2-Luís Barreira, Claudia Valls - Dynamical Systems by Example-Springer (2019)

Problem Books in Mathematics
Luís Barreira
Claudia Valls
Dynamical
Systems by
Example
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
ISSN 0941-3502 ISSN 2197-8506 (electronic)

ISBN 978-3-030-15914-6 ISBN 978-3-030-15915-3 (eBook)
https://doi.org/10.1007/978-3-030-15915-3
Library of Congress Control Number: 2019934778
Mathematics Subject Classification (2010): 37-01, 37Dxx, 37Bxx, 37Exx, 37Axx
© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
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
solutions, on the basic theory, topological dynamics, low-dimensional dynamics,

hyperbolic dynamics, symbolic dynamics, and basic ergodic theory. The text is
complemented by 50 figures.
We also provide a list of important topics, certainly incomplete, that were left out
of the book for reasons of scope together with recommendations for further reading:
• Holomorphic dynamics (see [13, 18, 36, 37, 38, 49]);
• Bifurcation theory and normal forms (see [5, 10, 21, 25]);
• Hamiltonian dynamics (see [1, 6, 10, 29, 39]);
• Discrete groups of isometries (see [4, 12, 31]);
• Dimension theory and multifractal analysis (see [7, 44]);
• Thermodynamic formalism and its applications (see [15, 32, 51]);
• Hyperbolicity and homoclinic bifurcations (see [42]);
• Hyperbolic dynamics and zeta functions (see [43]);
• Partial hyperbolicity and stable ergodicity (see [45]);
• Nonuniform hyperbolicity and smooth ergodic theory (see [9, 14, 47]);
• Hyperbolic systems with singularities and billiards (see [19, 30]);
• Algebraic dynamics and ergodic theory (see [52]);
• Infinite-dimensional dynamics (see [26, 27, 53, 55]).
Lisbon, Portugal Luís Barreira

February 2019 Claudia Valls
Contents
Part I Theory and Problems

I.1 Basic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
I.2 Topological Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
I.3 Low-Dimensional Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
I.4 Hyperbolic Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
I.5 Symbolic Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
I.6 Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Part II Problems and Solutions

II.1 Basic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
II.2 Topological Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
II.3 Low-Dimensional Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
II.4 Hyperbolic Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
II.5 Symbolic Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
II.6 Ergodic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
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.
Notions and Results
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 ń “ 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 .
Definition 1.2 Given a map f : X Ñ X, a set A Ď X is said to be:

1. f -invariant if f ´1 A “ A;
2. forward f -invariant if f pAq Ď A;
3. backward f -invariant if f ´1 A Ď A.
Let N0 “ N Y t0u be the set of all nonnegative integers.
© Springer Nature Switzerland AG 2019 3

L. Barreira and C. Valls, Dynamical Systems by Example, Problem Books
in Mathematics, https://doi.org/10.1007/978-3-030-15915-3 1
4 I.1 Basic Theory
Definition 1.3 Given a map f : X Ñ X and a point x P X, the set

(
γ ` pxq “ f n pxq : n P N0
is called the positive semiorbit of x. Moreover, when f is invertible, the sets

(
γ ´ pxq “ f ń pxq : n P N0
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
is called a q-periodic point of f . Moreover, x P X is called a periodic point if it is a

q-periodic point for some q P N. The fixed points are the 1-periodic points, that is,
those points x P X such that f pxq “ x.
A periodic point is said to have period q if it is q-periodic, but is not p-periodic
for any p ă q.
We also consider various classes of dynamical systems with discrete time.
Definition 1.5 Let „ be the equivalence relation on R given by
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.
Moreover, given an integer m ą 1, the expanding map Em : S1 Ñ S1 is defined by
Em rxs “ rmxs.
Definition 1.6 Given n P N, let „ be the equivalence relation on Rn given by
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
Given an n ˆ n matrix A with entries in Z, the endomorphism of the torus induced

by A is the map TA : Tn Ñ Tn defined by
TA rxs “ rAxs for rxs P Tn .
When TA is invertible, it is also called the automorphism of the torus induced by A.

Finally, we consider corresponding notions for continuous time.
Definition 1.7 A family Φ “ pϕt qtě0 of maps ϕt : X Ñ X, for t ě 0, such that ϕ0 “ id
and
ϕt`s “ ϕt ˝ ϕs for t, s ě 0
is called a semiflow on X. Moreover, a family Φ “ pϕt qtPR of maps ϕt : X Ñ X, for
t P R, such that ϕ0 “ id and
ϕt`s “ ϕt ˝ ϕs for t, s P R
is called a flow on X. A family Φ of maps is called a dynamical system with contin-

uous time if it is a flow or a semiflow.
We denote by v1 “ dv{dt the derivative of a function v “ vptq. It turns out that
any differential equation v1 “ f pvq with global unique solutions determines a flow.
Proposition 1.8 Let f : Rn Ñ Rn be a continuous function such that, given v0 P Rn ,
the initial value problem #
v1 “ f pvq,
vp0q “ v0
has a unique solution t ÞÑ vpt, v0 q defined on the whole R. Then the family of maps
ϕt : Rn Ñ Rn , for t P R, defined by ϕt pv0 q “ vpt, v0 q is a flow on Rn .
A point v0 P Rn with f pv0 q “ 0 is called a critical point of the equation v1 “ f pvq.
Definition 1.9 Given a semiflow Φ on X, a set A Ď X is said to be Φ -invariant if
ϕt´1 A “ A for t ě 0.
Moreover, given a flow Φ on X, a set A Ď X is said to be Φ -invariant if
ϕt´1 A “ A for t P R.
Definition 1.10 Given a semiflow Φ on X and a point x P X, the set
γ ` pxq “ tϕt pxq : t ě 0u
is called the positive semiorbit of x. Moreover, for a flow Φ on X, the sets
γ ´ pxq “ tϕ´t pxq : t ě 0u

6 I.1 Basic Theory
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
Fig. I.1.1 The first return time τ pxq.

I.1 Basic Theory 7
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
f p1q “ 3, f p2q “ 5, f p3q “ 4, f p4q “ 2 and f p5q “ 1
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
Fig. I.1.2 The map f in Problem 1.6.
Problem 1.7 Let f : I Ñ I be a strictly increasing map on an interval I Ď R. Show

that any periodic point of f is a fixed point.
8 I.1 Basic Theory
Problem 1.8 Let f : I Ñ I be a strictly decreasing map on an interval I Ď R. Show

that any periodic point of f is either a fixed point or a periodic point with period 2.
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.
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.
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.
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).
Problem 1.13 Given positive integers n and k, show that if x is a periodic point of f n
with period k, then it is a periodic point of f with period kn{l for some factor l of n
with pk, lq “ 1.
Problem 1.14 Let f : I Ñ I be a map on a closed interval I Ď R. Show that if the
positive semiorbit γ ` pxq of a point x P I is dense in I, then the set of points in I with
a dense positive semiorbit is dense in I.
Problem 1.15 Let f : X Ñ X be a continuous one-to-one map on a compact set.
Show that if the set P of periodic points of f is dense in X, then f is a homeomor-
phism.
Problem 1.16 Let f : X Ñ X be a continuous map on a compact set. Show that if
all points of X are periodic points of f , then f is a homeomorphism.
the set P of periodic points of f is dense in X and there exists p P N such that all
periodic points have period at most p, then f is a homeomorphism.
Problem 1.18 Consider an interval I “ pa, bq Ĺ S1 with 0 ă a ă b ă 1. Show that
there exists n P N such that E2n pIq “ S1 (see Definition 1.5), identifying I with the
set trxs : x P Iu Ĺ S1 (see Figures I.1.3 and I.1.4).
Problem 1.19 Show that given x P S1 and δ ą 0, there exist y P px ´ δ , x ` δ q Ď S1

and n P N such that
n
E pxq ´ E n pyq ě 1 .
2 2
4
Ť8
Problem 1.20 Given x P S , show that the union n“1 E2ń x is dense in S1 .
1
Problem 1.21 Consider the map f : R Ñ R defined by f pzq “ z3 on the set
R “ tz P C : |z| “ 1u.
Show that the set of periodic points of f is dense in R.

I.1 Basic Theory 9
E2 x
1 2 1 x
Fig. I.1.3 The expanding map E2 .
E22 x
1 4 1 2 3 4 1 x
Fig. I.1.4 The map E22 “ E4 .

10 I.1 Basic Theory
Problem 1.22 Consider the map f : T2 Ñ T2 defined by
f px, yq “ pEm pxq, Em pyqq.
Show that the set of periodic points of f is dense in T2 .

Problem 1.23 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 set of periodic points
of TA is Qn {Zn .
Problem 1.24 Let TA : Tn Ñ Tn be an endomorphism of the torus. Show that
card TA´1 x “ |det A| for all x P Tn ,
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).
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
card f ´1 x “ q for all x P S1 .
(The number q coincides with the degree deg f of f .)

Problem 1.28 Given a map f : X Ñ X, show that the complement of a backward
f -invariant set is forward f -invariant.
Problem 1.29 Let f : X Ñ X be a continuous map on a compact metric space. More-
over, let A Ď X be a closed backward f -invariant set and define
8
č
Λ“ f ń A.
n“0
Show that:
1. if U is an open neighborhood of Λ , then f ń 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 . Show that for each set I Ď R` 0 the union

ď (
px, yq P R2 : y6 ` x4 “ a
aPI
is invariant under the flow determined by the differential equation.

#
x1 “ ax ´ xy,
y1 “ ý ` x2 ´ 2y2 ,
on R2 for some a ą 0. Show that the solutions starting in the sets

" *
( x2
S1 “ px, yq P R : x “ 0
2
and S2 “ px, yq P R2 : y “
1 ` 2a
remain in these sets for all time.

12 I.1 Basic Theory
Problem 1.33 Find the flow determined by the equation x2 ` x “ 0.

Problem 1.34 Find whether the equation x1 “ xpx2 ` 1q determines a flow.
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.
#
x1 “ y,
y1 “ ´ sin x.
Show that the set (

X “ px, yq P R2 : y “ 0, x P p0, π q
is a Poincaré section and determine the corresponding first return time and Poincaré
map.
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
hpϕτ px,tq pxqq “ ψt phpxqq for x,t ą 0.
Fig. I.1.6 Phase portrait of the equation x1 “ ´x.
Fig. I.1.7 Phase portrait of the equation x1 “ ´x2 .
Problem 1.38 Let f , g : R2 Ñ R be C1 functions with f ą 0 such that
f px ` k, y ` lq “ f px, yq and gpx ` k, y ` lq “ gpx, yq
for all x, y P R and k, l P Z. Then the differential equation

I.1 Basic Theory 13
x1 “ f px, yq, y1 “ gpx, yq
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
ϕt p0, zq “ pxptq, yptqq “ pxpt, zq, ypt, zqq
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
(see Figure I.1.8) is invertible.
1, h z
0, z ϕt 0, z
1 x
Fig. I.1.8 The map h in Problem 1.38.
Problem 1.39 Consider the differential equation px1 , y1 q “ pα , β q on T2 “ R2 {Z2

with α , β ‰ 0. Determine whether any orbit of the flow determined by this equation
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
14 I.1 Basic Theory
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 ;
2. has at least one point in each of the sets

( (
Bϕ “ v P Aϕ : pL2 ϕ qpvq ď 0 and Cϕ “ v P Aϕ : pL2 ϕ qpvq ě 0 ;
3. intersects Aϕ transversally at each point in the sets

( (
v P Aϕ : pL2 ϕ qpvq ă 0 and v P Aϕ : pL2 ϕ qpvq ą 0 .
Chapter I.2
Topological Dynamics
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.
Notions and Results
We first recall some notions and results from topological dynamics.

