ppi q, δ {2q Ď f ´kUki P Un
k“0 k“0
Since diam U ă ε , for each n P N no element of Un can contain two distinct points
pi , p j P An because ˜ ¸
č
n´1
Vl :“ f l f ´kUk Ď Ul
k“0
and so
diamVl ď diamUl ă ε .
Therefore, Npn, ε q ď NpUn q.
II.2 Topological Dynamics 113
and so
1 1
lim sup log Npn, ε q ď inf lim log NpUn q.
nÑ8 n diam Uăε n
Hence,
1
hp f q “ lim lim sup log Npn, ε q
ε Ñ0 nÑ8 n
1
ď lim inf lim log NpUn q (II.2.10)
ε Ñ0 diam Uăε nÑ8 n
1
“ lim inf lim log NpUn q.
diam UÑ0 nÑ8 n
Combining (II.2.9) and (II.2.10), we conclude that the identity in item 4 holds.
card f ´1 U “ card U.
Hence, card U2 ď 2 card U (see Problem 2.38). One can show by induction that
and so
1
log card Un “ 0.
lim (II.2.12)
n
nÑ8
for each ε (taking a cover U by open intervals of length less than ε ) and so hp f q “ 0.
Problem 3.1 Show that the composition Rα ˝ Rβ of two rotations of the circle
Rα , Rβ : S1 Ñ S1 is also a rotation of the circle.
Solution We have
pRα ˝ Rβ qrxs “ Rα rx ` β s “ rx ` α ` β s
Problem 3.2 Show that given n ˆ n matrices A and B with entries in Z, the composi-
tion of the endomorphisms of the torus TA , TB : Tn Ñ Tn is again an endomorphism
of the torus and determine a matrix C such that TA ˝ TB “ TC .
Solution We have
TA rxs “ rAxs and TB rxs “ rBxs
for each rxs P Tn . Hence,
(see Figure I.3.1). Show that t0, 1{2u is a periodic orbit of f and that ρ p f q “ 1{2.
Problem 3.4 Given a P p0, 1{p2π qq, consider the map f : S1 Ñ S1 defined by
(see Figure I.3.2). Show that each orbit starting in p0, 1{2q is monotonous.
F 1 pxq “ 1 ` 2π a cosp2π xq ą 1 ´ 2π a ą 0.
Solution Let x0 be a periodic point of f with period p and let U0 , . . . ,Up´1 be the
connected component of the set S1 zγ px0 q. The sets Ui are open intervals and, without
loss of generality, we can assume that
when k Ñ ˘8.
Solution We first show that f has fixed points. Take x P r0, 1q and let F be a lift
of f with Fp0q P r0, 1q. Note that f pxq “ x if and only if Fpxq ´ x P Z. Since f is
an orientation-reversing homeomorphism, it follows from Problem 3.6 that Fp1q “
Fp0q ´ 1. Therefore, the continuous function g : R Ñ R defined by gpxq “ Fpxq ´ x
satisfies
gp1q “ gp0q ´ 2. (II.3.1)
Moreover, since F is decreasing (again because f is an orientation-reversing home-
omorphism), the map g is also decreasing and so it follows from (II.3.1) that it takes
exactly two integer values on r0, 1q. This shows that f has fixed points.
Now let G be a lift of f 2 . Since f has fixed points, the same happens with f 2 .
Hence, there exists x P S1 such that Gpxq ´ x P Z and so
Therefore,
118 II.3 Low-Dimensional Dynamics
Gn pxq ´ x
ρ pGq “ lim “ k,
nÑ8 n
which implies that ρ p f 2 q “ tρ pGqu “ 0.
ÿ
n´1
F n pxq ´ x “ rFpF i pxqq ´ F i pxqs
i“0
that
p F n pxq ´ x p
k` `ε ď ď k ` ` 1 ´ ε.
q n q
Thus, " * „ j
F n pxq ´ x p p
ρp f q “ lim P ` ε, ` 1 ´ ε ,
nÑ8 n q q
which contradicts the hypothesis that ρ p f q “ p{q. This shows that there exist x P R
such that Fpxq ´ x ´ p{q P Z.
On the other hand, since r0, 1s is compact, there exists a constant M ą 0 (indepen-
dent of x and n) such that
and
Fpxq ě Fpxn ´ kq “ Fpxn q ´ k.
Hence, |Fpxq ´ Fpxn q| ď k and it follows by induction that
|F m pxq ´ F m pxn q| ď k
|F n pxq ´ F n pxn q| ď k.
Since
F n pxq ´ x F n pxn q ´ xn F n pxq ´ F n pxn q x ´ xn
´ “ ´ ,
n n n n
we obtain n
F pxq ´ x F n pxn q ´ xn 2k
´ ď Ñ0
n n n
when n Ñ 8. This establishes the property in item 1.
2. Now take xn “ tGn pxqu. For each x P R we have
|x ´ xn | ď |x| ` 1 “: k
120 II.3 Low-Dimensional Dynamics
r{s ă ρ p f q ă p{q ă 1
for all x P R;
2. if g : S1 Ñ S1 is an orientation-preserving homeomorphism with
F q px ` pq “ F q pxq ` p
F 2q pxq “ F q pF q pxqq ě F q px ` pq
“ F q pxq ` p ě x ` 2p
F nq pxq ě x ` np for n P N.
Thus,
F nq pxq ´ x np p
ρ pFq “ lim ě lim “ . (II.3.2)
nÑ8 nq nÑ8 nq q
On the other hand, since Fp0q P p0, 1q, we have ρ pFq P r0, 1q and so it follows from
(II.3.2) that ρ p f q ě p{q, which contradicts the fact that ρ p f q ă p{q. This shows that
F q pxq ă x ` p. The second inequality in item 1 can be obtained analogously.
2. Since F q ´ id is periodic and continuous, it has a maximum. Hence, it follows
from the first inequality in item 1 that there exists δ ą 0 such that
Since the intervals f ´kpm´nq I, for k P N, have adjacent endpoints, the sequence
xk “ f ´kpm´nq p f n pxqq
Solution Take x, y P S1 and assume that f mn pxq Ñ x0 P ω pxq when n Ñ 8 for some
sequence mn Õ 8. By Problem 3.12, for each n P N there exists kn P N such that
Then f kn pyq Ñ x0 when n Ñ 8 and it follows from Proposition 2.4 that ω pxq Ď
ω pyq. Interchanging x and y we also obtain ω pyq Ď ω pxq and so ω pxq “ ω pyq.
Problem 3.17 Determine whether the maps E4 and R1{2 are topologically conjugate.
Solution By Problem 2.26, if the maps E4 and R1{2 were topologically conjugate,
then they would have the same number of fixed points. But while the expanding
map E4 has fixed points, such as the origin, the rotation R1{2 has none. Therefore,
the maps are not topologically conjugate.
Problem 3.18 Determine for which values of α P r0, 1q the rotations Rα and R2α
are topologically conjugate.
Solution If Rα and R2α are topologically conjugate, then ρ pRα q “ ρ pR2α q. Since
we obtain
α ” 2α mod 1, that is, α ” 0 mod 1.
But since α P r0, 1q, we conclude that α “ 0. In particular, for α P p0, 1q the maps
Rα and R2α are not topologically conjugate. On the other hand, for α “ 0 we have
Rα “ R2α “ id
Problem 3.19 Show that the map f in Problem 3.16 is topologically mixing.
(see Figure II.3.1 for the graph of f 2 ). Now let U Ĺ S1 be a nonempty open set.
Then there exists n P N and i P t0, 1, . . . , 2n ´ 1u such that
ˆ ˙
i i`1
I“ , Ď U.
2n 2n
Problem 3.20 Show that the map g : S1 Ñ S1 given by gpxq “ 4xp1 ´ xq is topolog-
ically mixing.
Solution By Problem 3.16, the map g is topologically conjugate to the map f given
by f pxq “ 1´|2x´1|. On the other hand, by Problem 3.19, the latter is topologically
mixing and so it follows from Problem 2.28 that g is also topologically mixing.
II.3 Low-Dimensional Dynamics 125
Solution Again by Problem 3.16, the map g is topologically conjugate to the map f
given by f pxq “ 1 ´ |2x ´ 1|. Let h : S1 Ñ S1 be a homeomorphism such that
f ˝ h “ h ˝ g on S1 .
On the other hand, it is shown in Problem 3.19 that there exists n P N such that
f n phpIqq “ S1 (it suffices to take n such that hpIq contains an interval of the form
pi{2n , pi ` 1q{2n q for some i). Then
Solution It follows from Problems 2.16 and 3.22 that NW p f q is the ω -limit set of
each of its points. Together with Problem 2.9 this implies that NW p f q is a closed
set without isolated points. Indeed, NW p f q has no periodic points (because ρ p f q is
irrational) and since
which again by Proposition 2.4 is equivalent to f pyq P ω pxq. This shows that the set
ω pxq is f -invariant.
If ω pxq “ S1 , then there is nothing to prove. Now assume that ω pxq ‰ŤS1 and
write the open set U “ S1 zω pxq as a disjoint union of open intervals U “ jPZ I j .
We claim that
α pyq, ω pyq Ď ω pxq for every y P U.
This follows from the fact that the orbit of y visits each interval I j at most once.
Indeed, since U is f -invariant, if f k1 pyq P I j1 and f k2 pyq P I j2 , then
f k2 ´k1 : I j1 Ñ I j2 is a homeomorphism.
If j1 “ j2 , then by Problem 1.5 the homeomorphism has a fixed point, thus con-
tradicting the fact that ρ p f q is irrational. The uniqueness of the minimal set is a
consequence of Problem 2.16. Indeed, any other minimal set would intersect U and
so in particular it could not be the ω -limit set of each one of its points.
Solution Given x P S1 , let M “ ω pxq. Note that ω pxq ‰ ∅. It follows from Prob-
lem 3.13 that M is independent of x. Hence, M is the ω -limit set of each of its points
and so in view of Problems 2.16 and 3.24, it is the unique minimal set of f .
Note that the boundary of M is a closed subset of M that is also invariant (con-
nected components of S1 zM are taken by f to connected components of S1 zM
because M is f -invariant). Since M is a minimal set, this implies that
II.3 Low-Dimensional Dynamics 127
BM “ M or BM “ ∅,
Problem 3.26 Let f : S1 Ñ S1 be a C2 map. Show that its derivative has bounded
variation.
| f 2 pxq| ď K for x P S1 .
Solution Since
we obtain śn´1 1 i
|p f n q1 pxq| | f p f pxqq|
log n 1 “ log śi“0
|p f q pyq| n´1 1 i
i“0 | f p f pyqq|
ź
n´1
| f 1 p f i pxqq|
“ log (II.3.4)
i“0
| f 1 p f i pyqq|
ÿ
n´1
| f 1 p f i pxqq|
“ log .
i“0
| f 1 p f i pyqq|
128 II.3 Low-Dimensional Dynamics
| f 1 p f i pxqq|
log ď D1 p f i pIqq.
| f 1 p f i pyqq|
Taking the supremum in (II.3.4) in x and y yields the desired inequality.
then f j´i pIq X I ‰ ∅ and so f j´i pIq “ I since the set S1 zX is f -invariant (because X
is f -invariant). But by Problems 1.4 and 1.5, this implies that f has a periodic point,
which is impossible because f has an irrational rotation number. This contradiction
shows thatř the intervals In are disjoint. Therefore, denoting by |In | the length of In ,
we obtain nPZ |In | ď 1. So, necessarily |In | Ñ 0 when |n| Ñ 8.
On the other hand,
ż
|In | ` |I´n | “ p|p f n q1 pxq| ` |p f ´n q1 pxq|q dx
I
ż a
ě2 |p f n q1 pxq| ¨ |p f ´n q1 pxq| dx
I
?
ě 2 C|I|,
which contradicts the fact that |In | Ñ 0. Hence, every orbit of f is dense.
Problem 3.29 Consider the piecewise linear map f : r1, 3s Ñ r1, 3s with
(see Figure I.3.8). Show that f has period points with all periods.
Solution Note that f is continuous and that it has a periodic orbit with period 3
(namely t1, 2, 3u). It follows from Sharkovsky’s theorem (Theorem 3.10) that f has
periodic points with all periods.
II.3 Low-Dimensional Dynamics 129
Solution Note that f is continuous and that t2{9, 4{9, 8{9u is a periodic orbit of f
with period 3. Hence, by Sharkovsky’s theorem (Theorem 3.10), the map f has
periodic points with all periods.
Problem 3.31 Show that the set X “ r´2, ´1s Y r1, 2s does not have the property
that any continuous map f : X Ñ X satisfies Sharkovsky’s ordering in Definition 3.9.
Problem 3.32 Show that a closed set X with the property that any continuous map
f : X Ñ X satisfies Sharkovsky’s ordering in Definition 3.9 is connected.
Note that p and q are periodic points of f with period two, but that f has no fixed
points. This shows that the map does not satisfy Sharkovsky’s ordering.
x1 “ xy ´ x3 ă 0.
This implies that the first component of each solution whose orbit is contained in the
second quadrant is strictly decreasing and so the solution cannot be periodic (neither
constant nor periodic nonconstant).