Definition 2.1 A continuous map f : X Ñ X is called a topological dynamical sys-
tem with discrete time. When f is a homeomorphism, that is, a bijective continuous
map with continuous inverse, it is also called an invertible topological dynamical
system.
Definition 2.2 A flow Φ (respectively, a semiflow Φ ) on X such that the map
pt, xq ÞÑ ϕt pxq is continuous on its domain is called a topological flow (respectively,
a topological semiflow). Any topological flow or semiflow is also called a topologi-
cal dynamical system with continuous time.
Definition 2.3 Given a map f : X Ñ X, the ω -limit set of a point x P X is defined by
č
ω pxq “ t f m pxq : m ě nu.
nPN
Moreover, when f is invertible, the α -limit set of a point x P X is defined by

č
α pxq “ t f ´m pxq : m ě nu.
nPN

16 I.2 Topological Dynamics
The following proposition gives a characterization of the ω -limit set.

Proposition 2.4 Given a map f : X Ñ X on a topological space, for each x P X the
following properties hold:
1. y P ω pxq if and only if there exists a sequence nk Õ 8 in N such that f nk pxq Ñ y
when k Ñ 8;
2. if f is continuous, then ω pxq is forward f -invariant.
When f is invertible, one can formulate a corresponding result for the α -limit set.
Proposition 2.5 Given an invertible map f : X Ñ X on a topological space, for each
x P X the following properties hold:
1. y P α pxq if and only if there exists a sequence nk Õ 8 in N such that f ńk pxq Ñ y
when k Ñ 8;
2. if f ´1 is continuous, then α pxq is backward f -invariant.
We also consider corresponding notions for continuous time.
Definition 2.6 Given a semiflow Φ on X, the ω -limit set of a point x P X is defined by
č
ω pxq “ tϕs pxq : s ą tu.
tą0
Moreover, given a flow Φ on X, the α -limit set of a point x P X is defined by

č
α pxq “ tϕs pxq : s ă tu.
tă0
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
ϕtk pxq Ñ y when k Ñ 8.
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
ϕtk pxq Ñ y when k Ñ 8.
Now we recall a few notions related to recurrence.

Definition 2.7 Given a map f : X Ñ X, a point x P X is said to be (forward) recurrent
for f if x P ω pxq. The set of recurrent points for f is denoted by Rp f q.
By Proposition 2.4, a point x P X is recurrent if and only if there exists a sequence
nk Õ 8 in N such that f nk pxq Ñ x when k Ñ 8.
Definition 2.8 Given a map f : X Ñ X on a topological space, a point x P X is said
to be nonwandering for f if for any open neighborhood U of x there exists n P N such
that f n pUq XU ‰ ∅. The set of nonwandering points for f is denoted by NW p f q.
I.2 Topological Dynamics 17
Definition 2.9 A map f : X Ñ X on a topological space is said to be:

1. topologically transitive if given nonempty open sets U,V Ĺ X, there exists an
integer n P N such that f ńU XV ‰ ∅;
2. topologically mixing if given nonempty open sets U,V Ĺ X, there exists an inte-
ger n P N such that f ´mU XV ‰ ∅ for all m ě n.
The remaining notions and results concern the topological entropy of a continu-
ous map on a compact metric space. It can be seen as a measure of the complexity
of the dynamics.
Definition 2.10 Let f : X Ñ X be a continuous map on a compact metric space with
distance d. For each n P N, we introduce a new distance dn “ dn, f on X by
(
dn px, yq “ max dp f k pxq, f k pyqq : 0 ď k ď n ´ 1 .
Let Npn, ε q “ N f pn, ε q be the largest number of points p1 , . . . , pm P X such that
dn ppi , p j q ě ε whenever i ‰ j.
The topological entropy of f is defined by

1
hp f q “ hd p f q “ lim lim sup log Npn, ε q.
ε Ñ0 nÑ8 n
Definition 2.11 Two maps f : X Ñ X and g : Y Ñ Y on topological spaces are said

to be topologically conjugate if there exists a homeomorphism H : X Ñ Y such that
H ˝ f “ g˝H on X,
that is, the diagram

f
X ÝÝÝÝÑ X
§ §
§ §
Hđ đH
g
Y ÝÝÝÝÑ Y
is commutative. Then H is called a topological conjugacy between f and g.
A topological conjugacy can be seen as a dictionary between two dynamics and
can sometimes be used to transfer a given property from one dynamics to the other.
It turns out that topologically conjugate continuous maps have the same topological
entropy.
Theorem 2.12 Let f : X Ñ X and g : Y Ñ Y be continuous maps on compact metric
spaces. If f and g are topologically conjugate, then hp f q “ hpgq.
Finally, we describe a few equivalents or alternative definitions of the topological
entropy.
Definition 2.13 Given n P N and ε ą 0:

1. let Mpn, ε q “ M f pn, ε q be the least number of points p1 , . . . , pm P X such that for
each x P X there exists i P t1, . . . , mu with dn px, pi q ă ε ;
2. let Cpn, ε q “ C f pn, ε q be the least cardinality of a cover of X by sets U1 , . . . ,Um
with (
sup dn px, yq : x, y P Ui ă ε for i “ 1, . . . , m.
Theorem 2.14 If f : X Ñ X is a continuous map on a compact metric space, then
1
hp f q “ lim lim inf log Npn, ε q
ε Ñ0 nÑ8 n
1
“ lim lim sup log Mpn, ε q
ε Ñ0 nÑ8 n
1
“ lim lim inf log Mpn, ε q
ε Ñ0 nÑ8 n
1
“ lim lim logCpn, ε q.
ε Ñ0 nÑ8 n
Problems
Problem 2.1 Show that if f : X Ñ X is a continuous map on a compact metric space,

then f pω pxqq “ ω pxq for all x P X.
Problem 2.2 Given a map f : X Ñ X on a metric space without isolated points,
show that if ω pxq ‰ X, then the positive semiorbit γ ` pxq is not dense.
Problem 2.3 Let f : X Ñ X be a continuous map. Given x P X, show that if ω pxq
contains infinitely many points, then it 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.
Problem 2.5 Let f : X Ñ X be a continuous map on a compact metric space. Given
x P X, show that if A “ ω pxq, then for any nonempty closed subset B Ĺ A we have
B X f pAzBq ‰ ∅.
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.
#
r1 “ rpr ´ 1q,
θ 1 “ 1,
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.
Problem 2.9 Let f : X Ñ X be a continuous map on a topological space and let

NW p f q be the set of nonwandering points. Show that:
1. NW p f q is closed;
2. NW p f q is forward f -invariant.
Problem 2.10 Let f : X Ñ X be a continuous map on a topological space. Show
that ω pxq Ď NW p f q for all x P X.
Problem 2.11 Let f : X Ñ X be a continuous map on a compact metric space. Show
that
lim dp f m pxq, NW p f qq “ 0
mÑ8
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 ńU 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.
Problem 2.20 Given a continuous map f : X Ñ X on a complete metric space with

a countable basis and without isolated points, show that the following properties are
equivalent:
1. f is topologically transitive;
2. the set of points with a dense positive semiorbit is dense.
Problem 2.21 Give an example of a continuous map f : X Ñ X on a finite set with
at least one dense positive semiorbit, but which is not topologically transitive.
Problem 2.22 Give an example of a continuous map f : X Ñ X on an infinite set
with at least one dense positive semiorbit, but which is not topologically transitive.
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.
Problem 2.24 Show that the map f in Problem 1.21 is topologically transitive.
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 .
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
p f ˆ f qpx, yq “ p f pxq, f pyqq
is also topologically mixing.

Problem 2.30 Let f : X Ñ X be a continuous map on a compact metric space pX, dq
such that
dp f pxq, f pyqq ď dpx, yq for all x, y P X.
Show that the topological entropy of f is zero.
Problem 2.31 Show that if f : X Ñ X is a homeomorphism on a compact metric

space pX, dq, then
N f pn, ε q “ N f ´1 pn, ε q
for all n P N and ε ą 0.
a closed forward f -invariant set Y Ď X, show that hp f q ě hp f |Y q.
Problem 2.33 Let f : X Ñ X and g : Y Ñ Y be continuous maps on compact metric
spaces, respectively, pX, dX q and pY, dY q satisfying f ˝ H “ H ˝ g for some continu-
ous onto map H : Y Ñ X. Show that hp f q ď hpgq.
spaces satisfying f ˝ H “ H ˝ g for some homeomorphism H : Y Ñ X. Show that
hp f q “ hpgq.
Problem 2.35 Compute the topological entropy of the map f in Problem 1.21.
Problem 2.36 Determine whether there exists a continuous map f : X Ñ X on a
compact metric space with infinite topological entropy.
Problem 2.37 Show that if the distances d and d 1 generate both the topologies of a
compact topological space X, then
hd p f q “ hd 1 p f q
for any continuous map f : X Ñ X.

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
# +
č
n´1
´k
Un “ f Uk : 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;
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
Mpn, δ {2q ě NpUn q;

3. if U is a finite open cover of X with

(
diam U :“ sup diamU : U P U ă ε ,
where diamU “ tdpx, yq : x, y P Uu, then Npn, ε q ď NpUn q;

4.
1
hp f q “ lim lim log NpUn q.
diam UÑ0 nÑ8 n
Problem 2.39 Let f : r0, 1s Ñ r0, 1s be a homeomorphism. Show that hp f q “ 0.
Problem 2.40 Let f : S1 Ñ S1 be a homeomorphism. Show that hp f q “ 0.
Chapter I.3
Low-Dimensional Dynamics
In this chapter we consider various dynamical systems on low-dimensional spaces

(namely of dimension 1 for discrete time and of dimension 2 for continuous time).
The reason for this separation is that the results and methods that can be used
with these dynamical systems, because of specific topological properties, fail on
higher-dimensional spaces. In particular, we consider homeomorphisms and diffeo-
morphisms of the circle, including the rotation number of an orientation-preserving
homeomorphism, continuous maps on a compact interval, and flows defined by au-
tonomous differential equations on the plane. We refer the reader to [2, 10, 22, 24]
for additional topics.
Notions and Results
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.
Proposition 3.2 Let f : S1 Ñ S1 be a homeomorphism. Then:

1. f has lifts;
2. if F and G are lifts of f , then there exists k P Z such that G ´ F “ k on R;
3. any lift of f is a homeomorphism on R.
Definition 3.3 A homeomorphism f : S1 Ñ S1 is said to be:
1. orientation-preserving if at least one of its lifts is an increasing function;
2. orientation-reversing if at least one of its lifts is a decreasing function.

24 I.3 Low-Dimensional Dynamics
Now we consider the notion of the rotation number of an orientation-preserving

homeomorphism.
Theorem 3.4 Let f : S1 Ñ S1 be an orientation-preserving homeomorphism. If F is
a lift of f , then for each x P R the limit
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.
We denote by tα u the fractional part of a number α P R.

Definition 3.5 The rotation number of an orientation-preserving homeomorphism
f : S1 Ñ S1 is defined by ρ p f q “ tρ pFqu for any lift F of f .
We note that the rotation number of an orientation-preserving homeomorphism f
is well defined, that is, the number tρ pFqu is the same for all lifts F of f .
Theorem 3.6 Let f : S1 Ñ S1 be an orientation-preserving homeomorphism. Then:
1. ρ p f q P Q if and only if f has at least one periodic point;
2. if ρ p f q “ p{q with pp, qq “ 1, then all periodic points of f have period q.
We also introduce the notion of a function with bounded variation.
Definition 3.7 Let P be the set of all partitions txk : k “ 0, . . . , nu of S1 composed of
points x0 ă x1 ă ¨ ¨ ¨ ă xn in S1 , with xn “ x0 , for some n P N. A function ϕ : S1 Ñ R
is said to have bounded variation if
ÿ
n
Var ϕ :“ sup |ϕ pxk q ´ ϕ pxk´1 q| ă `8.
P k“1
Theorem 3.8 (Denjoy) Let f : S1 Ñ S1 be an orientation-preserving C1 diffeomor-

phism whose derivative has bounded variation. If ρ p f q P RzQ, then f is topologi-
cally conjugate to the rotation Rρ p f q .
We also consider Sharkovsky’s theorem for a continuous map on a compact in-
terval. It relates the existence of periodic points with different periods with respect
to a certain ordering on N.
Definition 3.9 The Sharkovsky’s ordering ă on N is defined by
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
Theorem 3.10 (Sharkovsky) Let f : I Ñ I be a continuous map on a compact

interval I Ĺ R. If f has a periodic point with period p and q ă p, then f has a
periodic point with period q.
Finally, we formulate a result for autonomous differential equations on the plane.
It can sometimes be used to establish the existence of periodic orbits.
Theorem 3.11 (Poincaré–Bendixson) For the flow determined by a nonautono-
mous differential equation on R2 , if the positive semiorbit γ ` pxq of a point x P R2 is
bounded and ω pxq contains no critical points, then ω pxq is a periodic orbit.
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
f prxsq “ rx ` a sinp2π xqs
(see Figure I.3.2). Show that each orbit starting in p0, 1{2q is monotonous.
1 2
0 1 2 1
Problem 3.5 Let f : S1 Ñ S1 be a homeomorphism with a single periodic orbit.

Show that any other orbit is asymptotic to this periodic orbit (see Figure I.3.3 for an
example).
Problem 3.6 Let f : S1 Ñ S1 be an orientation-reversing homeomorphism and let F
be a lift of f . Show that Fpx`1q “ Fpxq´1 for all x P R (see Figures I.3.4 and I.3.5).
Problem 3.7 Let f : S1 Ñ S1 be an orientation-reserving homeomorphism. Show
that ρ p f 2 q “ 0.
Problem 3.8 Let f : S1 Ñ S1 be an orientation-preserving homeomorphism with
ρ p f q “ p{q for some integers p, q P N. Show that for each lift F of f there exists
x P R such that Fpxq ´ x ´ p{q P Z.
Problem 3.9 Let f , g : S1 Ñ S1 be homeomorphisms. Show that if F and G are lifts,
respectively, of f and g, then
FpGn pxqq ´ Gn pxq

lim “0
nÑ8 n
for each x P R.
1 2
0 1 2 1
Fig. I.3.3 Homeomorphism of the circle with the single periodic orbit t0, 1{2u.
0 1
Fig. I.3.4 An orientation-reversing homeomorphism of the circle.
Problem 3.10 Let f , g : S1 Ñ S1 be homeomorphisms and let F and G be lifts, re-

spectively, of f and g. Show that if f is an orientation-preserving homeomorphism,
then:
1. for each k P N we have
F n pxq ´ x F n pxn q ´ xn
lim “ lim
nÑ8 n nÑ8 n
for any x, xn P R with |x ´ xn | ď k for n P N;
Fig. I.3.5 A lift of an orientation-reversing homeomorphism.
2.
F n pGn pxqq ´ Gn pxq F n pxq ´ x

lim “ lim
nÑ8 n nÑ8 n
for each x P R.
r{s ă ρ p f q ă p{q ă 1
for some positive integers r, s, p, q. Show that:

1. any lift F of f with Fp0q P p0, 1q satisfies
F q pxq ă x ` p and F s pxq ą x ` r
for all x P R;
2. if g : S1 Ñ S1 is an orientation-preserving homeomorphism with
dp f , gq :“ max dp f pxq, gpxqq

xPS1
sufficiently small, then r{s ă ρ pgq ă p{q.

Problem 3.12 Let f : S1 Ñ S1 be an orientation-preserving homeomorphism. Show
that if ρ p f q P RzQ, then for any m, n P Z with m ‰ n we have
8
ď
S1 “ f ´k I, for I “ r f n pxq, f m pxqs.
k“0

that if ρ p f q P RzQ, then ω pxq “ ω pyq for any x, y P S1 .
that if there exists an open interval J Ĺ S1 such that the sets f k pJq, for k P Z, are
pairwise disjoint, then f is not conjugate to an irrational rotation.
ρ p f q P RzQ. Show that if there exists a minimal set K ‰ S1 (see Problem 2.14 for
the definition), then f is not conjugate to a rotation.
Problem 3.16 Show that the map h : S1 Ñ S1 given by
hpxq “ sin2 pπ x{2q
is a topological conjugacy between the maps f , g : S1 Ñ S1 defined by
f pxq “ 1 ´ |2x ´ 1| and gpxq “ 4xp1 ´ xq
(see Figures I.3.6 and I.3.7).

Problem 3.17 Determine whether the maps E4 and R1{2 are topologically conjugate.
Problem 3.18 Determine for which values of α P r0, 1q the rotations Rα and R2α
are topologically conjugate.
Problem 3.19 Show that the map f in Problem 3.16 is topologically mixing.
Problem 3.20 Show that the map g : S1 Ñ S1 given by gpxq “ 4xp1 ´ xq is topolog-
ically mixing.
Problem 3.21 Consider an interval I “ pa, bq Ĺ S1 with 0 ă a ă b ă 1. For the map g
in Problem 3.20, show that there exists n P N such that gn pIq “ S1 , identifying I with
trxs : x P Iu Ĺ S1 .
Problem 3.22 Let f : S1 Ñ S1 be an orientation-preserving homeomorphism and let
NW p f q be the nonwandering set for f . Show that if ρ p f q P RzQ, then NW p f q is a
minimal set.
Fig. I.3.7 The map g in Problem 3.16.


that if ρ p f q P RzQ, then either NW p f q “ S1 or NW p f q is a Cantor set (that is,
a closed set with empty interior and without isolated points).
irrational rotation number. Show that f has a unique minimal set K Ď S1 .
irrational rotation number. Show that either ω pxq is nowhere dense or ω pxq “ S1 .
Problem 3.26 Let f : S1 Ñ S1 be a C2 map. Show that its derivative has bounded
variation.
Problem 3.27 Let f : S1 Ñ S1 be a C1 diffeomorphism. Moreover, given n P N and
an interval I Ď S1 , let
|p f n q1 pxq|
Dn pIq “ sup log n 1 .
x,yPI |p f q pyq|
Show that
ÿ
n´1
Dn pIq ď D1 p f i pIqq.
i“0
Problem 3.28 Let f : S1 Ñ S1

be an orientation-preserving C1 diffeomorphism with
irrational rotation number. Show that if there exist a sequence nk P N with nk Õ 8
and a constant C ą 0 such that
|p f nk q1 pxq| ¨ |p f ńk q1 pxq| ą C
for all x P S1 and k P N, then every orbit of f is dense.

Problem 3.29 Consider the piecewise linear map f : r1, 3s Ñ r1, 3s with
f p1q “ 2, f p2q “ 3 and f p3q “ 1
(see Figure I.3.8). Show that f has period points with all periods.
Problem 3.30 Show that the map f : r0, 1s Ñ r0, 1s defined by

#
2x if x P r0, 1{2s,
f pxq “
2p1 ´ xq if x P r1{2, 1s
(see Figure I.3.6) 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.
1 2 3
Problem 3.33 Verify that the differential equation

#
x1 “ xy ´ x3 ,
y1 “ x ` y ´ x2
has no periodic solutions contained in the second quadrant.

#
r1 “ 4 ´ 3r cos θ ´ r4 ,
θ 1 “ ´1,
written in polar coordinates, has at least one periodic solution.

#
r1 “ rp1 ´ rq,
θ 1 “ 1 ` cos2 θ
has at least one periodic solution.

Problem 3.36 Consider the function f : R2 Ñ R2 defined by
` ˘
f px, yq “ y, ´x ` yp1 ´ 3x2 ´ 6y2 q
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 .
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.
#
x1 “ y,
y1 “ ´ f pxqy ´ gpxq
for some C1 functions f , g : R Ñ R with f ą 0. Show that the equation has no

periodic solutions.
#
x1 “ f px, yq,
y1 “ gpx, yq
on R2 . Show that if there exists a C1 function ϕ : R2 Ñ R such that
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
has no periodic solutions.

Chapter I.4
Hyperbolic Dynamics
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.
Notions and Results
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
for each x P Λ such that:

1.
dx f E s pxq “ E s p f pxqq and dx f E u pxq “ E u p f pxqq;
2 if v P E s pxq and n P N, then
dx f n v ď cλ n v;
3. if v P E u pxq and n P N, then
dx f ń v ď cλ n v.
The spaces E s pxq and E u pxq are called, respectively, the stable and unstable spaces
at x.
36 I.4 Hyperbolic Dynamics
Given E Ď R p and v P R p , let
dpv, Eq “ min v ´ w : w P E .
(
The distance between two subspaces E, F Ď R p is defined by

" *
dpE, Fq “ max max dpv, Fq, max dpw, Eq .
vPE,v“1 wPF,w“1
Theorem 4.2 If Λ Ĺ R p is a hyperbolic set for a diffeomorphism f : R p Ñ R p , then

the spaces E s pxq and E u pxq vary continuously with x P Λ , that is, if xm Ñ x when
m Ñ 8, with xm , x P Λ for each m P N, then
dpE s pxm q, E s pxqq Ñ 0 when m Ñ 8
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
H1 “ r0, 1s ˆ r0, as and H2 “ r0, 1s ˆ r1 ´ a, 1s,
and the vertical strips
V1 “ r0, as ˆ r0, 1s and V2 “ r1 ´ a, 1s ˆ r0, 1s,
for some constant a P p0, 1{2q. We assume that f is a C1 diffeomorphism on an open

neighborhood of the square Q “ r0, 1s2 such that
f pH1 q “ V1 and f pH2 q “ V2
(see Figure I.4.1), with the restriction of f to H1 Y H2 given by

#
pax, y{aq if px, yq P H1 ,
f px, yq “
páx ` 1, ý{a ` 1{aq if px, yq P H2 .
The set č č
Λ“ f n pQq “ f n pH1 Y H2 q
nPZ nPZ
is called Smale horseshoe.