130 II.3 Low-Dimensional Dynamics
r1 “ 4 ´ 3r cos θ ´ r4 ă 0
for any sufficiently large r. This implies that for any sufficiently small r1 ą 0 and
any sufficiently large r2 ą 0, the positive semiorbit of any point in the set
! a )
D “ px, yq P R2 : r1 ă x2 ` y2 ă r2 (II.3.5)
Solution Since θ 1 ą 0, it follows from the first equation that the origin is the only
critical point. Moreover,
II.3 Low-Dimensional Dynamics 131
and
rp1 ´ rq ă 0 for any sufficiently large r ą 0.
Hence, the positive semiorbit of any point in the set D in (II.3.5) with r1 sufficiently
small and r2 sufficiently large is contained in D. Again, we have the qualitative
behavior in Figure II.3.2. It follows from the Poincaré–Bendixson theorem (Theo-
rem 3.11) that the ω -limit set of any point in D is a periodic orbit.
and let
V px, yq “ x2 ` y2 and V9 px, yq “ ∇V px, yq ¨ f px, yq.
Show that:
1. V9 px, yq ď 0 whenever V px, yq ě 1{3;
2. V9 px, yq ě 0 whenever V px, yq ď 1{6;
3. there exists at least one periodic solution of the equation px1 , y1 q “ f px, yq in the
set (
D “ px, yq P R2 : 1{6 ď V px, yq ď 1{3 .
for all sufficiently small t ą 0. Similarly, by item 2, if p2 “ 1{6, then V9 ppq ě 0
and so
pxptq, yptqq2 ě 1{6
for all sufficiently small t ą 0. Therefore, any solution starting in D remains in D
for all positive time. Moreover, one can easily verify that 0 is the only critical point
and so D contains no critical points. Hence, it follows from the Poincaré–Bendixson
theorem (Theorem 3.11) that there exists at least one periodic orbit in D.
Problem 3.37 Show that any flow determined by the differential equation does not
have periodic solutions:
1. v1 “ f pvq for some continuous map f : Rn Ñ Rn , assuming that there exists a
function V : Rn Ñ R that is strictly decreasing along solutions;
2. v1 “ ´∇V pvq for some C1 function V : Rn Ñ R.
V pvpT qq ă V pvp0qq.
“ p f pxqy ` gpxqq dx ` y dy
żBD
“ ´y1 dx ` x1 dy
BD
żT
“ r´y1 ptqx1 ptq ` x1 ptqy1 ptqs dt “ 0.
0
This contradiction shows that there are no periodic solutions.
has the same sign almost everywhere on a simply connected open set U Ď R2 , then
the equation has no periodic orbits contained in U.
which is negative everywhere. Hence, it follows from Problem 3.39 that there are
no periodic solutions.
Chapter II.4
Hyperbolic Dynamics
Problem 4.1 Let Λ be a hyperbolic set with finitely many points. Show that Λ is
composed of periodic points.
Solution It suffices to note that since Λ is invariant, all its points are periodic and
so Λ is the union of finitely many periodic points.
Problem 4.2 Show that the second and third conditions in Definition 4.1 can be
replaced, respectively, by
1
dx f ´n v ě λ n v for x P Λ , v P E s pxq, n P N
c
and
1
dx f n v ě λ ´n v for x P Λ , v P E u pxq, n P N.
c
Solution We first consider the condition on the unstable space
is equivalent to
1
dx f ´n v ě λ n v for x P Λ , v P E s pxq, n P N.
c
Problem 4.3 Show that if Λ is a hyperbolic set for a diffeomorphism f , then for all
sufficiently large k P N the set Λ is a hyperbolic set for f k with constant c “ 1.
Solution Since Λ is hyperbolic set for f , there exist λ P p0, 1q, d ě 1 and a splitting
Tx M “ E s pxq ‘ E u pxq for each x P Λ such that
that is,
dx p f k qm v ď dpλ k qm x for v P E s pxq, m P N.
Similarly, it follows from (II.4.3) with n “ km that
dpλ k qm ď d m pλ k qm “ μ m ,
we conclude that for all sufficiently large k the set Λ is a hyperbolic set for the
diffeomorphism f k with constant c “ 1.
g´1 pΛ ˆ Λ q “ f ´1 pΛ q ˆ f pΛ q “ Λ ˆ Λ .
II.4 Hyperbolic Dynamics 137
associated with the hyperbolic set Λ . Although E u pxq and E s pxq are subspaces of
Tx R p , there is a linear isomorphism between E u pxq ‘ E s pxq and E u pxq ˆ E s pxq.
Hence, writing z “ px, yq and letting
we obtain
Tz pR p ˆ R p q “ pTx R p q ˆ pTy R p q
“ pE u pxq ‘ E s pxqq ˆ pE u pyq ‘ E s pyqq
» pE u pxq ˆ E s pyqq ‘ pE s pxq ˆ E u pyqq
“ E u pzq ‘ E s pzq.
Since ˆ ˙
dx f 0
dz g “ , (II.4.4)
0 dy f ´1
we have
dz gE u pzq “ dpx,yq gpE u pxq ˆ E s pyqq
“ dx f E u pxq ‘ dy f ´1 E s pyq
“ E u p f pxqq ‘ E s p f ´1 pyqq “ E u pgpzqq
and, analogously,
dz gE s pzq “ E s pgpzqq.
Finally, we establish exponential bounds along the spaces E u pzq and E s pzq. Since
Λ is a hyperbolic set, there exist λ P p0, 1q and c ą 0 such that for each
we have
dx f n u ď cλ n u and dy f ´n v ď cλ n v
for all n P N. On the other hand, by (II.4.4), we have
ˆ n ˙
d f 0
dz gn “ x
0 dy f ´n
and so, considering the norm pu, vq “ u ` v, we obtain
we have
dx f ´n u ď cλ n u and dy f n v ď cλ n v
for n P N, which gives
Problem 4.5 Let A be a p ˆ p matrix with real entries and without eigenvalues of
modulus 1. Show that there exist a splitting R p “ E s ‘ E u and constants λ P p0, 1q
and c ą 0 such that
and
A´n v ď cλ n v for v P E u , n P N.
pA ´ τ Idqk v “ 0 (II.4.5)
AE s “ E s and AE u “ E u .
II.4 Hyperbolic Dynamics 139
0 ε
¨ ˛
0
˚ .. .. ‹
. . ‹
N“˚ ‹.
˚
. . . ε‚
˚ ‹
˝
0 0
Moreover, let ¨ ˛
R1 0
S´1 BS “ ˝ . . . ‚
˚ ‹
0 Rk
be a Jordan canonical form of the matrix B “ A|F s , for some invertible matrix S “
Spε q with complex entries. Then
¨ n ˛
R1 0
Bn “ S ˝ . . . ‚S´1
˚ ‹
0 Rnk
and
n´1
ÿˆ ˙
n n´i i
Rnj “ λj N .
i“0
i
Therefore,
n´1
ÿˆ ˙
n
Bn ď S ¨ S´1 σ pBqn´i Ni
i“0
i
n ˆ ˙
ÿ n
ď D1 σ pBqn´i pD2 ε qi
i“0
i
˘n
“ D1 σ pBq ` D2 ε
`
A similar argument applies to A´n |E u . Summing up, there exist λ P p0, 1q and c ą 0
as desired.
Problem 4.6 For the Smale horseshoe Λ (see Definition 4.3), show that for any
two-sided sequence ω “ p¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q in t1, 2uZ the set
f ´nVin
č
Λω “
nPZ
(see Figure II.4.1). Due to the contraction and expansion of f on H1 Y H2 (see Def-
inition 4.3), each set Rn pω q is a square of side an and so diam Rn pω q Ñ 0 when
n Ñ 8. This implies that each intersection
f ´nVin
č
Λω “
nPZ
č
“ Rn pω q
nPN
contains at most one point. On the other hand, since Rn pω q is a decreasing sequence
of nonempty closed sets, each intersection is nonempty. This shows that
Problem 4.7 For the Smale horseshoe Λ , show that the map H : t1, 2uZ Ñ Λ
defined by
f ´nVin
č
Hp¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q “
nPZ
f ´n pV1 YV2 q
č
Λ“
nPZ
f ´nVin
ď č
“
ω Pt1,2uZ nPZ
ď
“ Hpω q
ω Pt1,2uZ
Vim pω q XVim pω 1 q “ ∅.
Hence,
˜ ¸ ˜ ¸
1 ´n ´n
č č
Hpω q X Hpω q “ f Vin pω q X f Vin pω 1 q “ ∅,
nPZ nPZ
Problem 4.8 For the Smale horseshoe Λ , show that for each m P N the number of
m-periodic points of the map f |Λ is at least 2m .
Solution Consider the sequence ω ppq “ p¨ ¨ ¨ ppp ¨ ¨ ¨ q P t1, 2uZ obtained from re-
peating indefinitely a vector p P t1, 2um . For ω “ ω ppq, the set Λω in Problem 4.6
satisfies
f ´n`mVin
č
f mΛω “
nPZ
f ´n`mVin´m “ Λω .
č
“
nPZ
Hence, the unique element x p of Λω ppq (see Problem 4.6) is an m-periodic point.
On the other hand, by Problem 4.7, for p, q P t1, 2um with p ‰ q we have x p ‰ xq .
Therefore, since the set t1, 2um has exactly 2m elements, the map f |Λ has at least 2m
m-periodic points.
ϕ1 ă ψ1 ă ϕ2 ă ψ2
142 II.4 Hyperbolic Dynamics
and
ϕ1 ă ψ1 ă ϕ2 ă ψ2
(see Figure I.4.5). Moreover, let f be a C1 diffeomorphism on an open neighborhood
of the square r0, 1s2 such that f pHi q “ Vi for i “ 1, 2 and
and some C1 functions g, h defined on compact subsets of R. Show that if sup |g1 | ă 1
and inf |h1 | ą 1, then č
Λ“ f n pH1 Y H2 q
nPZ
f ´1 Λ “ f ´1
č
f n pH1 Y H2 q
nPZ
č
“ fn´1
pH1 Y H2 q “ Λ .
nPZ
Let E s px, yq and E u px, yq be, respectively, the horizontal and vertical axes. For each
px, yq P Λ we consider the decomposition
and
dpx,yq f ´1 v ď pinf |h1 |q´1 v for v P E u px, yq.
Since sup |g1 | ă 1 and inf |h1 | ą 1, we conclude that Λ is a hyperbolic set for f with
constants c “ 1 and
Problem 4.10 Show that if a compact connected manifold M is a hyperbolic set for
a diffeomorphism f : M Ñ M, then all stable spaces have the same dimension.
II.4 Hyperbolic Dynamics 143
Solution Since the stable and unstable spaces E s pxq and E u pxq vary continuously
with x, their dimensions are locally constant. More precisely, for each x P M there
exists an open neighborhood Ux of x such that
for all y P Ux . Since M is compact, there exists a finite cover of M by sets Ux . Hence,
it follows from the connectedness of M, that all stable spaces have the same dimen-
sion.
Alternatively, let
Ak “ x P M : dim E s pxq “ k
(
and
Bk “ x P M : dim E s pxq ‰ k
(
M “ Ak Y Bk and Ak X Bk “ ∅.
Hence, dim E s pxq “ k and so x P Ak , which shows that Ak is closed. Now take xn P Bk
such that xn Ñ x P M when n Ñ 8. Since xn P Bk , we have
Hence, it follows from (II.4.8) that dim E s pxq ‰ k and so x P Bk . This shows that
Bk is closed. Since M is connected, we conclude that for each k either Ak “ ∅ or
Bk “ ∅. Hence, there exists k P t0, 1, . . . , dim Mu such that Ak “ M.
Problem 4.11 Let f : R Ñ R be a C1 map. Show that if there exist λ P p0, 1q and
c ą 0 such that |p f n q1 pxq| ď cλ n for all x P R and n P N, then
1
lim sup log | f n pIq| ă 0
nÑ8 n
Solution We have
ż y
dp f n pxq, f n pyqq “ p f n q1 pzq dz ď cλ n dpx, yq (II.4.9)
x
| f n pIq| ď cλ n |I|
144 II.4 Hyperbolic Dynamics
and so
1
lim sup log | f n pIq| ă 0.
nÑ8 n
Problem 4.12 Show that there exists no diffeomorphism f : S2 Ñ S2 for which the
whole S2 is a hyperbolic set with dim E u pxq “ t0u for all x P S2 .
Solution Assume that dim E u pxq “ t0u for all x P S2 . Then one can proceed as in
(II.4.9) to conclude that
Determine whether the set X of all bounded functions f : R Ñ R with the property
that f p0q ` f p1q “ 0 is a complete metric space with the distance d.
Solution We show that X is complete with the distance d. Let p fn qnPN be a Cauchy
sequence in X. Note that it is also a Cauchy sequence in the set of all bounded con-
tinuous functions equipped with the distance d, which is a complete metric space.