We also consider the characterization of a hyperbolic set in terms of invariant
families of cones.
Definition 4.4 Let f : M Ñ M be a C1 diffeomorphism and let Λ Ď M be a compact
f -invariant set. For each x P Λ , consider a splitting
I.4 Hyperbolic Dynamics 37
H2
H1 V1 V2
Fig. I.4.1 Horizontal and vertical strips.
Tx M “ F s pxq ‘ F u pxq
and an inner product x¨, ÿ1 “ x¨, ÿ1x 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¨, ÿ1 . Given γ P p0, 1q and x P Λ , we define cones by
Cs pxq “ pv, wq P F s pxq ‘ F u pxq : w1 ă γ v1 Y t0u

(
and
Cu pxq “ pv, wq P F s pxq ‘ F u pxq : v1 ă γ w1 Y t0u.
(
Theorem 4.5 Let f : M Ñ M be a C1 diffeomorphism and let Λ Ď M be a compact

f -invariant set. Then Λ is a hyperbolic set for f if and only if there exist a splitting
Tx M “ F s pxq ‘ F u pxq and an inner product x¨, ÿ1x on Tx M, for each x P Λ , and there
exist constants μ , γ P p0, 1q such that for each x P Λ :
1.
dx fCu pxq Ĺ Cu p f pxqq and dx f ´1Cs pxq Ĺ Cs p f ´1 pxqq;
2.
dx f v1 ě μ ´1 v1 for v P Cu pxq;
3.
dx f ´1 v1 ě μ ´1 v1 for v P Cs pxq.
Now we formulate two main results of the theory. We first introduce the notion
of a local topological conjugacy.
Definition 4.6 Two maps f : X Ñ X and g : Y Ñ Y on topological spaces are said

to be locally topologically conjugate, respectively, on open sets U Ď X and V Ď Y
if there exists a homeomorphism h : U Ñ V with hpUq “ V such that h ˝ f “ g ˝ h
on U.
The following result provides a local topological conjugacy to the linear part in
the neighborhood of a hyperbolic fixed point, that is, a fixed point x such that txu is
a hyperbolic set.
Theorem 4.7 (Grobman–Hartman) Let x P R p be a hyperbolic fixed point of a
C1 diffeomorphism f : R p Ñ R p . Then the maps f and dx f are locally topologically
conjugate, respectively, on open neighborhoods U of x and V of 0, that is, there
exists a homeomorphism h : U Ñ V with hpUq “ V such that
h ˝ f “ dx f ˝ h on U
E s (x) x
E u (x)
Fig. I.4.2 Behavior of the orbits of dx f .
Let Λ be a hyperbolic set for a C1 diffeomorphism f : R p Ñ R p . Given ε ą 0, for

each x P Λ we denote by Bpx, ε q the ball of radius ε centered at x and we consider
the sets
V s pxq “ y P Bpx, ε q : f n pyq ´ f n pxq ă ε for n ą 0
(
and
V u pxq “ y P Bpx, ε q : f n pyq ´ f n pxq ă ε for n ă 0 .
(
V s (x)
V u (x)
Fig. I.4.3 Behavior of the orbits of f .
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
TxV s pxq “ E s pxq and TxV u pxq “ E u pxq;
2. there exist ρ P p0, 1q and C ą 0 such that
f n pyq ´ f n pxq ď Cρ n y ´ x for y P V s pxq
and
f ń pyq ´ f ń 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
Tx M “ E s pxq ‘ E 0 pxq ‘ E u pxq
for each x P Λ such that:

1. E 0 pxq is the 1-dimensional space generated by the vector dt ϕt pxq|t“0 ;
d
2. for each t P R we have
dx ϕt E s pxq “ E s pϕt pxqq and dx ϕt E u pxq “ E u pϕt pxqq;
3. if v P E s pxq and t ą 0, then
dx ϕt v ď cλ t v;
4. if v P E u pxq and t ą 0, then
dx ϕ´t 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
Fig. I.4.4 The Möbius transformation T .
Finally, we consider the upper half plane
H “ z P C : Im z ą 0
(
and its unit tangent bundle
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
T piq “ z and T 1 piqi “ v

(see Figure I.4.4). More precisely, let x, y P R Y t8u be, respectively, the limits of
γ ptq “ T piet q when t Ñ ´8 and when t Ñ `8. Then:
1. for x, y P R with x ă y, we have
α yw ` x z´x

T pwq “ , with α “ ;
αw ` 1 zý
2. for x, y P R with x ą y, we have
yw ´ α x zý

T pwq “ , with α “ ;
w´α z´x
3. for x P R and y “ 8, we have
T pwq “ α w ` x, with α “ Im z;
4. for x “ 8 and y P R, we have

α
T pwq “ ´ ` y, with α “ Im z.
w
Definition 4.10 Let γ ptq “ T piet q with T as above. The geodesic flow ϕt : SH Ñ SH
on the unit tangent bundle of the upper half plane is defined by
ϕt pz, vq “ pγ ptq, γ 1 ptqq.
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 ń 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 ń 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.
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
An v ď cλ n v for v P E s , n P N
and
Ań 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 ńVin
č
Λω “
nPZ
contains exactly one point.

Problem 4.7 For the Smale horseshoe Λ , show that the map H : t1, 2uZ Ñ Λ de-
fined by
f ńVin
č
Hp¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q “
nPZ
is bijective (by Problem 4.6 the map is well defined).

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 .
Problem 4.9 Consider horizontal strips
Hi “ px, yq P r0, 1s2 : ϕi pxq ď y ď ψi pxq

(
and vertical strips
Vi “ px, yq P r0, 1s2 : ϕi pyq ď x ď ψi pyq ,

(
for some functions ϕi , ψi , ϕi , ψi : r0, 1s Ñ r0, 1s, for i “ 1, 2, such that
ϕ1 ă ψ1 ă ϕ2 ă ψ2
and
(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
f px, yq “ pgpxq, hpyqq for px, yq P H1 Y H2
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
is a hyperbolic set for f .
H̄2
H̄1 V̄1 V¯2
Fig. I.4.5 Horizontal and vertical strips in Problem 4.9.
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
for any compact interval I Ĺ R.

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 .
Problem 4.13 Given functions f , g : R Ñ R, let
dp f , gq “ sup | f pxq ´ gpxq| : x P R .

(
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
dp f , gq “ sup | f pxq ´ gpxq| : x P ra, bs .

(
Problem 4.15 Construct a sequence of bounded C1 functions fn : R Ñ R converg-

ing uniformly to a nondifferentiable function.
Problem 4.16 Let Λ be a compact set that is invariant under a C1 diffeomorphism f

and assume that there exist a splitting Tx M “ F s pxq ‘ F u pxq for each x P Λ , an inner
product x¨, ÿ1x on Tx M varying continuously with x P Λ and a constant γ P p0, 1q such
that
dx f v1 ą v1 for x P Λ , v P Cu pxqzt0u.
Show that there exists a constant μ P p0, 1q such that
dx f v1 ě μ ´1 v1 for x P Λ , v P Cu pxq.
Problem 4.17 Consider the set
A “ S1 ˆ px, yq P R2 : x2 ` y2 ď 1
(
and the map f : A Ñ S1 ˆ R2 defined by

ˆ ˙
1 1 1 1
f prθ s, x, yq “ r2θ s, x ` cosp2πθ q, y ` sinp2πθ q .
5 2 5 2
Show that f pAq Ĺ A and that f |A is one-to-one.

Problem 4.18 For the map f |A in Problem 4.17, show that Λ “ f n pAq is a
Ş
nPN
hyperbolic set for f .
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.
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.
Problem 4.21 Construct a topological conjugacy between the maps f , g : R2 Ñ R2
given by
f px, yq “ p2x, 3yq and gpx, yq “ p5x, 4yq
for px, yq P R2 .
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
31
B“ .
11
Problem 4.24 Show that the vector Xpxq “ dtd ϕt pxqˇt“0 in Definition 4.9 is neither
ˇ
contracted nor expanded by dx ϕt .

Problem 4.25 Let Λ be a hyperbolic set for a flow Φ on R p . Given a sequence

xm Ñ x when m Ñ 8, with xm , x P Λ , for each m P N, show that any sublimit of a
sequence vm P E s pxm q Ď R p with vm “ 1 is in E s pxq.
Problem 4.26 For a sequence xm as in Problem 4.25, show that there exists m P N
such that
dim E s px p q “ dim E s pxq q and dim E u px p q “ dim E s pxq q
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 “ ý ` x2 .
x1 “ xpa ´ yq,
y1 “ ý ` 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.
’
%
is a topological conjugacy between the flows determined by the differential equa-

tions
x1 “ ax and x1 “ bx.
In other words, if ϕt and ψt are the flows determined by the equations, then
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.
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
for px, yq ‰ p0, 0q satisfies

h ˝ ϕt “ ψt ˝ h.
Problem 4.33 For the differential equation
#
x1 “ ´x,
y1 “ y ` x2 ,
show that the map h : R2 Ñ R2 defined by
x2
ˆ ˙
hpx, yq “ x, y `
3
is a topological conjugacy between the flows determined by the equation and its
linearization at the origin.
#
x1 “ x,
y1 “ ý ` xn ,
with n ě 2, show that the map h : R2 Ñ R2 defined by
xn
ˆ ˙
hpx, yq “ x, y ´
n`1
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
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
In this chapter we consider various topics of symbolic dynamics, with emphasis

on their relation to hyperbolic dynamics. This includes expanding maps, the Smale
horseshoe, and its modifications. We also consider topological Markov chains and
their periodic points, irreducible and transitive matrices, topological transitivity, and
topological mixing as well as the notion of topological entropy. We refer the reader
to [16, 33, 34] for additional topics.
Notions and Results
We first recall a few basic notions.

Definition 5.1 Given an integer k ą 1, consider the set Σk` “ t1, . . . , kuN of one-
sided sequences
ω “ pi1 pω qi2 pω q ¨ ¨ ¨ q
with in pω q P t1, . . . , ku for each n P N. The one-sided shift map σ : Σk` Ñ Σk` is
defined by
σ pω q “ pi2 pω qi3 pω q ¨ ¨ ¨ q.
Given β ą 1, for each ω , ω 1 P Σk` let n “ npω , ω 1 q P N be the smallest positive
integer such that in pω q ‰ in pω 1 q and define
#
1 β ń if ω ‰ ω 1 ,
dβ pω , ω q “
0 if ω “ ω 1 .
One can show that the function dβ is a distance on Σk` .
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.

50 I.5 Symbolic Dynamics
Definition 5.3 Given an integer k ą 1, let A “ pai j q be a k ˆ k matrix with entries

ai j in t0, 1u and consider the set
(
ΣA` “ ω P Σk` : ain pω qin`1 pω q “ 1 for n P N .
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
σ pω q “ ω 1 , with in pω 1 q “ in`1 pω q for n P Z.
Given β ą 1, for each ω , ω 1 P Σk let n “ npω , ω 1 q P N be the smallest integer

such that
in pω q ‰ in pω 1 q or iń pω q ‰ iń pω 1 q,
and define #
1 β ń if ω ‰ ω 1 ,
dβ pω , ω q “
0 if ω “ ω 1 .
One can show that the function dβ is a distance on Σk .
Definition 5.6 Given an integer k ą 1, let A “ pai j q be a k ˆ k matrix with entries
ai j in t0, 1u and consider the set
(
ΣA “ ω P Σk : ain pω qin`1 pω q “ 1 for n P Z .
The restriction σ |ΣA : ΣA Ñ ΣA of σ |Σk to ΣA is called the two-sided topological

Markov chain with transition matrix A.
Finally, we recall a few additional notions and results, here formulated for sim-
plicity only for the one-sided case.
Definition 5.7 A k ˆ k matrix A is called:
1. irreducible if for each i, j P t1, . . . , ku there exists m “ mpi, jq P N such that the
entry pi, jq of Am is positive;
2. transitive if there exists m P N such that all entries of the matrix Am are positive.
Proposition 5.8 If the matrix A is irreducible (respectively, transitive), then the
topological Markov chain σ |ΣA` is topologically transitive (respectively, topolog-
ically mixing).
I.5 Symbolic Dynamics 51
Proposition 5.9 If the matrix A is transitive, then hpσ |ΣA` q ą 0.

We also recall the notion of the zeta function of a dynamical system with discrete
time.
Definition 5.10 Given a map f : X Ñ X with
(
an :“ card x P X : f n pxq “ x ă `8
for each n P N, its zeta function is defined by
ÿ8
an zn
ζ pzq “ exp
n“1
n
for each z P C such that the series converges.
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 ¨¨ïn :“ ω P Σk` : i j pω q “ i j for j “ 1, . . . , n .
Problem 5.2 Show that any cylinder Ci1 ¨¨ïn (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
and that it generates the same topology as the distance dβ .

#
1 1{npω , ω 1 q if ω ‰ ω 1 ,
dpω , ω q “
0 if ω “ ω 1

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
Fig. I.5.1 Graph associated to the matrix A in Problem 5.14.

Problem 5.14 Determine whether the matrix

¨ ˛
001
A “ ˝1 1 0‚
111
is irreducible or transitive (see Figure I.5.1 for its graph).

Problem 5.15 Show that the matrix
¨ ˛
0 0 1 0
˚0 0 0 1‹
A“˚
˝0
‹
1 0 0‚
0 1 1 0
is not transitive (see Figure I.5.2 for its graph).
1 3
4 2
Fig. I.5.2 Graph associated to the matrix A in Problem 5.15.
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
has positive topological entropy.

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
1. Compute hpσ |ΣA` q and hpσ |ΣB` q.

2. Find the fixed points of σ |ΣA` and σ |ΣB` .
Problem 5.18 Find topological Markov chains σ |ΣA` and σ |ΣB` both with zero
topological entropy that are not topologically conjugate.
Problem 5.19 Find topological Markov chains σ |ΣA` and σ |ΣB` with
hpσ |ΣA` q “ hpσ |ΣB` q ‰ 0
that are not topologically conjugate.

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 ń Vin
nPZ
defines a bijective map H : t1, 2uZ Ñ Λ .

Problem 5.21 Show that the bijection H : Σ2 Ñ Λ in Problem 5.20 is a topological
conjugacy between the shift map σ and the map f in Problem 4.9.
Problem 5.22 Show that the set
(
ω P Σk : ω is q-periodic for some even q P N
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.
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
Fig. I.5.3 An expanding map f : S1 Ñ S1 .
Problem 5.32 An orientation-preserving C1 map f : S1 Ñ S1 is called an expanding

map if
λ :“ min f 1 pxq ą 1
xPS1
(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
Ii0 ...in´1 “ rp j , p j`1 s, for j “ 0, . . . , 2n ´ 1,
satisfy:
1. |Ii0 ¨¨ïn´1 | ď λ ń ;
2. Ii0 ...in´1 in Ĺ Ii0 ...in´1 ; and
3. f m pIi0 ¨¨ïn q “ Iim ¨¨ïn for each m ď n.
f (x)
p1
Fig. I.5.4 Preimages of the point p under f 2 .
Problem 5.33 Let f : S1 Ñ S1 be an expanding Ť map (see Problem 5.32) with

deg f “ 2. Show that the set of all preimages nPN f ń 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` .
Problem 5.35 Let f : S1 Ñ S1 be an expanding map with deg f “ k ě 2. Show that
there exists a continuous onto map H : Σk` Ñ S1 such that H ˝ σ “ f ˝ H on Σk` .
Problem 5.36 Show that the map H in Problem 5.35 is not a homeomorphism.
Problem 5.37 Show that the periodic points of any expanding map f : S1 Ñ S1 are
dense in S1 .
Problem 5.38 Show that any expanding map f : S1 Ñ S1 is topologically transitive.
Problem 5.39 Determine whether any expanding map f : S1 Ñ S1 is topologically
mixing.
Problem 5.40 For the map f in Problem 5.35, show that hp f q ď log k.
Chapter I.6
Ergodic Theory
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.
Notions and Results
We start by recalling some basic notions of measure theory. Let X be a set.

Definition 6.1 A set A of subsets of X is called a σ -algebra on X if:
1. ∅, X P A;
XzB P A whenever B P A;
2. Ť
3. 8n“1 Bn P A whenever Bn P A for all n P N.
Given a set A of subsets of X, the σ -algebra generated by A is the smallest σ -

algebra on X that contains all elements of A. In particular, the Borel σ -algebra on R,
denoted by B, is the σ -algebra generated by the open intervals of R. We shall always
assume that R is equipped with the Borel σ -algebra B. More generally, the Borel
σ
ś-algebra on Rn , denoted by Bn , is the σ -algebra generated by the open rectangles
n
i“1 i i q, with ai ă bi for i “ 1, . . . , n. Clearly, B1 “ B.
pa , b
Definition 6.2 Given a σ -algebra A on X, a function μ : A Ñ r0, `8s is called a
measure on X (with respect to A) if μ p∅q “ 0 and
˜ ¸
8
ď 8
ÿ
μ Bn “ μ pBn q
n“1 n“1

58 I.6 Ergodic Theory
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
for all ai ă bi and i “ 1, . . . , n. It is called the Lebesgue measure on Rn .

We assume from now on in this section that we are given some measure space
pX, A, μ q.
Definition 6.3 A function ϕ : X Ñ R is said to be A-measurable if ϕ ´1 B P A for
every B P B (where B is the Borel σ -algebra on R).
To introduce the notion of integral of a measurable function, we first consider the
class of simple functions. The characteristic function χB : X Ñ t0, 1u of a set B Ď X
is defined by #
1 if x P B,
χB pxq “
0 if x R B.
Definition 6.4 Given sets B1 , . . . , Bn P A and numbers a1 , . . . , an P R, the function
ÿ
n
s“ ak χBk
k“1
is called a simple function.

Note that any simple function is measurable. Now we introduce the notion of
integral of a nonnegative measurable function.
Definition 6.5 The Lebesgue integral of a measurable function ϕ : X Ñ R`
0 is
defined by # +
ż ÿ
n ÿ
n
ϕ d μ “ sup ak μ pBk q : ak χBk ď ϕ .
X k“1 k“1
The integral of an arbitrary measurable function can now be introduced as fol-

lows. Given a function ϕ : X Ñ R, let
ϕ ` “ maxtϕ , 0u and ϕ ´ “ maxt´ϕ , 0u.
Definition 6.6 An A-measurable function ϕ : X Ñ R is said to be μ -integrable if

ż ż
ϕ ` d μ ă 8 and ϕ ´ d μ ă 8.
X X
The Lebesgue integral of a μ -integrable function ϕ : X Ñ R is defined by

ż ż ż
ϕ dμ “ ϕ dμ ´ ϕ´ dμ.
`
X X X
I.6 Ergodic Theory 59
We also consider the class of measurable maps.

Definition 6.7 A map f : X Ñ X is said to be A-measurable if f ´1 B P A for every
B P A.
Definition 6.8 Given an A-measurable map f : X Ñ X, a measure μ on X is said
to be f -invariant if
μ p f ´1 Bq “ μ pBq for all B P A.
Then we also say that f preserves μ .
Now we formulate two basic but also fundamental results of ergodic theory.
Theorem 6.9 (Poincaré’s recurrence theorem) Let f : X Ñ X be an A-measurable
map preserving a finite measure μ on X. For each set A P A, we have
` (˘
μ x P A : f n pxq P A for infinitely many integers n P N “ μ pAq.
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
exists for almost every x P X, the function ϕ f is μ -integrable and

ż ż
ϕ f dμ “ ϕ dμ.
X X
Finally, we introduce the notion of metric entropy of a measurable map with

respect to an invariant probability measure.
Definition 6.11 A finite set ξ Ď A is called a partition of X (with respect to μ ) if:
`Ť ˘
1. μ CPξ C “ 1;
2. μ pC X Dq “ 0 for any C, D P ξ with C ‰ D.
Given a partition ξ of X, we define
ÿ
Hμ pξ q “ ´ μ pCq log μ pCq,
CPξ
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 .
Definition 6.12 Given an A-measurable map f : X Ñ X preserving a probability

measure μ on X, for each partition ξ of X let
1
hμ p f , ξ q “ inf Hμ pξn q.
nPN n
Definition 6.13 Given an A-measurable map f : X Ñ X preserving a probability

measure μ on X, the metric entropy of f with respect to μ is defined by
hμ p f q “ sup hμ p f , ξ pnq q
nPN
for any sequence ξ pnq of partitions of X such that:

1. given C P ξ pnq , there exist C1 , . . . ,Cm P ξ pn`1q satisfying
ˆ ďm ˙ ˆď
m ˙
μ Cz Ci “ μ Ci zC “ 0;
i“1 i“1
Ť pnq
2. the union nPN ξ generates the σ -algebra A.
One can show that the supremum giving hμ p f q does not depend on each specific
sequence of partitions ξ pnq .
Problems
Problem 6.1 For a measure space pX, A, μ q, show that:

1. if A, B P A and A Ď B, then μ pAq ď μ pBq;
2. if Bn P A for all n P N, then
˜ ¸
8
ď 8
ÿ
μ Bn ď μ pBn q.
n“1 n“1

1. if A, B P B, then
` ˘ ` ˘
μ pAq ` μ pBq “ μ A Y B ` μ A X B ;
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:
1. ra, bs with a ă b;
2. pa, bs with a ă b.
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.
lim sup ϕn and lim inf ϕn

nÑ8 nÑ8
(here assumed to be finite everywhere) are also A-measurable functions.

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
ϕn pxq Ñ ϕ pxq when n Ñ 8
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
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.
Fig. I.6.1 A map f as in Problem 6.14.

#
x{p1 ´ xq if x P r0, 1{2s,
f pxq “
p1 ´ xq{x if x P r1{2, 1s
(see Figure I.6.2) preserves the measure μ on r0, 1s defined by

ż
dx
μ pAq “ .
A x
Problem 6.16 Consider the Gauss map f : r0, 1s Ñ r0, 1s defined by

#
t1{xu if x ‰ 0,
f pxq “
0 if x “ 0
(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
Problem 6.18 Given α P R, show that the map f : T2 Ñ T2 defined by
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
Fig. I.6.3 The Gauss map.
is the continued fraction of a number x P p0, 1q, then:

1. f pxq “ ra2 , a3 , . . .s;
2. letting
p´1 “ 0, p0 “ 1, pm “ am pm´1 ` pm´2
and
q´1 “ 1, q0 “ 0, qm “ am qm´1 ` qm´2
for m P N, we have
pm qm´1 ´ pm´1 qm “ p´1qm´1 for m ě 0;
3. for m P N we have
1 pm
ra1 , a2 , . . . , am s :“ “ .
a1 ` 1
1
qm
a2 `
¨¨¨` a1m
Problem 6.20 Let P “ ppi j q be a k ˆ k matrix with entries pi j ě 0 for i, j “ 1, . . . , k

such that
ÿk
pi j “ 1 for j “ 1, . . . , k.
j“1
Moreover, take p1 , . . . , pk P p0, 1q such that

ÿ
k ÿ
k
pi “ 1 and pi pi j “ p j
i“1 i“1
for j “ 1, . . . , k. We define a measure μ on the σ -algebra on Σk` generated by the

cylinders (see Problem 5.1) by
μ pCi1 ¨¨ïn q “ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in
for each n P N and i1 , . . . , in P t1, . . . , ku. Show that:

ř
1. μ pCi1 ¨¨ïn q “ kj“1 μ pCi1 ¨¨ïn j q;
2. μ is σ -invariant;
3. μ pΣk` q “ 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.
Show that supp μ “ ΣA` .