Hence, there exists a continuous function f : R Ñ R such that dp fn , f q Ñ 0 when
n Ñ 8. Moreover, it follows from the inequality
| fn pxq ´ f pxq| ď dp fn , f q
that
f pxq “ lim fn pxq for each x P R.
nÑ8
Problem 4.14 Show that the set of all C1 functions f : ra, bs Ñ R is not a complete
metric space with the distance
Solution We show that the limit of C1 functions need not be differentiable. For an
example, take ra, bs “ r´1, 1s and for each n P N let
II.4 Hyperbolic Dynamics 145
b
fn pxq “ px ´ 1{2q2 ` 1{n for x P r´1, 1s
f pxq “ |x ´ 1{2|.
f1
f 10
1 1 x
gn pxq “ 2 x
n 2 if ´ 1n ď x ď 1n ,
x ´ 2n1
if x ě 1n
’
%
f1
f2
Since the set tpx, vq : x P Λ , v P Sx u is compact, there exists μ P p0, 1q such that
1
dx f v ě μ ´1
v1
and so
dx f v1 ě μ ´1 v1 for v P Cu pxqzt0u.
This establishes the desired property.
A “ S1 ˆ px, yq P R2 : x2 ` y2 ď 1
(
Solution We have
´1 1 ¯2 ´ 1 1 ¯2
x ` cosp2πθ q ` y ` sinp2πθ q
5 2 5 2
1 2 1 2 1 1 1
“ x ` y ` ` x cosp2πθ q ` y sinp2πθ q (II.4.11)
25 25 4 5 5
1 2 2 1
ď px ` y q ` ` ă 1
2
25 5 4
and so f pAq Ĺ A. To show that f is one-to-one, assume that
f prθ1 s, x1 , y1 q “ f prθ2 s, x2 , y2 q.
Then
2θ1 ” 2θ2 mod 1,
1 1 1 1
x1 ` cosp2πθ1 q “ x2 ` cosp2πθ2 q, (II.4.12)
5 2 5 2
1 1 1 1
y1 ` sinp2πθ1 q “ y2 ` sinp2πθ2 q.
5 2 5 2
It follows from the first relation that either θ1 “ θ2 or θ1 “ θ2 ` 1{2.
When θ1 “ θ2 it follows from the second and third relations in (II.4.12) that
x1 “ x2 and y1 “ y2 . Now assume that θ1 “ θ2 ` 1{2. In this case the second and
third relations in (II.4.12) become
1 1
px1 ´ x2 q “ cosp2πθ1 q and py1 ´ y2 q “ sinp2πθ1 q.
5 5
Therefore,
148 II.4 Hyperbolic Dynamics
Problem 4.18 For the map f |A in Problem 4.17, show that Λ “ nPN f n pAq is a
Ş
hyperbolic set for f .
Solution The map f is continuous on the compact set A and so f pAq is also com-
pact. Moreover, it follows from (II.4.11) that f pAq is contained in the interior of A.
Therefore, Λ is the intersection of a decreasing sequence of compact sets and so it
is nonempty and compact. One can also verify that f is a diffeomorphism on some
open neighborhood of Λ . Finally,
˜ ¸
´1 ´1
č
f Λ“f f pAq
n
nPN
č č
“ f pAq “ A X
n
f n pAq
ně0 nPN
č
“ f pAq “ Λ
n
nPN
and so Λ is f -invariant.
Now we consider the splitting
1 1 1 1
u “ 2u, v “ ´ u sinp2πθ q ` v, w “ u cosp2πθ q ` w.
2 5 2 5
Then
u2 sin2 p2πθ q 1 1
v2 ` w2 “ ` v2 ´ uv sinp2πθ q
4 25 5
u2 cos2 p2πθ q 1 1
` ` uw cosp2πθ q ` w2
4 5 25
u2 1 1
“ ` pv2 ` w2 q ` upw cosp2πθ q ´ v sinp2πθ qq
4 25 5
u2 1 1 2 1
ď ` ¨ u ` upw cosp2πθ q ´ v sinp2πθ qq.
4 25 4 5
We have
w cosp2πθ q ´ v sinp2πθ q ď w ` v
? a
ď 2 v2 ` w2
1
ď? u
2
and so ˆ ˙
1 1 1 1
v2 ` w2 ď ` ` ? u2 ă u2 .
4 100 5 2 4
This shows that
dz fCu pzq Ĺ Cu p f pzqq (II.4.15)
(see Figure II.4.4).
Moreover,
4
dz f pu, v, wq2 ą 4u2 ě pu, v, wq2 (II.4.16)
1 ` 1{4
for pu, v, wq P R3 with u ‰ 0. Now consider a 2-dimensional space
150 II.4 Hyperbolic Dynamics
dz f Cu (z)
Cu ( f ( z))
Fig. II.4.4 The cone Cu p f pzqq and its image dz fCu pzq.
for each z P Λ . It exists because by (II.4.15) the set Gu pzq is the intersection of
a decreasing sequence of closed sets and so it is nonempty. Moreover, since the
closed unit ball S3 in R3 is compact, the set Gu pzq X S3 is also compact and so any
sequence of linearly independent vectors
m
d f ´n pzq f nCu p f ´n pzqq,
č
1 , v2 P
vm m
n“0
pu, v, wq ´ vu “ vs P E s pzq,
we also have
1
dz f ´n ppu, v, wq ´ vu q ě pu, v, wq ´ vu Ñ `8
5n
when n Ñ 8 (because vs ‰ 0). This contradiction shows that E u pzq “ Gu pzq. Since
II.4 Hyperbolic Dynamics 151
dz f Gu pzq “ Gu p f pzqq
and so also
dz f E u pzq “ E u p f pzqq.
Putting everything together, we conclude that Λ is a hyperbolic set with stable and
unstable spaces F s pzq and E u pzq at each z P Λ (see (II.4.14) and (II.4.16)).
Problem 4.19 Show that if Tn is a hyperbolic set for an automorphism of the torus
TA : Tn Ñ Tn , then TA has positive topological entropy.
k =2
k=1
k=0
Fig. II.4.5 Sets TA´k BpTAk ppi q, ε q for k “ 0, 1, 2 and their intersection B2 ppi , ε q (in gray).
(see Figure II.4.5). It follows from results in linear algebra that for each sufficiently
small δ ą 0 there exists C ą 0 such that
152 II.4 Hyperbolic Dynamics
p|λi |´1 ` δ qm ε n .
ź
the n-volume of the ball Bm ppi , ε q is at most D
|λi |ą1
λ “ mint|λi | : |λi | ą 1u ą 1.
Then
Mpm, ε q ě D´1 p|λi |´1 ` δ q´m ε ´n
ź
|λi |ą1
ě D pλ ´1 ` δ q´dm ε ´n .
´1
hpTA q ě d log λ ą 0.
Problem 4.20 For the maps f , g : R Ñ R defined by f pxq “ ax and gpxq “ bx, with
a, b ą 1, show that there exists a homeomorphism h : R Ñ R such that h ˝ f “ g ˝ h.
h ˝ f n “ gn ˝ h
One can easily verify that h is a homeomorphism, with inverse h´1 : R Ñ R given
by $
&x
’ log a{ log b if x ą 0,
´1
h pxq “ 0 if x “ 0,
´|x|log a{ log b if x ă 0.
’
%
Moreover,
hp f pxqq “ paxqlog b{ log a
“ elog b xlog b{ log a
“ bxlog b{ log a
“ gphpxqq
for x ą 0. One can show analogously that the same identity also holds for x ď 0.
and $
&x
’ log 4{ log 3 if x ą 0,
h2 pxq “ 0 if x “ 0,
´|x|log 4{ log 3 if x ă 0
’
%
h1 ˝ f2 “ f5 ˝ h1 and h2 ˝ f3 “ f4 ˝ h2 ,
We have
hp f px, yqq “ p2xqlog 5{ log 2 , p3yqlog 4{ log 3
` ˘
“ gphpx, yqq
for x ą 0. One can show analogously that the same identity holds for x ď 0 and so
h ˝ f “ g ˝ h.
Problem 4.22 Determine the 2-periodic points of the automorphism of the torus TA
for the matrix ˆ ˙
21
A“ .
11
Problem 4.23 Determine the 2-periodic points of the endomorphism of the torus TB
for the matrix ˆ ˙
31
B“ .
11
II.4 Hyperbolic Dynamics 155
Problem 4.24 Show that the vector Xpxq “ dtd ϕt pxqˇt“0 in Definition 4.9 is neither
ˇ
Solution Since pϕt qtPR is a flow, we have ϕt˝s “ ϕt ˝ ϕs for all t, s P R. Hence,
d d
ϕt`s pxqˇt“0 “ ϕt pϕs pxqqˇt“0 “ Xpϕs pxqq
ˇ ˇ
dt dt
and
d
Xpϕs pxqq “ ϕt`s pxqˇt“0
ˇ
dt
d
“ ϕs pϕt pxqqˇt“0
ˇ
dt
d
“ dϕt pxq ϕs ˇt“0 ϕt pxqˇt“0
ˇ ˇ
dt
“ dx ϕs Xpxq.
Therefore,
dx ϕs Xpxq “ Xpϕs pxqq,
which yields the desired property.
Solution Since the closed unit sphere in R p is compact, the sequence vm has sub-
limits. Moreover, since vm P E s pxm q, there exist constants c ą 0 and λ P p0, 1q such
that
dxm ϕt vm ď cλ t vm
for all m P N and t ě 0. Letting m Ñ 8, we obtain
dx ϕt v ď cλ t v
156 II.4 Hyperbolic Dynamics
for all t ě 0 and any sublimit v of the sequence vm . This implies that v has no
component in E 0 pxq ‘ E u pxq and so v P E s pxq.
Problem 4.26 For a sequence xm as in Problem 4.25, show that there exists m P N
such that
for any p, q ą m.
Solution Since the dimensions dim E s pxm q and dim E u pxm q take only finitely many
values, there exists a subsequence ym of xm such that the numbers dim E s pym q and
dim E u pym q are independent of m. Let k “ dim E s pym q (which is independent of m)
and let
v1m , . . . , vkm P E s pym q
be an orthonormal basis for E s pym q. Since the closed unit sphere in R p is com-
pact, the sequence pv1m , . . . , vkm q has sublimits and each sublimit pv1 , . . . , vk q is an
orthonormal set. It follows from Problem 4.25 that
Comparing with (II.4.20), we conclude that l “ k. This establishes the desired prop-
erty.
Problem 4.27 Find the stable and unstable invariant manifolds at the origin for the
differential equation #
x1 “ x,
y1 “ ´y ` x2 .
Solution Note that the origin is the only critical point of the differential equation.
Since the matrix of the linearization at the origin
#
x1 “ x,
y1 “ ´y,
II.4 Hyperbolic Dynamics 157
has eigenvalues ˘1, the origin is a hyperbolic fixed point. The stable and unstable
spaces are, respectively,
(the vertical and horizontal axes). Moreover, for each b P R the set
x2
" *
b
px, yq P R : y ´ “
2
(II.4.21)
3 x
is invariant under the flow determined by the differential equation. Indeed,
˙1
x2 b ˇˇ
ˆ ˆ ˙
2x 1 b 1 ˇˇ
y´ ´ “ ´y ` x ´ x ` 2 x ˇ 2
2
3 x y´x {3“b{x 3 x y´x {3“b{x
ˇ 2
2
ˆ ˙ˇ
x b ˇ
“ ´y ` x2 ´ 2 `
3 x y´x2 {3“b{x
ˇ
x2 b x2 b
“´ ´ ` x2 ´ 2 ` “ 0.
3 x 3 x
For b “ 0 the set in (II.4.21) is tangent to E u at the origin. Hence, it is the unstable
invariant manifold at the origin.
Furthermore, the line x “ 0 is invariant under the flow determined by the differ-
ential equation and is tangent to E s at the origin. Hence, it is the stable invariant
manifold at the origin.
Problem 4.28 Find the stable and unstable invariant manifolds at the origin for the
differential equation #
x1 “ xpa ´ yq,
y1 “ ´y ` x2 ´ 2y2 ,
for each given a ą 0.
Solution The origin is a critical point of the differential equation. The linearization
at the origin is #
x1 “ ax,
y1 “ ´y
with a ą 0, and so p0, 0q is a hyperbolic fixed point for each a ą 0. The stable and
unstable spaces are, respectively,
x2
x“0 and y“
1 ` 2a
158 II.4 Hyperbolic Dynamics
are invariant under the flow determined by the differential equation. Since they are
tangent, respectively, to E s and E u at the origin, they are the stable and unstable
invariant manifolds at the origin.
h ˝ ϕt “ ψt ˝ h for t P R.
Solution We have
ϕt pxq “ eat x and ψt pxq “ ebt x.
Moreover, h is a homeomorphism, with inverse h´1 : R Ñ R given by
$
&x
’ a{b if x ą 0,
´1
h pxq “ 0 if x “ 0,
´|x|a{b if x ă 0.
’
%
Finally, we have
hpϕt pxqq “ hpeat xq
˘b{a
“ e3t x
`
“ ebt xb{a
and
ψt phpxqq “ ψt pxb{a q “ ebt xb{a
for x ą 0, with analogous identities for x ď 0.
Problem 4.30 Let A and B be n ˆ n matrices. Show that if there exists a topological
conjugacy between the flows determined by the differential equations
v1 “ Av and v1 “ Bv
that is a diffeomorphism, then there exists also a topological conjugacy between the
flows that is a linear map.
II.4 Hyperbolic Dynamics 159
which shows that the matrix C is invertible. Now we consider the invertible linear
map g : Rn Ñ Rn given by gpvq “ Cv. It follows from (II.4.22) that
g ˝ ϕt “ ψt ˝ g.
Problem 4.31 Let A and B be n ˆ n matrices. Show that if there exists a topologi-
cal conjugacy between the flows determined by the differential equations in Prob-
lem 4.30 that is a linear map, then the matrices A and B are conjugate, that is, there
exists an invertible n ˆ n matrix C such that A “ C´1 BC.
Solution Assume that (II.4.23) holds for some invertible linear map g : Rn Ñ Rn
and all t P R and v P Rn . Writing gpvq “ Cv for some invertible n ˆ n matrix C, we
obtain
CeAt “ eBt C.
Finally, taking t “ 0 yields the identity CA “ BC, which establishes the desired
property.
Problem 4.32 Let ϕt and ψt be the flows determined, respectively, by the linear
equations with matrices
´1 0 ´1 1
ˆ ˙ ˆ ˙
A“ and B “ .
0 ´1 0 ´1
Therefore,
x2
ˆ ˙
hpx, yq “ x, y `
3
is a topological conjugacy between the flows determined by the equation and its
linearization at the origin.
Solution The solution of the equation with pxp0q, yp0qq “ px0 , y0 q is given by
x02 t
xptq “ x0 e´t , yptq “ y0 et ` pe ´ e´2t q (II.4.24)
3
with t P R. The linearization at the origin of the equation is
II.4 Hyperbolic Dynamics 161
#
x1 “ ´x,
y1 “ y,
with t P R. By (II.4.24) and (II.4.25), the flows determined by the two equations are,
respectively,
x2
ˆ ˙
ϕt px0 , y0 q “ x0 e´t , y0 et ` 0 pet ´ e´2t q
3
and
ψt px0 , y0 q “ px0 e´t , y0 et q.
On the other hand, h is a homeomorphism, with inverse given by
´ x2 ¯
h´1 px, yq “ x, y ´ .
3
Since
e´t 0
ˆ ˙ˆ ˙
x0
ψt phpx0 , y0 qq “
0 et y0 ` x02 {3
x0 e´t
ˆ ˙
“
e py0 ` x02 {3q
t
and
x2
ˆ ˙
hpϕt px0 , y0 qq “ h x0 e´t , y0 et ` 0 pet ´ e´2t q
3
2
px0 e´t q2
ˆ ˙
x
“ x0 e´t , y0 et ` 0 pet ´ e´2t q `
3 3
ˆ
x 2 ˙
“ x0 e´t , et py0 ` 0 q ,
3
we conclude that
h ˝ ϕt “ ψt ˝ h for t P R. (II.4.26)
xn
ˆ ˙
hpx, yq “ x, y ´
n`1
is a topological conjugacy between the flows determined by the equation and its
linearization at the origin.
Solution The solution of the equation with pxp0q, yp0qq “ px0 , y0 q is given by
x0n
xptq “ x0 et , yptq “ y0 e´t ´ pe´t ´ ent q (II.4.27)
n`1
with t P R. The linearization at the origin of the equation is
#
x1 “ x,
y1 “ ´y,
with t P R. By (II.4.27) and (II.4.28), the flows determined by the two equations are,
respectively,
xn
ˆ ˙
ϕt px0 , y0 q “ x0 et , y0 e´t ´ 0 pe´t ´ ent q
n`1
and
ψt px0 , y0 q “ px0 et , y0 e´t q.
Moreover, one can easily verify that h is a homeomorphism. Since
ˆ t ˙ˆ ˙
e 0 x0
ψt phpx0 , y0 qq “
0 e´t y0 ´ x0n {pn ` 1q
x0 et
ˆ ˙
“ ´t
e py0 ´ x0n {pn ` 1qq
and
xn
ˆ ˙
hpϕt px0 , y0 qq “ h x0 et , y0 e´t ´ 0 pe´t ´ ent q
n`1
xn px0 et qn
ˆ ˙
“ x0 et , y0 e´t ´ 0 pe´t ´ ent q ´
n`1 n`1
ˆ
x n ˙
“ x0 et , e´t py0 ´ 0 q ,
n`1
we find that property (II.4.26) holds.
II.4 Hyperbolic Dynamics 163
and let
T1,c pzq “ T´ 1 c ¯ pzq “ z ` c for c P R,
01
T2,c pzq “ T´ c 0
¯ pzq “ c2 z for c P Rzt0u,
0 1{c
1
T3,c pzq “ T´ 0 ´1 ¯ pzq “ ´ for c P R.
1 c z`c
Show that #
T1,ab ˝ T2,a if c “ 0,
TA “
T1,a{c ˝ T2,1{c ˝ T3,d{c if c ‰ 0.
Solution If c “ 0, then
az ` b
TA pzq “
d
a b
“ z`
d d
“ a2 z ` ab
“ pT1,ab ˝ T2,a qpzq.
On the other hand, if c ‰ 0, then
cpaz ` bq
TA pzq “
cpcz ` dq
caz ` cb
“ 2
c z ` cd
caz ` ad ´ ad ` bc
“
c2 z ` cd
apcz ` dq 1
“ ´
cpcz ` dq cpcz ` dq
a 1{c2
“ ´
c z ` d{c
“ pT1,a{c ˝ T2,1{c ˝ T3,d{c qpzq.
Problem 4.36 Show that any straight line or circle is determined by an equation of
the form
Azz ` Bz ` Bz `C “ 0 with A,C P R, B P C,
respectively, with A “ 0 and with A ‰ 0 and AC ă BB.
164 II.4 Hyperbolic Dynamics
When A “ 0, clearly (II.4.29) determines a straight line. On the other hand, when
A ‰ 0 one can rewrite (II.4.29) in the form
´ B1 ¯2 ´ B2 ¯2 B21 B22
A x` `A y´ ´ ´ `C “ 0,
A A A A
that is,
´ B1 ¯2 ´ B2 ¯2 B21 B22 C BB C
x` ` y´ “ 2´ 2´ “ 2 ´ . (II.4.30)
A A A A A A A
In order that (II.4.30) determines a circle we must have
BB C
´ ą 0.
A2 A
Problem 4.37 Show that the map T pzq “ 2z takes straight lines into straight lines
and circles into circles.
z “ T ´1 pwq “ w{2
satisfies ?
0 “ Aww{4 ` Bw{2 ` Bw{ 2 `C
1
“ A1 ww ` B1 w ` B w `C1 ,
with
A1 “ A{4, B1 “ B{2, C1 “ C.
Note that A1 ,C1 P R. Moreover, A1 “ 0 if and only if A “ 0 and so T maps straight
lines into straight lines. Finally, if A ‰ 0 and AC ă BB, then
1
A1C1 “ AC{4 ă BB{4 “ B1 B
Problem 4.38 Show that the map T pzq “ ´1{pz ` 1q takes straight lines and circles
into straight lines or circles.
z “ T ´1 pwq “ ´1 ´ 1{w
II.4 Hyperbolic Dynamics 165
satisfies
´ 1 ¯´ 1¯ ´ 1¯ ´ 1¯
0 “ A ´1 ´ ´1 ´ ` B ´1 ´ ` B ´1 ´ `C
w w w w
1
“ Apw ` 1qpw ` 1q ` Bp´ww ´ wq ` Bp´ww ´ wq `C
ww
1 1
“ pA1 ww ` B1 w ` B w `C1 q,
ww
with
A1 “ A ´ B ´ B, B1 “ A ´ B, C1 “ C ` A.
Note that A1 ,C1 P R. If A1 “ 0, then T maps straight lines and circles into straight
lines. On the other hand, if A1 ‰ 0, then T maps straight lines and circles into circles
1
(clearly A1C1 ă B1 B since T is bijective).
Problem 4.39 Show that the components of the geodesic flow ϕt : SH Ñ SH given
by
ϕt pz, vq “ pγ ptq, γ 1 ptqq
(see Definition 4.10), satisfy
2 Re γ 1 ptq Im γ 1 ptq
pRe γ 1 ptqq1 “
Im γ ptq
and
pIm γ 1 ptqq2 ´ pRe γ 1 ptqq2
pIm γ ptqq1 “ .
Im γ ptq
Solution Let
az ` b
T pzq “ .
cz ` d
We have
aiet ` b
γ ptq “ T piet q “ .
ciet ` d
Since ad ´ bc “ 1, we obtain
ipbc ´ adqet
γ 1 ptq “
pcet ´ idq2
´iet
“
pcet ´ idq2
iet
“ .
pciet ` dq2
Hence,
166 II.4 Hyperbolic Dynamics
iet
γ 1 ptq “
pciet ` dq2
iet pd ´ ciet q2
“ (II.4.31)
pd 2 ` c2 e2t q2
2cde2t et pd 2 ´ c2 e2t q
“ ` i ,
pd 2 ` c2 e2t q2 pd 2 ` c2 e2t q2
while
aiet ` b paiet ` bqpd ´ ciet q
γ ptq “ “
ciet ` d d 2 ` c2 e2t
bd ` ace 2t pad ´ bcqet
“ 2 `i 2 (II.4.32)
d `c e2 2t d ` c2 e2t
bd ` ace2t et
“ 2 `i 2 .
d `c e2 2t d ` c2 e2t
It follows from (II.4.31) and (II.4.32) that
2cde2t et pd 2 ´ c2 e2t q
Re γ 1 ptq “ , Im γ 1 ptq “ , (II.4.33)
pd 2 ` c2 e2t q2 pd 2 ` c2 e2t q2
and
bd ` ace2t et
Re γ ptq “ , Im γ ptq “ . (II.4.34)
d 2 ` c2 e2t d 2 ` c2 e2t
Therefore,
2cde2t
ˆ ˙
1 d 1
pRe γ ptqq “
dt pd 2 ` c2 e2t q2
4cdpd 2 ´ c2 e2t qe2t
“
pd 2 ` c2 e2t q3
and
2 Re γ 1 ptq Im γ 1 ptq 2cde2t et pd 2 ´ c2 e2t q ´ et ¯´1
“2 2 ¨
Im γ ptq pd ` c2 e2t q2 pd 2 ` c2 e2t q2 d 2 ` c2 e2t
4cdpd 2 ´ c2 e2t qe2t
“ .
pd 2 ` c2 e2t q3
Finally, we also have
d ´ et pd 2 ´ c2 e2t q ¯
pIm γ 1 ptqq1 “
dt pd 2 ` c2 e2t q2
et pd 4 ´ 6c2 d 2 e2t ` c4 e4t q
“
pd 2 ` c2 e2t q3
II.4 Hyperbolic Dynamics 167
and
pIm γ 1 ptqq2 ´ pRe γ 1 ptqq2 e2t pd 4 ´ 6c2 d 2 e2t ` c4 e4t q ´ et ¯´1
“
Im γ ptq pd 2 ` c2 e2t q4 d 2 ` c2 e2t
et pd 4 ´ 6c2 d 2 e2t ` c4 e4t q
“ .
pd 2 ` c2 e2t q3
This establishes the desired identities.
Problem 4.40 For the geodesic flow ϕt : SH Ñ SH, show that the quantities
|γ 1 ptq| Re γ 1 ptq
H1 “ and H2 “
Im γ ptq pIm γ ptqq2
are independent of t.
Re γ 1 ptq 2dce2t
H2 “ “ “ 2cd.
pIm γ ptqq2 e2t
Chapter II.5
Symbolic Dynamics
Problem 5.1 Show that for each ω P Σk` and r ą 0 the open ball
(
Bdβ pω , rq “ ω 1 P Σk` : dβ pω 1 , ω q ă r
is a cylinder, that is, there exist integers i1 , . . . , in P t1, . . . , ku with n “ nprq such that
(
Bdβ pω , rq “ Ci1 ¨¨¨in :“ ω P Σk` : i j pω q “ i j for j “ 1, . . . , n .
1 log r
β ´npω ,ω q ă r ðñ npω , ω 1 q ą ´
log β
Problem 5.2 Show that any cylinder Ci1 ¨¨¨in (see Problem 5.1) is simultaneously
open and closed.
i j pω 1 q “ i j for j P t1, . . . , nu
Problem 5.3 Given β , β 1 ą 1, show that the distances dβ and dβ 1 generate the same
topology on Σk` .
Solution It suffices to verify that given ω P Σk` and r ą 0, for each ω 1 P Bdβ pω , rq
there exists r1 ą 0 such that
Indeed, one can then interchange the roles of β and β 1 to also conclude that given
ω P Σk` and r ą 0, for each ω 1 P Bdβ 1 pω , rq there exists r1 ą 0 such that
This implies that the two distances dβ and dβ 1 generate the same topology on Σk` .