Problem 6.22 Let f : X Ñ X be an A-measurable map preserving a finite measure μ
on X. Given a set A P A with μ pAq ą 0, show that there exist integers m, n ě 0 with
m ‰ n such that
f ´m A X f ń A ‰ ∅.
on X. Given a set A P A with μ pAq ą 0, show that the map τA : X Ñ N given by
τA pxq “ mintn P N : f n pxq P Au
is well defined for almost every x P A.

Problem 6.24 Show that a function ϕ : X Ñ R satisfies ϕ ˝ f “ ϕ for some map
f : X Ñ X if and only if the set ϕ ´1 α is f -invariant for every α P R.
Problem 6.25 Given α P Q, consider the rotation Rα : S1 Ñ S1 . Let U “ ra, bs be
an interval in S1 with 0 ď a ă b ď 1 and define
(
Tn pxq “ card 1 ď k ď n : Rkα pxq P U .
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
2. the former property holds for all continuous functions ϕ : S1 Ñ R;

3. for each x P S1 we have
Tn pxq
lim “ b ´ a.
nÑ8 n
Problem 6.26 Let f : X Ñ X be an A-measurable map. A probability measure μ

on X is said to be ergodic (with respect to f ) if any f -invariant set has either mea-
sure 0 or measure 1. Show that if
1 ÿ ` ´k ˘
n´1
lim μ f A X B “ μ pAqμ pBq
nÑ8 n
k“0
for any sets A, B P A, then the measure μ is ergodic.

Problem 6.27 Let f : X Ñ X be an A-measurable map. Show that an f -invariant
measure μ on X is ergodic (see Problem 6.26) if and only if
˜ ¸
8
ď
ń
μ f A “1
n“1
for any set A P A with μ pAq ą 0.

Problem 6.28 Let f : X Ñ X be an A-measurable map preserving an ergodic prob-
ability measure μ on X. Given a set A P A with μ pAq ą 0, show that
ż
τA d μ “ 1
A
with the function τA as in Problem 6.23.

on X is said to be mixing (with respect to f ) if
` ˘
lim μ f ń A X B “ μ pAqμ pBq
nÑ8
for any sets A, B P B and is said to be weakly mixing (with respect to f ) if
1 ÿ ` í ˘
n´1
lim μ f A X B ´ μ pAqμ pBq “ 0
nÑ8 n
i“0
for any sets A, B P B. Show that if μ is mixing, then it is weakly mixing.

Problem 6.30 Let f : X Ñ X be an A-measurable map. Show that if a probability
measure μ on X is weakly mixing (see Problem 6.29), then it is ergodic.
Problem 6.31 Let f : X Ñ X be an A-measurable map preserving a mixing proba-
bility measure μ on X (see Problem 6.29). Show that f is topologically mixing on
the support supp μ of μ .
Problem 6.32 Let μ be a probability measure on X. Show that if ξ is a partition

of X, then
Hμ pξ q ď log card ξ .
Problem 6.33 Let μ be a probability measure on X and let ξ and η be partitions
of X. Show that if η is a refinement of ξ (that is, if for each D P η there exists C P ξ
such that μ pDzCq “ 0), then
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.
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 í ξ “ lim Hμ f í ξ .
nPN n nÑ8 n
i“0 i“0
Problem 6.37 Let μ be a probability measure on X and let ξ , ζ and η be partitions

of X. Show that if ζ is a refinement of η (see Problem 6.33) and
ÿ μ pC X Dq
Hμ pξ |ζ q “ ´ μ pC X Dq log ,
μ pDq
CPξ ,DPζ
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 6.39 For the measure μ in Problem 6.20, compute hμ pσ q.

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.
Part II
Problems and Solutions
Chapter II.1
Basic Theory
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.
The (real) solutions are:

1. x “ 0 when a “ 0;
2. x “ 0 and x “ p2{aq1{3 when a ‰ 0.
Hence, the map f has nonzero fixed points if and only if a P Rzt0u.
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.
We have g1 pxq “ ex ´ 1 ą 0 for x ą 0 and so
ex ´ x “ gpxq ą gp0q ą 0 for x ą 0.
This shows that f has no fixed points in R` . Finally, one can show by induction that
f n pxq ą f n´1 pxq ą ¨ ¨ ¨ ą x
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.
72 II.1 Basic Theory
Solution Take p such that f 2 ppq “ p. We have q “ f ppq ‰ p. Now let
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
f pcq ´ c ď 0 and f pdq ´ d ě 0,
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.
Solution Note that

f paq ě a and f pbq ď b
(see Figure II.1.2). Proceeding as in Problem 1.4, 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 .
a b
Fig. II.1.2 A map f satisfying ra, bs Ě f pra, bsq.
Problem 1.6 Consider the continuous map f : r1, 5s Ñ r1, 5s with
f p1q “ 3, f p2q “ 5, f p3q “ 4, f p4q “ 2 and f p5q “ 1
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.
Solution 1. One can easily verify that
t1, 3, 4, 2, 5u (II.1.1)
is a periodic orbit with period 5.

1 2 3 4 5
Fig. II.1.3 The map f 3 .
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
f 2 pxq P r4, 5s and f 3 pxq P r1, 2s,
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 1.7 Let f : I Ñ I be a strictly increasing map on an interval I Ď R. Show

that any periodic point of f is a fixed point.
Solution We proceed by contradiction. Assume that x P I is a periodic point of f

that is not a fixed point. Then
x ą f pxq or x ă f pxq.
We consider only the first case since the second case is completely analogous. So
assume that x ą f pxq. Since f is strictly increasing, proceeding inductively we ob-
tain
x ą f pxq ą f 2 pxq ą ¨ ¨ ¨ ą f n pxq ą ¨ ¨ ¨
and so x ‰ f n pxq for all n P N. But this is impossible because x was assumed to be a
periodic point. This contradiction establishes the desired property.
Problem 1.8 Let f : I Ñ I be a strictly decreasing map on an interval I Ď R. Show

that any periodic point of f is either a fixed point or a periodic point with period 2.
Solution Since f is decreasing, the map g “ f 2 is increasing. Let x be a periodic

point of f with even period. Then x is also a periodic point of g. By Problem 1.7,
it is in fact a fixed point of g and so, all periodic points of f with even period have
period 2.
Now let x be a periodic point of f with odd period greater than 2. Then x is also
a periodic point of g, with the same odd period. However, by Problem 1.7, this is
impossible and so all periodic points of f with odd period are fixed points. This
establishes the desired property.
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.
Solution Note that x has period 2n with respect to f if and only if
f 2n pxq “ x and f i pxq ‰ x for i “ 1, . . . , 2n ´ 1.
Similarly, x has period n with respect to f 2 if and only if
f 2n pxq “ x and f 2i pxq ‰ x for i “ 1, . . . , n ´ 1.
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 .
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,
p f n qm{pm,nq pxq “ p f m qn{pm,nq pxq “ x,
which shows that p divides m{pm, nq. Therefore, p “ m{pm, nq.

Problem 1.13 Given positive integers n and k, show that if x is a periodic point of f n
with period k, then it is a periodic point of f with period kn{l for some factor l of n
with pk, lq “ 1.
Solution Since
x “ p f n qk pxq “ f kn pxq,
the period of x with respect to f is kn{l for some positive integer l. By Problem 1.12,
we have
kn{l
“ k.
pkn{l, nq
Therefore, ń n ¯ n
n
“ pkn{l, nq “ k, l “ pk, lq,
l l l l
which shows that pk, lq “ 1 and so l divides n (since it divides kn).
Problem 1.14 Let f : I Ñ I be a map on a closed interval I Ď R. Show that if the
positive semiorbit γ ` pxq of a point x P I is dense in I, then the set of points in I with
a dense positive semiorbit is dense in I.
Solution Since γ ` pxq is dense in I and the interval has no isolated points, for each
n P N the set (
γ ` pxqz x, f pxq, . . . , f n´1 pxq
is also dense in I. In other words, the positive semiorbit of each element of γ ` pxq is
dense in I. This shows that the set of points with a dense positive semiorbit is dense
in I.
Problem 1.15 Let f : X Ñ X be a continuous one-to-one map on a compact set.

Show that if the set P of periodic points of f is dense in X, then f is a homeomor-
phism.
Solution We first show that f is onto. Clearly,
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
because P is dense. This shows that f is onto.

To show that f is a homeomorphism it remains to verify that f is open, that is,
that the image f pUq of any open set U Ď X is also open. Since f is onto and one-to-
one, we have
f pXzUq “ Xz f pUq
and since f is continuous, the image Xz f pUq of the compact set XzU is also com-
pact. Therefore, f pUq is an open set.

all points of X are periodic points of f , then f is a homeomorphism.
Solution By Problem 1.15, it suffices to show that f is one-to-one. Take x, y P X

with x ‰ y. If f pxq “ f pyq, then
f j pxq “ f j pyq for all j P N. (II.1.2)
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.

the set P of periodic points of f is dense in X and there exists p P N such that all
periodic points have period at most p, then f is a homeomorphism.
Solution Again by Problem 1.15, it suffices to show that f is one-to-one. Take

x, y P X with x ‰ y. If f pxq “ f pyq, then property (II.1.2) holds. Moreover,
x “ lim xn and y “ lim yn

nÑ8 nÑ8
for some sequences xn , yn P P. Since f is continuous, we conclude that

` ˘
0 “ f j pxq ´ f j pyq “ lim f j pxn q ´ f j pyn q
nÑ8
for all j P N. Hence, given ε ą 0, there exists k P N such that

| f j pxn q ´ f j pyn q| ď ε for n ě k (II.1.3)
and j “ 1, . . . , p2 . Denoting by n j and m j , respectively, the periods of x j and y j , we

have n j m j ď p2 and it follows from (II.1.3) that
|x j ´ y j | “ | f n j m j px j q ´ f n j m j py j q| ď ε
for j ě k. This implies that
x ´ y “ lim pxn ´ yn q “ 0,
nÑ8
which shows that f is one-to-one.
Problem 1.18 Consider an interval I “ pa, bq Ĺ S1 with 0 ă a ă b ă 1. Show that

there exists n P N such that E2n pIq “ S1 (see Definition 1.5), identifying I with the
set trxs : x P Iu Ĺ S1 (see Figures I.1.3 and I.1.4).
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 .
Problem 1.19 Show that given x P S1 and δ ą 0, there exist y P px ´ δ , x ` δ q Ď S1

and n P N such that
n
E pxq ´ E n pyq ě 1 .
2 2
4
Solution Take x P S1 and δ ą 0. By Problem 1.18, there exists n P N such that
` ˘
E2n px ´ δ , x ` δ q “ 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 ń 1
n“1 E2 x is dense in S .
Solution Note that the set

8
ď
A“ E2ń x
n“1
is dense in S1 if and only if each open interval I Ĺ S1 contains a point y P A. Now

consider an interval I “ pa, bq Ĺ S1 with 0 ă a ă b ă 1. By Problem 1.18, there
exists n P N such that E2n pIq “ S1 . In particular, there exists y P I such that E2n pyq “ x
and so y P A. This shows that A is dense in S1 .
Problem 1.21 Consider the map f : R Ñ R defined by f pzq “ z3 on the set
R “ tz P C : |z| “ 1u.
Show that the set of periodic points of f is dense in R.

q
Solution Note that f q pzq “ z3 for each q P N. Indeed, proceeding inductively we
obtain ` q´1 ˘3 q
f q pzq “ f p f q´1 pzqq “ z3 “ z3 .
Hence, z P R is a q-periodic point of f if and only if
q q ´1
f q pzq “ z ðñ z3 “ z ðñ z3 “ 1.
Therefore, a point z “ e2π iθ , with θ P p0, 1s, is q-periodic if and only if

p
θ“ for p “ 1, . . . , 3q ´ 1.
3q ´ 1
In particular, given z P R and ε ą 0, by taking q sufficiently large we find that there
exist periodic points ε -close to z. This implies that the set of periodic points of f is
dense in R.
Problem 1.22 Consider the map f : T2 Ñ T2 defined by
f px, yq “ pEm pxq, Em pyqq.
Show that the set of periodic points of f is dense in T2 .