Take ω 1 P Bdβ pω , rq. By Problem 5.1 we have
for some integers i1 , . . . , inprq P t1, . . . , ku. Therefore, ω 1 P Bdβ pω , rq if and only if
in pω 1 q “ in for n ď nprq and so
ÿ8
1
dpω , ω 1 q “ |i pω q ´ in pω 1 q|
n n
n“1
2
Bd pω 1 , r1 q Ď Bdβ pω , rq (II.5.2)
II.5 Symbolic Dynamics 171
For the first property note that, by Problem 5.1, ω 1 P Bdβ pω , rq, that is, dβ pω , ω 1 q ă r
if and only if there exists nprq P N such that
ÿ8
1
dpω 2 , ω q “ |i pω 2 q ´ in pω q|
n n
n“1
2
8
ÿ 1 k´1
ď n
pk ´ 1q “ m ă r
n“m`1
2 2
provided that r2 is sufficiently small. This shows that ω 2 P Bd pω , rq, which estab-
lishes property (II.5.3).
Then
1 1 1
dpω , ω 2 q “ , dpω , ω 1 q “ , dpω 1 , ω 2 q “ .
n1 n2 n3
172 II.5 Symbolic Dynamics
1 1
dpω 2 , ω 1 q ě ě ą r1
n nprq
1
dpω 2 , ω q ă ăr
m
provided that r2 is sufficiently small. This shows that ω 2 P B d pω , rq, which estab-
lishes property (II.5.3).
Problem 5.6 Show that the distances d and d in Problems 5.4 and 5.5 are not equiv-
alent.
Solution Consider the constant sequence ω “ p111 ¨ ¨ ¨ q and for each m P N define
a new sequence ωm P Σk` by
#
1 if j ă m,
i j pωm q “
2 if j ě m.
Note that the smallest positive integer j such that i j pω q ‰ i j pωm q is npω , ωm q “ m.
Now assume that d and d are equivalent. Then there exists c ą 0 such that
or, equivalently,
1 1 1
c´1 ď ď c m´1 for m P N.
2m´1 m 2
II.5 Symbolic Dynamics 173
Solution First note that the open balls in the distance d are again cylinders. Indeed,
ω 1 P B d pω , rq if and only if
1 1
ăr ðñ npω , ω 1 q ą
npω , ω 1 q r
To show that Σk` is compact, we first equip t1, . . . , ku with the topology in which
all subsets of t1, . . . , ku are open. Then the product topology on Σk` “ t1, . . . , kuN
coincides with the topology generated by the cylinders. It follows from Tychonoff’s
theorem that pΣk` , dq is a compact topological space because it is the product of
compact topological spaces, with the product topology.
Solution We first give an argument using the fact that pΣk` , dq is a compact metric
space. Let pωm qmPN be a Cauchy sequence in Σk` . Since Σk` is compact, there exists
a subsequence kn with ωkn Ñ ω when n Ñ 8. Given ε ą 0, there exists N P N such
that
dpωm , ωkn q ă ε {2 for m, kn ą N
and
dpωkn , ω q ă ε {2 for kn ą N.
Therefore,
i p pω q “ i p pωNppq q for p P N.
Problem 5.9 Show that the shift map σ |Σk is topologically transitive.
with j´m , . . . , jm P t1, . . . , ku, are open balls in the distance dβ and that they generate
the topology. Thus, it suffices to consider these sets in the notion of topological
transitivity. Given two sets D j´m ¨¨¨ jm and Dk´m ¨¨¨km , we claim that there exists p P N
such that
σ ´p D j´m ¨¨¨ jm X Dk´m ¨¨¨km ‰ ∅.
Given ω “ p¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q P D j´m ¨¨¨ jm , take p ě 2m ` 2 and
with center at k0 , for some l1 , . . . , l p´2m´1 P t1, . . . , ku. Then σ p pω q P D j´m ¨¨¨ jm and
so
ω 1 P σ ´p D j´m ¨¨¨ jm X Dk´m ¨¨¨km .
This shows that the map σ |Σk is topologically transitive.
Problem 5.10 For a topological Markov chain σ |ΣA` or σ |ΣA with transition matrix
A “ pai j q, show that:
1. the number of blocks pi1 ¨ ¨ ¨ in q of length n with ai p i p`1 “ 1 for p “ 1, . . . , n ´ 1
such that i1 “ i and in “ j is equal to the pi, jq entry of the matrix An ;
2. the number of n-periodic points is trpAn q.
Solution 1. Note that ai p i p`1 “ 1 for p “ 1, . . . , n´1 if and only if ai1 i2 ¨ ¨ ¨ ain´1 in “ 1.
Hence, the number of blocks of length n with i1 “ 1 and in “ j is equal to
ÿ
aii2 ¨ ¨ ¨ ain´1 j “ pAn qi j ,
i2 ¨¨¨in´1
ÿ
k
pAn qii “ trpAn q.
i“1
Problem 5.11 Given a k ˆ k matrix A with entries ai j in t0, 1u, show that the fol-
lowing properties are equivalent:
1. there exists n P N such that ai1 i2 ¨ ¨ ¨ ain´1 in “ 0 for all i1 , . . . , in P t1, . . . , ku;
2. An “ 0 for some n P N;
3. ΣA “ ∅.
Problem 5.12 For a topological Markov chain σ |ΣA` , compute the number of peri-
odic orbits with period p prime in terms of the transition matrix A.
176 II.5 Symbolic Dynamics
trpA p q ´ tr A
.
p
Problem 5.13 For a topological Markov chain σ |ΣA` , compute the number of peri-
odic orbits with period p “ p1 p2 , with p1 , p2 prime, in terms of the transition
matrix A.
which implies that the number of periodic orbits with period p is equal to
pnq pnq
Solution Let ai j be the entries of An . We show by induction on n that ak1 “ 0 for
all n ě 0 and 1 ď k ď 4. This is clear for n “ 1. Now assume that the property holds
for some n P N. Then
pn`1q
ÿ
4
p1q pnq
ak1 “ ak j a j1 “ 0 for 1 ď k ď 4.
j“1
Problem 5.16 Show that the topological Markov chain σ |ΣA` with transition matrix
¨ ˛
1010
˚1 1 1 0‹
A“˚ ‹
˝1 1 0 1‚
0111
Problem 5.17 Consider the topological Markov chains σ |ΣA` and σ |ΣB` with tran-
sition matrices, respectively,
ˆ ˙ ˆ ˙
11 10
A“ and B “ .
10 01
Problem 5.18 Find topological Markov chains σ |ΣA` and σ |ΣB` both with zero
topological entropy that are not topologically conjugate.
Solution Consider the topological Markov chains σ |ΣA` and σ |ΣB` with transition
matrices, respectively,
ˆ ˙ ˆ ˙
10 10
A“ and B “ .
00 01
Then
hpσ |ΣA` q “ log ρ pAq “ log 1 “ 0
and
hpσ |ΣB` q “ log ρ pBq “ log 1 “ 0.
On the other hand, the topological Markov chains σ |ΣA` and σ |ΣB` are not topologi-
cally conjugate because ΣA` contains the single sequence p11 ¨ ¨ ¨ q, while ΣB` contains
the two sequences p11 ¨ ¨ ¨ q and p22 ¨ ¨ ¨ q.
Problem 5.19 Find topological Markov chains σ |ΣA` and σ |ΣB` with
Solution Consider the topological Markov chains σ |ΣA` and σ |ΣB` with transition
matrices, respectively,
¨ ˛ ¨ ˛
1000 1000
˚0 0 0 0‹ ˚0 1 0 0‹
A“˚ ‹ ˚
˝0 0 1 1‚ and B “ ˝0 0 1 1‚.
‹
0010 0010
Since ?
1` 5
ρ pAq “ ρ pBq “ ,
2
it follows from Proposition 5.4 that
?
1` 5
hpσ |ΣA` q “ hpσ |ΣB` q “ log .
2
On the other hand, the number of fixed points of σ |ΣA` is tr A “ 2, while the number
of fixed points of σ |ΣB` is tr B “ 3. Hence, the topological Markov chains are not
topologically conjugate.
II.5 Symbolic Dynamics 179
Problem 5.20 For the set Λ in Problem 4.9, show that the correspondence
č
ω “ p¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q ÞÑ Λω “ f ´n Vin
nPZ
Solution As in Problem 4.6, the intersection Λω is a nonempty closed set for each
ω P t1, 2uZ since it is the intersection of the decreasing sequence of nonempty closed
sets
čn
Rn pω q “ f ´m Vim . (II.5.7)
m“´n`1
On the other hand, the projections of Rn pω q on the horizontal and vertical axes are
bounded, respectively, by sup |g1 |n and pinf |h1 |q´n . Since
Finally, we have
č
Hpσ pω qq “ f ´n Vin`1 pω q
nPZ
č
“ f 1´n Vin pω q
nPZ
“ f pHpω qq.
is dense in Σ2 .
dpω , ω p q ă β ´p Ñ 0 when p Ñ 8.
Problem 5.23 Compute the zeta function of the shift map σ |Σk .
II.5 Symbolic Dynamics 181
Solution We have
(
an “ card ω P Σk : σ n pω q “ ω “ kn
and so
ÿ8 n n
k z
ζ pzq “ exp .
n“1
n
Since ˜ ¸1
ÿ8 n n 8
ÿ
k z k
“ kn zn´1 “ (II.5.8)
n“1
n n“1
1 ´ kz
Problem 5.24 Compute the zeta function of the topological Markov chain σ |ΣA`
with transition matrix ¨ ˛
1 01
A “ ˝1 1 1‚.
1 11
Problem 5.25 Compute the zeta function of the automorphism of the torus T2
induced by the matrix ˆ ˙
51
A“ .
92
and so
ÿ8
pτ n ` τ ´n ´ 2qzn
ζ pzq “ exp
n“1
n
˜ ¸
ÿ8 8 8 n
pτ zqn ÿ pτ ´1 zqn ÿ z
“ exp ` ´2 .
n“1
n n“1
n n“1
n
p1 ´ zq2 p1 ´ zq2
ζ pzq “ ´1
“
p1 ´ τ zqp1 ´ τ zq 1 ´ 7z ` z2
?
for |z| ă p7 ´ 3 5q{2.
Ť
Problem 5.26 Given a nonempty finite set F in ně2 t1, . . . , kun (that is, a set of
blocks with finite length), consider the set F1 Ĺ Σk of all sequences in Σk containing
no blocks in F. Determine two sets F1 and F2 with k “ 2 such that F11 “ F21 ‰ ∅
with F1 ‰ F2 .
II.5 Symbolic Dynamics 183
Solution Take
F1 “ tp22qu and F2 “ tp221q, p222qu.
We show that F11
“ F21 .
A sequence ω P F11 does not contain the block p22q and so
it also cannot contain the blocks p221q or p222q. Hence, ω P F21 . On the other hand,
a sequence ω P F21 does not contain the block p22q, since otherwise it would also
contain p221q or p222q, which is impossible. Hence, F11 “ F21 .
Solution Take ω P F11 X F21 . Since ω P F11 , it contains no block in F1 and since
ω P F21 , it contains no block in F2 . Therefore, ω contains no block in F1 Y F2 , that
is, ω P pF1 Y F2 q1 .
Now take ω P pF1 Y F2 q1 . Then ω contains no block in F1 Y F2 . This implies that
it contains no block in Fi for i “ 1, 2 and so ω P F11 X F21 .
Ť
Problem 5.28 Given a nonempty finite set of blocks F in ně2 t1, . . . , kun , show
that F1 Ĺ Σk is compact and σ -invariant.
Solution The invariance is clear because the image and preimage of any sequence
under the shift map σ : Σk Ñ Σk contains exactly the same blocks as the original
sequence.
To show that F1 is compact, it suffices to show that it is closed. Take ω P Σk zF1
containing a block in F of length p and without loss of generality centered at the
origin. Let ` ˘ (
U “ Bdβ ω , β ´p “ ω 1 P Σk : dβ pω , ω 1 q ă β ´p .
Then dβ pω , ω 1 q ă β ´p if and only if npω , ω 1 q ą p (again npω , ω 1 q is the smallest
positive integer n such that in pω q ‰ in pω 1 q or i´n pω q ‰ i´n pω 1 q). In particular, no
ω 1 P U is in F1 and so U is an open neighborhood of ω that does not intersect F1 .
This shows that F1 is closed.
Ť
Problem 5.29 Given a nonempty finite set of blocks F in ně2 t1, . . . , kun , show
that if F contains only blocks of length 2, then F1 “ ΣA for some k ˆ k matrix A with
entries in t0, 1u.
Problem 5.30 Show that for k “ 2 and F “ tp22qu, the matrix A in Problem 5.29 is
ˆ ˙
11
A“ .
10
Problem 5.31 Compute the number of n-periodic points of the restriction σ |F1 of
σ : Σ2 Ñ Σ2 to F1 for the set F “ tp22qu.
184 II.5 Symbolic Dynamics
Solution By Problem 5.30, the number of n-periodic points of σ |F1 is equal to the
number of n-periodic points of the topological Markov chain σ |ΣA with transition
matrix A as in Problem 5.30. On the other hand, by Problem 5.10, the latter is equal
to trpAn q.
Note that the eigenvalues of A are
? ?
1` 5 1´ 5
λ“ and μ “ .