Solution We have
f q px, yq “ pEmq pxq, Emq pyqq “ pEmq pxq, Emq pyqq.
Hence, a point px, yq P T2 is q-periodic if and only if
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.

matrix A whose spectrum contains no root of 1. Show that the set of periodic points
of TA is Qn {Zn .
Solution Let rxs P Tn be a periodic point of TA . Then there exist q P N and y P Zn

such that Aq x “ x ` y, that is,
pAq ´ Idqx “ y.
Since the spectrum of A contains no root of 1, the matrix Aq ´ Id is invertible and so
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 .
Assuming, without loss of generality, that q1 ą q2 , we obtain
Aq1 ´q2 pAq2 xq ´ 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 .
Problem 1.24 Let TA : Tn Ñ Tn be an endomorphism of the torus. Show that
card TA´1 x “ |det A| for all x P Tn ,
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).
Solution First note that TA “ TP TD TQ . Since
|det P| “ |det Q| “ 1,
the maps TP and TQ are automorphisms of the torus and so
card TA´1 x “ card TD´1 pTP´1 xq
for all x P Tn . Hence, it suffices to show that
card TD´1 x “ |det D| for all x P Tn .
Let k1 , . . . , kn be the (integer) entries on the diagonal of D. The identity TD rys “ x,

with x “ rxs and x P r0, 1qn , is equivalent to
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,
card TD´1 x “ |k1 | ¨ ¨ ¨ |kn | “ |det D|.

matrix A whose spectrum contains no root of 1. Show that the number of q-periodic
points of TA is equal to |detpAq ´ Idq|.
Solution A point rxs P Tn is q-periodic if and only if
rAq xs “ rxs ðñ rpAq ´ Idqxs “ 0.
Since the matrix Aq ´ Id is invertible (because by hypothesis the spectrum of A

contains no root of 1), it induces the endomorphism of the torus TAq Íd . Hence, the
number of solutions of the equation
TAq Íd rxs “ rpAq ´ Idqxs “ 0
is equal to card TA´1

q Íd 0, which by Problem 1.24 is |detpA ´ Idq|.
q
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.
Solution First note that the eigenvalues

? ?
τ “ p3 ` 5q{2 and τ ´1 “ p3 ´ 5q{2
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
“ |pτ q ´ 1qpτ ´q ´ 1q|

“ τ q ` τ ´q ´ 2
“ 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
card f ´1 x “ q for all x P S1 .
(The number q coincides with the degree deg f of f .)
Solution Since f 1 never vanishes, it is always positive or always negative. Hence,

the map f is either increasing or decreasing. Therefore, as x P S1 increases its image
f pxq travels along S1 always in the positive direction or always in the negative direc-
tion. When x returns to the initial point, the image f pxq also returns to the initial
image and since it travels always in the positive direction or always in the negative
direction, all points in S1 are attained the same number of times (more precisely, the
number of times that f pxq travels along S1 either in the positive or in the negative
direction). Finally, note that since | f 1 | is continuous, it has a maximum and so the
number of times that f pxq travels along S1 is finite.
Problem 1.28 Given a map f : X Ñ X, show that the complement of a backward

f -invariant set is forward f -invariant.
Solution Let A be a backward f -invariant set. Given x P f pXzAq, there exists

y P XzA with f pyq “ x. Since A is backward f -invariant, this implies that x R A
(otherwise f pyq “ x P A and so y P A). Hence,
f pXzAq Ď XzA.
Problem 1.29 Let f : X Ñ X be a continuous map on a compact metric space. More-

over, let A Ď X be a closed backward f -invariant set and define
8
č
Λ“ f ń A.
n“0
Show that:
1. if U is an open neighborhood of Λ , then f ń A Ď U for any sufficiently large n;
2. the set Λ is f -invariant.
f 1A
Λ 2A
f
Fig. II.1.4 Preimages of the set A and their intersection.
Solution 1. Since A is backward f -invariant, the sets Bn “ f ń A satisfy Bn`1 Ď Bn

for all n P N (see Figure II.1.4). Moreover,
8
č
Λ“ Bn .
n“0
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 ń A “ Bn Ď U for all n ě N.
2. We have
8
č
f ´1Λ “ f ń A.
n“1
But since the sequence Bn “ f ń A is decreasing, we conclude that

8
č 8
č
f ń A “ f ń A “ Λ .
n“1 n“0
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).
Solution The expanding map E3 : S1 Ñ S1 is given by

$
’
&3x if x P r0, 1{3q,
E3 pxq “ 3x ´ 1 if x P r1{3, 2{3q,
’
%
3x ´ 2 if x P r2{3, 1q.
Note that if A contains some point different from 0, then
E3´1 A X pS1 zJq ‰ ∅.
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.

#
x1 “ 6y5 ,
y1 “ ´4x3
on R2 . Show that for each set I Ď R` 0 the union

ď (
px, yq P R2 : y6 ` x4 “ a
aPI
is invariant under the flow determined by the differential equation.
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
py6 ` x4 q1 “ 6y5 y1 ` 4x3 x1

“ 6y5 p´4x3 q ` 4x3 p6y5 q “ 0.
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.

#
x1 “ ax ´ xy,
y1 “ ý ` x2 ´ 2y2 ,
on R2 for some a ą 0. Show that the solutions starting in the sets

" *
( x2
S1 “ px, yq P R2 : x “ 0 and S2 “ px, yq P R2 : y “
1 ` 2a
remain in these sets for all time.
Fig. II.1.5 The sets Ca “ tpu, vq P R2 : y6 ` x4 “ au.
Solution In the coordinates px, yq the set S1 is given by x “ 0. Moreover, x1 |x“0 “ 0

on this set, which shows that any solution starting in S1 remains in S1 for all time.
Now consider the set S2 . We introduce new coordinates px, zq with
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 ¯
“ ý ` 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.
Problem 1.33 Find the flow determined by the equation x2 ` x “ 0.
Solution One can easily verify that the solutions of the equation are
xptq “ c1 cost ` c2 sint for c1 , c2 ,t P R.
Letting y “ x1 , one can rewrite x2 ` x “ 0 in the form

Fig. II.1.6 Phase portrait of equation (II.1.5).
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 “ ć1 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.
Problem 1.34 Find whether the equation x1 “ xpx2 ` 1q determines a flow.
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
whenever xp0q ą 0 and t ą 0. Finally, letting xptq Ñ `8 gives

1
tď ,
2xp0q2
which shows that the equation does not determine a flow because some solutions
are not defined on the whole R.
Fig. II.1.7 Phase portrait of the equation x1 “ xpx2 ` 1q.
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
ϕt px0 , y0 q “ px0 cost ` y0 sint, ´x0 sint ` y0 costq
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
ϕt px0 , 0q R R` ˆ t0u for t P p0, 2π q,
and so the corresponding Poincaré map is the identity map .

#
x1 “ y,
y1 “ ´ sin x.
Show that the set (

X “ px, yq P R2 : y “ 0, x P p0, π q
is a Poincaré section and determine the corresponding first return time and Poincaré
map.
Solution The critical points of the equation are px, yq “ pnπ , 0q, for n P Z. Note that
in the neighborhoods of the points
px, yq “ pp2n ` 1qπ , 0q and px, yq “ p2nπ , 0q,
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.
Fig. II.1.8 Phase portrait of the equation in Problem 1.36.
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
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
hpϕτ px,tq pxqq “ ψt phpxqq for x,t ą 0.

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.
Problem 1.38 Let f , g : R2 Ñ R be C1 functions with f ą 0 such that
f px ` k, y ` lq “ f px, yq and gpx ` k, y ` lq “ gpx, yq
for all x, y P R and k, l P Z. Then the differential equation
x1 “ f px, yq, y1 “ gpx, yq
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
ϕt p0, zq “ pxptq, yptqq “ pxpt, zq, ypt, zqq
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
(see Figure I.1.8) is invertible.
Solution Let pxpt, wq, ypt, wqq be the solution of the equation
px1 , y1 q “ ´p f , gqpx, yq with px, yqp0q “ p1, wq.

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,
that is, h is the inverse of h.
Problem 1.39 Consider the differential equation px1 , y1 q “ pα , β q on T2 “ R2 {Z2

with α , β ‰ 0. Determine whether any orbit of the flow determined by this equation
is dense.
Solution The solutions of the equation px1 , y1 q “ pα , β q on R2 are
pxptq, yptqq “ pxp0q, yp0qq ` pα t, β tq.
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.
On the other hand, if α {β P Q, then writing α {β “ p{q with p, q P Z we have
pα t, β tq “ pp, qq P Z2 for t “ q{β .
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
pα t, β tq “ pn, α tq, with α “ α {β P RzQ.
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 ;
0 1
Fig. II.1.9 A solution of the equation px1 , y1 q “ pα , β q on T2 .
2. has at least one point in each of the sets

( (
Bϕ “ v P Aϕ : pL2 ϕ qpvq ď 0 and Cϕ “ v P Aϕ : pL2 ϕ qpvq ě 0 ;
3. intersects Aϕ transversally at each point in the sets

( (
v P Aϕ : pL2 ϕ qpvq ă 0 and v P Aϕ : pL2 ϕ qpvq ą 0 .
Solution 1. Note that if vptq is a periodic solution of the equation, then there exist
t “ t1 and t “ t2 such that
ϕ pvpt1 qq ď ϕ pvptqq ď ϕ pvpt2 qq
(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
pLϕ qpvptqq “ x f pvptqq, ∇ϕ pvptqqy

“ xv1 ptq, ∇ϕ pvptqqy (II.1.9)
d
“ ϕ pvptqq “ 0.
dt
This shows that each periodic orbit has at least two points in Aϕ .
2. Assume that ϕ ˝ v has a local maximum at t “ t1 and a local minimum at 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
pL2 ϕ qpvptqq “ x f pvptqq, ∇ϕ pvptqqy.
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
Problem 2.1 Show that if f : X Ñ X is a continuous map on a compact metric space,

then f pω pxqq “ ω pxq for all x P X.
Solution Take y P ω pxq. By Proposition 2.4 there exists a sequence nk Õ 8 in N

satisfying
y “ lim f nk pxq.
kÑ8
Since f is continuous, we have
f pyq “ lim f nk `1 pxq

kÑ8
and, again by Proposition 2.4, f pyq P ω pxq. In other words,
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
Problem 2.2 Given a map f : X Ñ X on a metric space without isolated points,

show that if ω pxq ‰ X, then the positive semiorbit γ ` pxq is not dense.
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.
96 II.2 Topological Dynamics
B(y, rk + 1 )
y B(y, rk )
f nk (x)
Fig. II.2.1 The points f nk pxq P Bpy, rk qzBpy, rk`1 q.
Problem 2.3 Let f : X Ñ X be a continuous map. Given x P X, show that if ω pxq

contains infinitely many points, then it has no isolated points.
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
lim f mi pxq “ a and lim f ni pxq “ b,

iÑ8 iÑ8
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,
lim zi “ a and lim f ki pzi q “ b.

iÑ8 iÑ8
Ť
Since zi P U for all i ě N, we conclude that b P iPN f ki pUq.

x P X, show that if A “ ω pxq, then for any nonempty closed subset B Ĺ A we have
B X f pAzBq ‰ ∅.
Solution Assume that B X f pAzBq “ ∅ for some nonempty closed subset B Ĺ A.