2 2
Considering the eigenvectors, respectively, pλ , 1q and pμ , 1q we obtain
ˆ ˙ˆ ˙ ˆ ˙
λ μ λ 0 1 1 ´μ
A“ ?
1 1 0 μ 5 ´1 λ
and so ˆ ˙ˆ n ˙ ˆ ˙
λ μ λ 0 1 1 ´μ
A “
n ? .
1 1 0 μn 5 ´1 λ
Therefore,
trpAn q “ λ n ` μ n “ λ n ` p´1qn λ ´n .
(see Figure I.5.3 for an example). Let i0 ¨ ¨ ¨ in´1 be the base-2 representation of j
and let p1 , . . . , p2n “ p0 be the preimages of the unique fixed point p of f under f n
(see Figure I.5.4). Show that if f : S1 Ñ S1 is an expanding map with deg f “ 2 (see
Problem 1.27), then the intervals
satisfy:
1. |Ii0 ¨¨¨in´1 | ď λ ´n ;
2. Ii0 ...in´1 in Ĺ Ii0 ...in´1 ;
3. f m pIi0 ¨¨¨in q “ Iim ¨¨¨in for each m ď n.
degp f n q “ pdeg f qn “ 2n ,
that is, all points have exactly 2n preimages under f n . We denote these preimages
by p j for j “ 1, . . . , 2n , numbering them in anticlockwise direction on S1 so that
p2n “ p0 “ p. These points determine 2n intervals
and so ż p j`1
1“ p f n q1 pxq dx ě λ n |Ii0 ¨¨¨in´1 |.
pj
f n`1 pq j q “ f p f n pq j qq “ f pqq “ p
and
f n`1 pp j q “ f p f n pp j qq “ f ppq “ p,
it follows that q j and p j are endpoints of some intervals I j0 ¨¨¨ jn . Since q j P rp j , p j`1 s,
we conclude that
For j “ i0 ¨ ¨ ¨ in we have
j ” i1 ¨ ¨ ¨ in mod 2n´1
and so f pIi0 ¨¨¨in q “ Ii1 ¨¨¨in . The last property follows now by induction.
Solution By Problem 5.32, the endpoints of each interval Ii0 ¨¨¨in´1 belong to f ´n p.
Moreover,
|Ii0 ¨¨¨in´1 | ď λ ´n Ñ 0 when n Ñ 8
and so the distance between two consecutive
Ť elements of f ´n p tends to zero when
´n
n Ñ 8. This readily implies that the set nPN f p is dense in S1 .
Problem 5.34 Let f : S1 Ñ S1 be an expanding map with deg f “ 2. Show that there
exists a continuous onto map H : Σ2` Ñ S1 such that H ˝ σ “ f ˝ H on Σ2` .
186 II.5 Symbolic Dynamics
using the same notation as in Problem 5.32. By item 2 in that problem the inter-
vals Ii1 ¨¨¨in form a decreasing sequence of closed sets and so their intersection is
nonempty. Moreover, since
|Ii1 ¨¨¨in | Ñ 0 when n Ñ 8,
Ş
the set nPN Ii1 ¨¨¨in contains exactly one point, which we continue to denote by Hpω q.
Now we show that H is continuous. Assume that dβ pω , ω 1 q “ β ´n . Writing
this means that n is the smallest positive integer such that jn ‰ jn1 . Finally, letting
jl “ jl1 “ il ` 1 for l ă n, we obtain
and so č
f pxq “ Ii2 ¨¨¨in . (II.5.10)
nPN
Therefore,
f pHpω qq “ f pxq “ Hpσ pω qq. (II.5.11)
satisfy:
1. |Ii0 ¨¨¨in´1 | ď λ ´n ;
2. Ii0 ...in´1 in Ĺ Ii0 ...in´1 ;
3. f m pIi0 ¨¨¨in q “ Iim ¨¨¨in for each m ď n.
As in Problem 5.34, let il “ jl ´ 1 for each l and define a map H : Σk` Ñ S1 by
č
Hp j1 j2 ¨ ¨ ¨ q “ Ii1 ¨¨¨in .
nPN
It follows from the former properties that the map H is well defined, continuous,
and onto. Finally, we show that H ˝ σ “ f ˝ H. Let x “ Hpω q. Then x P Ii1 ¨¨¨in and
for all n P N. This implies that properties (II.5.10) and (II.5.11) hold.
Problem 5.36 Show that the map H in Problem 5.35 is not a homeomorphism.
Solution Note that Σk` is the union of two disjoint nonempty closed sets, but that
the same does not happen with S1 .
Problem 5.37 Show that the periodic points of any expanding map f : S1 Ñ S1 are
dense in S1 .
Solution Let k “ deg f . By Problem 5.35, there exists a continuous onto map
H : Σk` Ñ S1 such that
H ˝ σ “ f ˝ H.
Hence, if σ n pω q “ ω , then
“ Ii0 ¨¨¨in´1 X S1 ‰ ∅.
Solution Given nonempty open sets U,V Ĺ S1 , take intervals Ii0 ¨¨¨in´1 and Ii1 ¨¨¨i1
0 n´1
as in (II.5.12). Note that for each m ą n we have
Problem 5.40 For the map f in Problem 5.35, show that hp f q ď log k.
Solution By Problem 5.35, there exists a continuous onto map H : Σk` Ñ S1 such
that H ˝ σ “ f ˝ H on Σk` . Hence, it follows from Problem 2.33 that
Solution 1. Since
2. if A, B P B and ` ˘
μ pAzBq Y pBzAq “ 0,
then μ pAq “ μ pBq.
Solution 1. Note that
μ pA Y Bq “ μ pAq ` μ pBzAq
and so
μ pA Y Bq ` μ pA X Bq “ μ pAq ` μ pBzAq ` μ pA X Bq
“ μ pAq ` μ pBq.
2. It follows from the hypothesis that
μ pAzBq “ μ pBzAq “ 0.
Hence,
μ pAq “ μ pAzBq ` μ pA X Bq “ μ pA X Bq
and
μ pBq “ μ pBzAq ` μ pA X Bq “ μ pA X Bq,
which shows that μ pAq “ μ pBq.
Problem 6.3 Show that the Borel σ -algebra B on R is generated by the set of all
intervals:
1. ra, bs with a ă b;
2. pa, bs with a ă b.
Solution 1. Since B is generated by the open intervals, it suffices to note that
8 ”
ď 1 1ı
pa, bq “ a` ,b´
n“1
n n
and
8 ´
č 1 1¯
ra, bs “ a´ ,b`
n“1
n n
for all a, b P R. Indeed, this implies that the σ -algebra generated by the closed in-
tervals coincides with B.
2. Similarly, it suffices to note that
8 ´
ď 1ı
pa, bq “ a, b ´
n“1
n
II.6 Ergodic Theory 191
and
8 ´
č 1¯
pa, bs “ a, b `
n“1
n
for all a, b P R.
Problem 6.4 Show that the Borel σ -algebra B on R is generated by the set of all
intervals:
1. pa, `8q with a P R;
2. p´8, bq with b P R;
3. ra, `8q with a P R;
4. p´8, bs with b P R.
and
8
ď
pa, `8q “ pa, a ` nq
n“1
for all a, b P R.
2. Similarly, we have
8 ´
č 1¯
pa, bq “ p´8, bqz ´8, a `
m“1
m
and
8
ď
p´8, bq “ p´n, bq
n“1
for all a, b P R.
3. In this case, note that
ď8 ” ¯
1
pa, bq “ a ` , `8 zrb, `8q
n“1
n
and
8 ´
1 ¯
8 č
ď
ra, `8q “ a´ ,n
n“1 m“1
m
for all a, b P R.
192 II.6 Ergodic Theory
4. Finally, we have
8 ´
ď 1ı
pa, bq “ ´8, b ´ zp´8, as
n“1
n
and
8 ´
1¯
8 č
ď
p´8, bs “ ´n, b `
n“1 m“1
m
for all a, b P R.
Problem 6.5 Show that the Borel σ -algebra on R coincides with the σ -algebra gen-
erated by the open sets in R.
Problem 6.6 Given a set C generating the Borel σ -algebra B on R, show that a
function ϕ : X Ñ R is A-measurable if and only if ϕ ´1 B P A for every B P C.
Hence, it follows from Problems 6.4 and 6.6 that the function ϕ 2 is A-measurable.
Solution By Problems 6.4 and 6.6, it suffices to show that the set
(
Ba “ x P R : ϕ pxq ą a
belongs to the Borel σ -algebra B for each a P R. Let b “ inf Ba . Then either b P Ba
and so Ba “ rb, `8q (see Figure II.6.1 for an example) or b R Ba and so Ba “
pb, `8q. In both cases one has Ba P B.
b
Fig. II.6.1 Preimage of the interval pa, `8q under a discontinuous increasing function.
Solution We have
" *
(
x P X : sup ϕn pxq ą a “ x P X : ϕn pxq ą a for some n P N
nPN
ď
“ tx P X : ϕn pxq ą au P A
nPN
since each function ϕn is A-measurable. Hence, it follows from Problems 6.4 and 6.6
that the function supnPN ϕn is A-measurable.
To show that the infimum is A-measurable it suffices to note that
inf ϕn “ ´ supp´ϕn q.
nPN nPN
Problem 6.11 For a measure space pX, A, μ q with μ pXq ă `8, let ϕn : X Ñ R, for
n P N, and ϕ : X Ñ R be A-measurable functions such that
for all x P X. Show that given ε ą 0, there exists a set B P A with μ pBq ă ε such that
ϕn Ñ ϕ when n Ñ 8
uniformly on XzB.
Note that for each q the sequence Bn pqq decreases with n and that
8
č
Bn pqq “ ∅.
n“1
Now let
II.6 Ergodic Theory 195
Given ε ą 0 and q P N, take nq P N such that μ pBnq pqqq ă ε {2q and let
8
ď
B“ Bnq pqq.
q“1
ÿ8 8
` ˘ ÿ ε
μ pBq ď μ Bnq pqq ă q
“ ε.
q“1 q“1
2
1
|ϕn pxq ´ ϕ pxq| ă .
q
Solution Let (
Bt “ x P X : |ϕ pxq| ě t .
Then ż ż
t c μ pBt q “ t c χBt d μ ď |ϕ |c d μ ,
X X
which yields the desired inequality.
196 II.6 Ergodic Theory
Problem 6.13 Show that the measure μ on S1 obtained from the Lebesgue mea-
sure λ on r0, 1s is invariant under the expanding map Em : S1 Ñ S1 .
Solution We identify S1 with r0, 1s. For each set B Ď r0, 1s in B, let
!” x ` i ı )
Bi “ :xPB
m
for i “ 1, . . . , m. Then
ď
m
Em´1 B “ Bi .
i“1
ÿ
m
μ pEm´1 Bq “ λ pBi q
i“1
ÿm
λ pBq
“
i“1
m
“ λ pBq “ μ pBq
ř Given finitely many or countably many intervals pai , bi q Ĺ r0, 1s, for
Problem 6.14
i P I, with iPI pbi ´ ai q “ 1, consider a map f : r0, 1s Ñ r0, 1s such that
x ´ ai
f pxq “
bi ´ ai
for each x P pai , bi q and i P I (see Figure I.6.1 for an example). Show that f preserves
the Lebesgue measure λ on r0, 1s.
Since the intervals p0, aq generate the Borel σ -algebra on r0, 1s, it follows that the
measure λ is f -invariant.
II.6 Ergodic Theory 197
Since the intervals ra, bs generate the Borel σ -algebra on r0, 1s (see Problem 6.3), it
follows that the measure μ is f -invariant.
(see Figure I.6.3). Show that f preserves the measure μ on r0, 1s defined by
ż
dx
μ pAq “ .
A `x
1
Solution First note that it suffices to consider the intervals A “ r0, aq for a P p0, 1q
since these generate the Borel σ -algebra on r0, 1s. We have
8 ”
ď 1 1¯
f ´1 r0, aq “ ,
n“1
n`a n
` ˘ ÿ 8 ´” 1 1 ¯¯
μ f ´1 r0, aq “ μ ,
n“1
n`a n
ÿ8 ż 1{n
dx
“
n“1 1{pn`aq
1 `x
ˆ ´ ˙
1¯ ´ 1 ¯
ÿ8
“ log 1 ` ´ log 1 `
n“1
n n`a
ÿ8 ˆ ˙
n`1 n
“ log ´ log
n“1
n`1`a n`a
ża
1 dx
“ ´ log “ “ μ pr0, aqq
1`a 0 1 `x
Problem 6.17 Show that the map g : r0, 1s Ñ r0, 1s given by gpxq “ 4xp1 ´ xq pre-
serves the measure μ on r0, 1s defined by
ż
dx
μ pAq “ a .
A xp1 ´ xq
for a ą 0. Since ˆ ˙
1 1 2
xp1 ´ xq “ ´ x ´ ,
4 2
we obtain ˆ ˙˙1
ˆ
1 x ´ 1{2
parcsinp2x ´ 1qq “ arcsin
1{2
1
“a (II.6.1)
1{4 ´ px ´ 1{2q2
1
“a .
xp1 ´ xq
Hence, żb
dx
μ pra, bsq “ a
a xp1 ´ xq
“ arcsinp2b ´ 1q ´ arcsinp2a ´ 1q.
On the other hand, since
II.6 Ergodic Theory 199
”1 1? 1 1? ı ”1 1? 1 1? ı
g´1 ra, bs “ ´ 1 ´ a, ´ 1´b Y ` 1 ´ b, ` 1´a
2 2 2 2 2 2 2 2
(and the intervals on the right-hand side are disjoint), we obtain
ˆ” ı˙
` ´1 ˘ 1 1? 1 1?
μ g ra, bs “ μ ´ 1 ´ a, ´ 1´b
2 2 2 2
ˆ” ı˙
1 1? 1 1?
`μ ` 1 ´ b, ` 1´a
2 2 2 2
? ?
“ arcsinp´ 1 ´ bq ´ arcsinp´ 1 ´ aq
? ?
` arcsinp 1 ´ aq ´ arcsinp 1 ´ bq
? ?
“ 2 arcsinp 1 ´ aq ´ 2 arcsinp 1 ´ bq.
Since the intervals ra, bs generate the Borel σ -algebra on r0, 1s (see Problem 6.3), it
follows that the measure μ is g-invariant.
px, yq ÞÑ px ` α , x ` yq
preserves the measure m on T2 obtained from the Lebesgue measure on r0, 1s2 .
Solution Let
Problem 6.19 Let f be the Gauss map in Problem 6.16. Show that if
1
x “ ra1 , a2 , . . .s “
a1 ` a2 `¨¨¨
1
3. for m P N we have
1 pm
ra1 , a2 , . . . , am s :“ “ .
a1 ` 1
1
qm
a2 `
¨¨¨` a1m
pm qm´1 ´ pm´1 qm “ pam pm´1 ` pm´2 qqm´1 ´ pm´1 pam qm´1 ` qm´2 q
“ ´ppm´1 qm´2 ´ pm´2 qm´1 q
“ ´p´1qm´2 “ p´1qm´1 ,
3. Again the identity is immediate for m “ 1. Now assume that it holds with m
replaced by m ´ 1 and note that
„ j
1
ra1 , . . . , am s “ a1 , . . . , am´1 ` .
am
ÿ
k ÿ
k
pi “ 1 and pi pi j “ p j
i“1 i“1
Solution 1. We have
ÿ
k ÿ
k
μ pCi1 ¨¨¨in j q “ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in pin j
j“1 j“1
2. Note that
ď
k
σ ´1Ci1 ¨¨¨in “ C ji1 ¨¨¨in
j“1
` ˘ ÿk
μ σ ´1Ci1 ¨¨¨in “ μ pC ji1 ¨¨¨in q
j“1
ÿ
k
“ p j p ji1 ¨ ¨ ¨ pin´1 in
j“1
Since the cylinders generate the σ -algebra on Σk` , this implies that the measure μ is
σ -invariant.
3. We have
ÿ
k ÿ
k
μ pΣk` q “ μ pCi q “ pi “ 1.
i“1 i“1
Problem 6.21 For the measure μ in Problem 6.20, consider the k ˆ k matrix A with
entries #
1 if pi j ą 0,
ai j “
0 if pi j “ 0.
(by the definition of A). Since the cylinders generate the topology on ΣA` , we con-
clude that pi1 i2 ¨ ¨ ¨ q P supp μ and so ΣA` Ď supp μ .
On the other hand, given pi1 i2 ¨ ¨ ¨ q P Σk` zΣA` , there exists n P N with ain in`1 “ 0
and thus,
pin in`1 “ 0 and μ pCi1 ¨¨¨in`1 q “ 0.
Since pi1 i2 ¨ ¨ ¨ q P Ci1 ¨¨¨in`1 , we conclude that pi1 i2 ¨ ¨ ¨ q R supp μ and so supp μ Ď ΣA` .
f ´m A X f ´n A “ ∅
but this is impossible because the measure μ is finite. This contradiction yields the
desired property.
ϕ ´1 α “ f ´1 pϕ ´1 α q “ pϕ ˝ f q´1 α .
or, equivalently,
ϕ pxq “ α if and only if ϕ p f pxqq “ α .
Therefore, ϕ ´1 α is f -invariant for every α P R if and only if
ϕ p f pxqq “ ϕ pxq
Moreover, let μ be the measure on S1 obtained from the Lebesgue measure on r0, 1s.
Show that:
1. if ϕ pxq “ e2π imx with m P Z, then
ż
1ÿ
n
lim ϕ pRα pxqq “
k
ϕ dμ for all x P S1 ;
nÑ8 n S1
k“1
Solution 1. Let am “ e2π imα . For the function ϕ pxq “ e2π imx we have
For m ‰ 0 we have
ÿ
n ÿ
n
1 ´ anm
ϕ pRkα pxqq “ ϕ pxq akm “ ϕ pxqam
k“1 k“1
1 ´ am
and so
1ÿ
n
lim ϕ pRkα pxqq “ 0. (II.6.2)
nÑ8 n
k“1
1ÿ 1ÿ
n n
lim ϕ pRkα pxqq “ lim 1 “ 1. (II.6.3)
nÑ8 n nÑ8 n
k“1 k“1
Finally, using the formula eiy “ cos y ` i sin y and integrating we obtain
ż #
0 if m ‰ 0,
ϕ dμ “
S 1 1 if m “ 0.
II.6 Ergodic Theory 205
Comparing with (II.6.2) and (II.6.3), it follows that the desired property holds for
the function ϕ pxq “ e2π imx .
2. First recall that the set of all linear combinations of the functions e2π imx is dense
in the set of all continuous functions ϕ : S1 Ñ S1 equipped with the distance
(
ϕ ´ ψ “ sup dpϕ pxq, ψ pxqq : x P S1 .
and so ż ż
ϕ dμ ´ ε ď ψ1 d μ
S1 S1
1ÿ
n
“ lim ψ1 pRkα pxqq (II.6.4)
nÑ8 n
k“1
1ÿ
n
ď lim inf ϕ pRkα pxqq
nÑ8 n k“1
(it follows readily from item 1 that the property also holds for linear combinations
of the functions e2π imx ). One can show in a similar manner that
ż
1ÿ
n
lim sup ϕ pRα pxqq ď
k
ϕ dμ ` ε. (II.6.5)
nÑ8 n k“1 S1
Since ε is arbitrary, it follows from (II.6.4) and (II.6.5) that the property in item 1
holds for all continuous functions ϕ : S1 Ñ R.
3. First observe that
ÿn
Tn pxq “ χU pRkα pxqq.
k“1
ψ1 ď χU ď ψ2
and
206 II.6 Ergodic Theory
ż
pψ2 ´ ψ1 q d μ ă ε .
S1
Proceeding as before, we find that the property in item 1 also holds with ϕ replaced
by χU . Hence,
ż
Tn pxq
lim “ χU d μ “ b ´ a for all x P S1 .
nÑ8 n S1
1 ÿ ` ´k ˘
n´1
lim μ f A X B “ μ pAqμ pBq
nÑ8 n
k“0
1 ÿ ` ´k ˘
n´1
μ pAqμ pXzAq “ lim μ f A X pXzAq
nÑ8 n
k“0
1ÿ ` ˘
n´1
“ lim μ A X pXzAq “ 0
nÑ8 n
k“0
Clearly, f ´1C “ C and since μ is ergodic, either μ pCq “ 0 or μ pCq “ 1. Note that
8 č
ď 8
BzC “ Bz f ´k B
n“1 k“n
μ p f ´n Aq “ μ pAq ą 0
μ pCq “ μ pBq ą 0,
μ pBq “ μ pCq “ 1.
Now assume that the set B in (II.6.6) has full measure for any set A P A with
μ pAq ą 0. Given an f -invariant set A P A with μ pAq ą 0, we have
f ´n A “ A for all n P N.
μ pAq “ μ pBq “ 1.
Solution By Problem 6.23 the function τA is well defined for almost every x P A
and so the integral is also well defined.
Consider the sets (
An “ x P A : τA pxq “ n
and (
Bn “ x P X : f n pxq P A, f k pxq R A for k “ 1, . . . , n ´ 1
for each n P N. Clearly, An Ď Bn . Moreover,
An X Am “ ∅ and Bn X Bm “ ∅
˜ ¸
ď8
μ Bn “ 1.
n“1
and so ˜ ¸
8
ď 8
ÿ
1“μ Bn “ μ pBn q
n“1 n“1
8 ÿ
ÿ 8 ÿ
ÿ k
“ μ pAk q “ μ pAk q
n“1 kěn k“1 n“1
ÿ8 ż
“ k μ pAk q “ τA d μ .
k“1 A
1 ÿ ` ´i ˘
n´1
lim μ f A X B ´ μ pAqμ pBq “ 0
nÑ8 n
i“0
1 ÿ ` ´i ˘ 2p n ´ p
n´1
μ f A X B ´ μ pAqμ pBq ď ` ε
n i“0 n n
1 ÿ ` ´i ˘
n´1
lim sup μ f A X B ´ μ pAqμ pBq ď ε
nÑ8 n i“0
1 ÿ
n´1
lim μ pAq ´ μ pAq2 “ 0
nÑ8 n
i“0
and so
μ pAq “ μ pAq2 .
Hence, either μ pAq “ 0 or μ pAq “ 1, which shows that the measure μ is ergodic.
Solution Let U,V Ĺ X be nonempty open sets intersecting the support supp μ . Then
μ pUq, μ pV q ą 0 and since μ is mixing, we have
` ˘
lim μ f ´nU XV “ μ pUqμ pV q ą 0.
nÑ8
Hence,
μ p f ´nU XV q ą 0
for all sufficiently large n, which implies that
f ´nU XV ‰ ∅
Note that ÿ
Hμ pξ q “ ´ ψ pμ pCqq.
CPξ
Since
ψ 2 pxq “ 1{x ą 0 for x ‰ 0,
the function ψ is strictly convex and so
ˆÿ
p ˙ ÿ
p
ψ ai xi ď ai ψ pxi q (II.6.8)
i“1 i“1
řp
for any numbers x1 , . . . , x p , a1 , . . . , a p P r0, 1s with i“1 ai “ 1. Moreover, (II.6.8) is
an equality if and only if all numbers xi corresponding to a nonzero coefficient ai
are equal. Therefore,
ÿ
Hμ pξ q “ ´ ψ pμ pCqq
CPξ
1 ÿ
“ ´ card ξ ψ pμ pCqq
card ξ
CPξ
ˆÿ ˙
μ pCq
ď ´ card ξ ψ
card ξ
CPξ
ˆ ˙
1
“ ´ card ξ ψ
card ξ
1
“ ´ log “ log card ξ .
card ξ
Hμ pξ q ď Hμ pη q.
and so ˜ ¸
ÿ
ψ pμ pCqq “ ψ μ pDq
DĎC
ÿ ÿ
“ μ pDq log μ pDq
DĎC DĎC
ÿ
ě μ pDq log μ pDq
DĎC
ÿ
“ ψ pμ pDqq.
DĎC
Problem 6.34 Let μ be a probability measure on X. Show that if ξ and η are parti-
tions of X and (
ξ _η “ C XD : C P ξ,D P η ,
then
Hμ pξ _ η q ď Hμ pξ q ` Hμ pη q.
Solution We have
ÿ
Hμ pξ _ η q “ ´ μ pC X Dq log μ pC X Dq
CPξ ,DPη
ÿ „ j
μ pC X Dq
“´ μ pC X Dq log ` log μ pCq
μ pCq
CPξ ,DPη
ÿÿ ˆ ˙
μ pC X Dq
“´ μ pCqψ
DPη CPξ
μ pCq
ÿ
´ μ pC X Dq log μ pCq
CPξ ,DPη
again with ψ as in (II.6.7). Since the function ψ is convex (see Problem 6.32), we
obtain
212 II.6 Ergodic Theory
˜ ¸
ÿ ÿ μ pC X Dq ÿ
Hμ pξ _ η q ď ´ ψ μ pCq ´ μ pCq log μ pCq
DPη
μ pCq
CPξ CPξ
ÿ
“´ ψ pμ pDqq ` Hμ pξ q
DPη
“ Hμ pη q ` Hμ pξ q.
f ´1 ξ “ t f ´1C : C P ξ u,
then
Hμ p f ´1 ξ q “ Hμ pξ q.
Hμ pξn`m q “ Hμ pξn _ f ´n ξm q
ď Hμ pξn q ` Hμ p f ´n ξm q
“ Hμ pξn q ` Hμ pξm q.
This shows that the sequence cn “ Hμ pξn q is subadditive (see (II.2.7)). Therefore,
by (II.2.8), the limit
1 1
lim Hμ pξn q “ inf Hμ pξn q “ hμ p f , ξ q
nÑ8 n nPN n
exists.
II.6 Ergodic Theory 213
then
Hμ pξ |ζ q ď Hμ pξ |η q.
Solution We have
ÿ μ pC X Dq
Hμ pξ |ζ q “ ´ μ pC X Dq log
μ pDq
CPξ ,DPζ
ÿ μ pC X Dq μ pC X Dq
“´ μ pD X Eq log
μ pDq μ pDq
CPξ ,DPζ ,EPη
ÿ ÿ μ pD X Eq ˆ μ pC X Dq ˙
“´ μ pEq ψ .
μ pEq μ pDq
CPξ ,EPη DPζ
and so
ÿ ˆ ˙
μ pC X Eq
Hμ pξ |ζ q ď ´ μ pEqψ
μ pEq
CPξ ,EPη
ÿ μ pC X Eq
“´ μ pC X Eq log
μ pEq
CPξ ,EPη
“ Hμ pξ |η q.
Hμ pξ _ ζ |η q “ Hμ pξ |ζ _ η q ` Hμ pζ |η q.
214 II.6 Ergodic Theory
Solution We have
ÿ μ pC X D X Eq
Hμ pξ _ ζ |η q “ ´ μ pC X D X Eq log
μ pDq
CPξ ,DPζ ,EPη
ÿ μ pC X D X Eq
“´ μ pC X D X Eq log
μ pD X Eq
CPξ ,DPζ ,EPη
ÿ μ pD X Eq
´ μ pC X D X Eq log .
μ pDq
CPξ ,DPζ ,EPη
Therefore,
ÿ μ pD X Eq
Hμ pξ _ ζ |η q “ Hμ pξ |ζ _ η q ´ μ pD X Eq log
μ pDq
DPζ ,EPη
“ Hμ pξ |ζ _ η q ` Hμ pζ |η q.
ÿ ÿ
n´1
´ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in log pi j i j`1
i1 ¨¨¨in j“1
ÿ
k ÿ
k ÿ
k
“´ pi log pi ´ pn ´ 1q pi pi j log pi j .
i“1 i“1 j“1
pnq
Now consider the partitions ξ pnq “ ξn . We have ξm “ ξn`m´1 and so
1
hpσ , ξ pnq q “ inf Hμ pξn`m´1 q
mPNm
˜ ¸
1 ÿ ÿ
k k ÿ k
“ inf ´ pi log pi ` pn ` m ´ 2q pi pi j log pi j
mPN m
i“1 i“1 j“1
ÿ
k ÿ
k
“´ pi pi j log pi j .
i“1 j“1
II.6 Ergodic Theory 215
Since the partitions ξ pnq satisfy the hypotheses of Definition 6.13, we obtain
ÿ
k ÿ
k
hμ pσ q “ sup hpσ , ξ pnq q “ ´ pi pi j log pi j . (II.6.10)
nPN i“1 j“1
Problem 6.40 For the measure μ in Problem 6.20 with pi j “ p j for all i, j “ 1, . . . , k,
compute hμ pσ q.
ÿ
k ÿ
k
hμ pσ q “ ´ pi pi j log pi j
i“1 j“1
ÿ
k ÿ
k
“´ pi p j log p j
i“1 j“1
ÿ
k
“´ p j log p j .
j“1
References
20. C. Chicone, Ordinary Differential Equations with Applications. Texts in Applied Mathemat-
ics, vol. 34 (Springer, NY, 1999)
21. S.-N. Chow, J. Hale, Methods of Bifurcation Theory. Grundlehren der Mathematischen Wis-
senschaften, vol. 251 (Springer, NY, 1982)
22. E. Coddington, N. Levinson, Theory of Ordinary Differential Equations (McGraw-Hill, NY,
1955)
23. Y. Coudène, Ergodic Theory and Dynamical Systems. Universitext (Springer, London, 2016)
24. W. de Melo, S. van Strien, One-Dimensional Dynamics. Ergebnisse der Mathematik und ihrer
Grenzgebiete, vol. 25 (Springer, Berlin, 1993)
25. J. Guckenheimer, P. Holmes, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of
Vector Fields. Applied Mathematical Sciences, vol. 42 (Springer, NY, 1983)
26. J. Hale, L. Magalhães, W. Oliva, Dynamics in Infinite Dimensions. Applied Mathematical
Sciences, vol. 47 (Springer, NY, 2002)
27. J. Hale, S. Verduyn Lunel, Introduction to Functional-Differential Equations. Applied Math-
ematical Sciences, vol. 99 (Springer, NY, 1993)
28. M. Irwin, Smooth Dynamical Systems. Pure and Applied Mathematics, vol. 94 (Academic
Press, NY, 1980)
29. A. Katok, B. Hasselblatt, Introduction to the Modern Theory of Dynamical Systems. Encyclo-
pedia of Mathematics and its Applications, vol. 54 (Cambridge University Press, Cambridge,
1995)
30. A. Katok, J.-M. Strelcyn, with the collaboration of F. Ledrappier, F. Przytycki, Invariant Man-
ifolds, Entropy and Billiards: Smooth Maps with Singularities. Lecture Notes in Mathematics,
vol. 1222 (Springer, Berlin, 1986)
31. S. Katok, Fuchsian Groups. Chicago Lectures in Mathematics (University of Chicago Press,
IL, 1992)
32. G. Keller, Equilibrium States in Ergodic Theory. London Mathematical Society Student Texts,
vol. 42 (Cambridge University Press, Cambridge, 1998)
33. B. Kitchens, Symbolic Dynamics: One-Sided, Two-Sided and Countable State Markov Shifts.
Universitext (Springer, Berlin, 1998)
34. D. Lind, B. Marcus, An Introduction to Symbolic Dynamics and Coding (Cambridge Univer-
sity Press, Cambridge, 1995)
35. R. Mañé, Ergodic Theory and Differentiable Dynamics. Ergebnisse der Mathematik und ihrer
Grenzgebiete, vol. 8 (Springer, Berlin, 1987)
36. C. McMullen, Complex Dynamics and Renormalization. Annals of Mathematics Studies, vol.
135 (Princeton University Press, NJ, 1994)
37. J. Milnor, Dynamics in One Complex Variable. Annals of Mathematics Studies, vol. 160
(Princeton University Press, NJ, 2006)
38. S. Morosawa, Y. Nishimura, M. Taniguchi, T. Ueda, Holomorphic Dynamics. Cambridge
Studies in Advanced Mathematics, vol. 66 (Cambridge University Press, Cambridge, 2000)
39. J. Moser, Stable and Random Motions in Dynamical Systems. Annals of Mathematics Studies,
vol. 77 (Princeton University Press, NJ, 1973)
40. Z. Nitecki, Differentiable Dynamics: An Introduction to the Orbit Structure of Diffeomor-
phisms (MIT Press, MA, 1971)
41. J. Palis, W. de Melo, Geometric Theory of Dynamical Systems: An Introduction (Springer,
NY, 1982)
42. J. Palis, F. Takens, Hyperbolicity and Sensitive Chaotic Dynamics at Homoclinic Bifurca-
tions: Fractal Dimensions and Infinitely Many Attractors. Cambridge Studies in Advanced
Mathematics, vol. 35 (Cambridge University Press, Cambridge, 1993)
43. W. Parry, M. Pollicott, Zeta Functions and the Periodic Orbit Structure of Hyperbolic Dynam-
ics. Astérisque, vol. 187–188 (Soc. Math. France, Montrouge, 1990)
44. Ya. Pesin, Dimension Theory in Dynamical Systems: Contemporary Views and Applications.
Chicago Lectures in Mathematics (Chicago University Press, IL, 1997)
45. Ya. Pesin, Lectures on Partial Hyperbolicity and Stable Ergodicity. Zürich Lectures in
Advanced Mathematics (European Mathematical Society, Zürich, 2004)
References 219
46. K. Petersen, Ergodic Theory. Cambridge Studies in Advanced Mathematics, vol. 2 (Cam-
bridge University Press, Cambridge, 1983)
47. M. Pollicott, Lectures on Ergodic Theory and Pesin Theory on Compact Manifolds. Lon-
don Mathematical Society Lecture Note Series, vol. 180 (Cambridge University Press, Cam-
bridge, 1993)
48. M. Pollicott, M. Yuri, Dynamical Systems and Ergodic Theory. London Mathematical Society
Student Texts, vol. 40 (Cambridge University Press, Cambridge, 1998)
49. F. Przytycki, M. Urbanski, Conformal Fractals: Ergodic Theory Methods. London Mathemat-
ical Society Lecture Note Series, vol. 371 (Cambridge University Press, Cambridge, 2010)
50. C. Robinson, Dynamical systems: Stability, Symbolic Dynamics, and Chaos (CRC Press, OH,
1994)
51. D. Ruelle, Thermodynamic Formalism. Encyclopedia of Mathematics and its Applications,
vol. 5 (Addison-Wesley, MA, 1978)
52. K. Schmidt, Dynamical Systems of Algebraic Origin. Progress in Mathematics, vol. 128
(Birkhäuser, Basel, 1995)
53. G. Sell, Y. You, Dynamics of Evolutionary Equations. Applied Mathematical Sciences, vol.
143 (Springer, NY, 2002)
54. M. Shub, Global Stability of Dynamical Systems (Springer, NY, 1986)
55. R. Temam, Infinite-Dimensional Dynamical Systems in Mechanics and Physics. Applied
Mathematical Sciences, vol. 68 (Springer, NY, 1997)
56. P. Walters, An Introduction to Ergodic Theory. Graduate Texts in Mathematics, vol. 79
(Springer, Berlin, 1982)
Index
Symbols C
rxs, 4, 23 Cpn, ε q, 18
Λ , 36 Cs pxq, 37
ΣA , 50 Cu pxq, 37
ΣA` , 50 Characteristic function, 58
Σk , 50 Circle, 4, 23
Σk` , 49 rotation, 4
α -limit set, 15, 16 Cones, 37
α pxq, 15, 16 Conjugacy, 17, 24
γ pxq, 4, 6 Critical point, 5
γ ` pxq, 4, 5
γ ´ pxq, 4, 5, 25 D
ω -limit set, 15, 16, 25 dn px, yq, 17
ω pxq, 15, 16 dpE, Fq, 36
π pxq, 23 Denjoy’s theorem, 24
Diffeomorphism, 35
ă, 24
hyperbolic fixed point, 38
ρ p f q, 24
hyperbolic set, 35–37
σ , 49, 50
Differential equation, 5, 25
σ -algebra, 57
Distance, 36, 49, 50
τ pxq, 6
Dynamical system, 5, 15
ξ pnq , 60 continuous time, 5, 15
ξn , 59 discrete time, 3, 4, 6, 15, 51
ζ pzq, 51 topological, 15
f ´1 A, 3 Dynamics
hyperbolic, 35, 135
A low-dimensional, 23, 115
Automorphism of the torus, 5 symbolic, 49, 169
Autonomous differential equation, 5, 25 topological, 15, 95
E
B E 0 pxq, 39
Backward invariant set, 3 E s pxq, 35, 39
Basic theory, 3, 71 E u pxq, 35, 39
Birkhoff’s ergodic theorem, 59 Em , 4
Borel σ -algebra, 57 Endomorphism of the torus, 5
Bounded variation, 24 Entropy
© Springer Nature Switzerland AG 2019 221
L. Barreira and C. Valls, Dynamical Systems by Example, Problem Books
in Mathematics, https://doi.org/10.1007/978-3-030-15915-3
222 Index
metric, 60 measure, 59
topological, 17, 18, 49–51 set, 3, 5
Equivalence relation Irreducible matrix, 50
on R, 4
on Rn , 4 L
Ergodic Lebesgue
measure, 66, 206 integral, 58
theory, 57, 189 measure, 58
Expanding map, 4, 55, 184 Lift, 23, 24
Limit set, 15, 16, 25
F Local topological conjugacy, 38
First return time, 6 Low-dimensional dynamics, 23, 115
Fixed point, 4
hyperbolic, 38 M
Flow, 5, 25 Mpn, ε q, 18
geodesic, 41 Möbius transformation, 40
hyperbolic set, 39 Manifold
topological, 15 stable, 39
Forward unstable, 39
invariant set, 3 Map
recurrent point, 16 expanding, 4, 55, 184
Fractional part, 24 measurable, 59
Function Poincaré, 6
characteristic, 58 shift, 49, 50
integrable, 58 topologically mixing, 17, 50
measurable, 58 topologically transitive, 17, 50
simple, 58 Markov chain, 50
zeta, 51 Matrix
irreducible, 50
G transition, 50
Geodesic flow, 41 transitive, 50
Grobman–Hartman theorem, 38 Measurable
function, 58
H map, 59
Hμ pξ q, 59 Measure, 57
hp f q, 17 ergodic, 66, 206
hμ p f q, 60 invariant, 59
hμ p f , ξ q, 60 Lebesgue, 58
Homeomorphism mixing, 66, 208
orientation-preserving, 23 space, 58
orientation-reversing, 23 weakly mixing, 66, 208
Horizontal strip, 36 Metric entropy, 60
Horseshoe, 36 Minimal set, 19, 101
Hyperbolic Mixing measure, 66, 208
dynamics, 35, 135
fixed point, 38 N
set, 35–37, 39 Npn, ε q, 17
NW p f q, 16
I Negative semiorbit, 4, 6
Integrable function, 58 Nonwandering point, 16
Integral, 58
Invariant O
backward, 3 Orbit, 4, 6
forward, 3 Orientation-preserving
Index 223