Then there exist open sets U and V such that
U XV “ ∅, BĹU and f pAzBq Ĺ V. (II.2.1)

II.2 Topological Dynamics 97
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
and it follows from (II.2.1) that f pW q XU “ ∅. Since
A “ ω pxq “ pAzBq Y B Ĺ W YU,
by Proposition 2.4 there exists an integer k0 ą 0 such that f n pxq P W YU for n ě k0 .

Moreover, f n pxq P W and f n`1 pxq P U for infinitely many integers n ě k0 because
both AzB and B are nonempty. Thus, there exists an increasing sequence pni qiPN
such that
f ni pxq P W, f ni `1 pxq P U and y “ lim f ni pxq P W .

iÑ8
Therefore, ´ ¯
f pyq “ f lim f ni pxq “ lim f ni `1 pxq P U
iÑ8 iÑ8
and so f pyq P f pW q X U, which yields a contradiction since f pW q X U “ ∅. This

shows that B X f pAzBq ‰ ∅.
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.
Solution Note that if x “ 0, then x1 “ 0 and so
ω p0q “ α p0q “ t0u.
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).
Fig. II.2.2 Phase portrait of the equation x1 “ ´xp1 ` cos2 xq.

#
r1 “ rpr ´ 1q,
θ 1 “ 1,
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.
Solution Consider the sets

(
Cr “ px, yq P R2 : x2 ` y2 “ r2
for r ě 0. Any orbit starting in C0 or C1 remains, respectively, in C0 or C1 . On the

other hand, r1 ă 0 for r P p0, 1q and r1 ą 0 for r ą 1 (see Figure II.2.3). Hence, for
each p P Cr we have
α ppq “ ω ppq “ tp0, 0qu for r “ 0,

α ppq “ C1 , ω ppq “ tp0, 0qu for r P p0, 1q,
α ppq “ ω ppq “ C1 for r “ 1,
α ppq “ C1 , ω ppq “ ∅ for r ą 1.
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.
Solution The only critical point is the origin and is a repelling

? focus since the eigen-
values of the Jacobian matrix at the origin are p1 ˘ 3iq{2. Moreover, for |x| ă 1
the function
y2 1
Hpx, yq “ ´ logp1 ´ x2 q
2 2
increases along solutions because
` ˘
9 yq “ ∇Hpx, yq ¨ px2 ´ 1qpy ´ xq, x “ x2 .
Hpx,
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
Problem 2.9 Let f : X Ñ X be a continuous map on a topological space and let

NW p f q be the set of nonwandering points. Show that:
1. NW p f q is closed;
2. NW p f q is forward f -invariant.
Solution 1. We show that the complement of NW p f q is open. If x R NW p f q, then

there exists an open neighborhood U of x such that f n pUq X U “ ∅ for all n P N.
Note that any point y P U is also outside of NW p f q. Hence, the complement of
NW p f q is open.
2. Take x P NW p f q and let V be an open neighborhood of f pxq. Then U “ f ´1V

is an open neighborhood of x and there exists n P N such that f n pUq XU ‰ ∅. But
then
f p f n pUq XUq Ď f n pV q XV,
which implies that f n pV q XV ‰ ∅. Hence, f pxq P NW p f q and so NW p f q is forward
f -invariant.
Problem 2.10 Let f : X Ñ X be a continuous map on a topological space. Show

that ω pxq Ď NW p f q for all x P X.
Solution Let x P X and y P ω pxq. Moreover, let U be an open neighborhood of y.

Since y P ω pxq, by Proposition 2.4 there exist positive integers nk Õ `8 such that
f nk pxq Ñ y. In particular, there exist nk0 ă nk1 such that
f nk0 pxq P U and f nk1 pxq P U.
Now let z “ f nk0 pxq and take n “ nk1 ´ nk0 . Then
z P f n pUq XU ‰ ∅,
which shows that y P NW p f q.
Problem 2.11 Let f : X Ñ X be a continuous map on a compact metric space. Show

that
lim dp f m pxq, NW p f qq “ 0
mÑ8
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.
Solution If there exist positive integers nk Õ 8 such that f nk pxq Ñ x when k Ñ 8,

then
f nk p f pxqq “ f p f nk pxqq Ñ f pxq P ω pxq
when k Ñ 8 because f is continuous. This shows that Rp f q is forward f -invariant.
Problem 2.13 Given a map f : X Ñ X, show that Rp f q Ď NW p f q.
Solution Take x P Rp f q. Then there exists a sequence xn P ω pxn q such that xn Ñ x

when n Ñ 8. Now let U be an open neighborhood of x. Then there exists m P N
with xm P U. Since xm P ω pxm q, it also exists n P N such that f n pxm q P U. Hence,
f n pUq XU ‰ ∅ and so x P NW p f q, which establishes the desired property.
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 ‰ ∅.
Note that A is closed and forward f -invariant since
f pAq Ď f pM1 q X f pM2 q Ď M1 X M2 “ A.
But then A is a nonempty closed forward f -invariant proper subset both of M1

and M2 , which contradicts the fact that M1 and M2 are minimal sets.
Problem 2.15 Give an example of an ω -limit set that is a minimal set.
Solution Consider a periodic orbit of a homeomorphism. Note that it is the ω -limit

set of any point in the orbit and that it is a minimal set since it contains no proper
nonempty closed invariant 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.
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
(3 ñ 1). Finally, assume that M “ ω pxq for every x P M. Let N be a nonempty

closed forward f -invariant subset of M and take y P N. Then M “ ω pyq Ď N. Hence,
N “ M and so M is a minimal set.
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 ‰ ∅.
Solution First assume that f is topologically transitive. Let U and V be nonempty
open sets. Then there exists n P N such that f ńV XU ‰ ∅. Hence,
∅ ‰ f n p f ńV XUq
Ď f n p f ńV q X f n pUq
Ď V X f n pUq
and so V X f n pUq ‰ ∅.
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 ńV , which shows that U X f ńV ‰ ∅ 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
ń
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 ńV ‰ ∅. 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 ń pUq.
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.
Solution We proceed by contradiction. Since f is topologically transitive, given

nonempty open sets U,V Ĺ X, there exists an integer n P N such that f ńU XV ‰ ∅.
Now let C Ĺ X be a nonempty closed f -invariant subset that has nonempty interior.
Then there exist a nonempty open set W Ĺ C and n P N such that
f ń 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 ń pXzCq Ď XzC,
which contradicts to (II.2.2) (because W Ĺ C). Thus, C has empty interior.
Problem 2.20 Given a continuous map f : X Ñ X on a complete metric space with

a countable basis and without isolated points, show that the following properties are
equivalent:
1. f is topologically transitive;
2. the set of points with a dense positive semiorbit is dense.
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 íUn .
iPN
Since f is topologically transitive, given a nonempty open set U, there exists i P N

such that f íUn X U ‰ ∅ and so Vn X U ‰ ∅. Thus, each open set Vn is dense and
Ş
their intersection V “ nPN Vn is also dense (because complete metric spaces have
the property that any countable intersection of dense open sets is dense).
We show that V is in fact the set of points with a dense positive semiorbit. Indeed,
x P V if and only if for each n P N there exists i P N such that f i pxq P Un , that is, if and
only if for each n P N we have γ ` pxq XUn ‰ ∅, which is equivalent to γ ` pxq “ X.
(2 ñ 1). Take a point x such that γ ` pxq “ X and let U,V Ĺ X be nonempty open
sets. Since X has no isolated points, γ ` pxq visits infinitely often U and V . Hence,
there exist m, n P N with m ą n such that
f m pxq P U and f n pxq P V.
Therefore,
x P f ´mU X f ńV “ f ń p f ´pmńqU XV q
and so the set f ´pmńqU XV is nonempty. Thus, f is topologically transitive.
Problem 2.21 Give an example of a continuous map f : X Ñ X on a finite set with

at least one dense positive semiorbit, but which is not 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.
Problem 2.22 Give an example of a continuous map f : X Ñ X on an infinite set

with at least one dense positive semiorbit, but which is not topologically transitive.
Solution Consider the set
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 ńU 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.
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 ‰ ∅.
Hence, f is topologically transitive.
Fig. II.2.5 Graph of the map f 2 .
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
is also in V . Moreover, f n ppq “ e2π iϕ and so p P f ńU. Therefore, f ńU XV ‰ ∅.

This shows that the map f is topologically transitive.
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 .
Solution If Rm m2
α pxq “ Rα pxq for some integers m1 ą m2 , then
1
m1 α ´ m2 α “ m
for some m P Z. But then α “ m{pm1 ´ m2 q, which is impossible since α P RzQ.

Thus, the points Rmα pxq are pairwise distinct.
Now assume that there exists a closest point Rm α pxq to x with m ‰ 0. Since the
iterates xn “ Rnm
α pxq are equally spaced on S 1 and since x ‰ x for all n ‰ 0, there
n
exists n P Z such that xn is between Rm α pxq and x. But this is impossible since then
xn would be closer to x than Rm α pxq. Hence, there exists a sequence mk P Z with
m
|mk | Ñ 8 such that Rα k pxq ‰ x for all k and
m
Rα k pxq Ñ x when k Ñ 8. (II.2.3)
nm
Now consider the iterates yn “ Rα k pxq for a given k. Again since these are equally
spaced on S1 , it follows from (II.2.3) that there are points of the orbit of x arbitrarily
close to each given point y P S1 . In other words, the orbit of x is dense in 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.
Solution Let H : X Ñ Y be a topological conjugacy between the maps f and g.

Then
H ˝ f “ g ˝ H on X
and proceeding inductively we obtain
H ˝ f q “ gq ˝ H on X
for all q P N. Therefore, if x P X is a q-periodic point of f , that is, f q pxq “ x, then
Hpxq “ Hp f q pxqq “ gq pHpxqq,
that is, Hpxq P Y is a q-periodic point of g. Similarly, if y P Y is a q-periodic point

of g, then
H ´1 pyq “ H ´1 pgq pyqq “ f q pH ´1 pyqq,
that is, H ´1 pyq P X is a q-periodic point of f . Since the map H is bijective, this
yields the desired result.
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.
Solution Clearly, p is a fixed point of f if and only if q “ hppq is a fixed point of g.

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.
Solution Let h : Y Ñ X be a homeomorphism such that
f ˝h “ h˝g
for some maps f : X Ñ X and g : Y Ñ Y . Now assume that f is topologically mixing.

Given nonempty open sets U,V Ĺ Y , the images hpUq and hpV q are also open and
so there exists n P N such that
f ´m hpUq X hpV q ‰ ∅ for m ě n.
It follows from the identity

f ´m ˝ h “ h ˝ g´m
that
hpg´mU XV q “ hpg´mUq X hpV q
“ f ´m hpUq X hpV q ‰ ∅
and so
g´mU XV ‰ ∅ for m ě n.
This shows that g is 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
p f ˆ f qpx, yq “ p f pxq, f pyqq
is also topologically mixing.
Solution Let U1 ,U2 ,V1 ,V2 Ď X be nonempty open sets and define U “ U1 Û2 and
V “ V1 ˆV2 . Since f is topologically mixing, there exists m P N such that
f ńU1 XV1 ‰ ∅ and f ńU2 XV2 ‰ ∅
for all n ě m. Then

p f ˆ f qń pU1 Û2 q X pV1 ˆV2 q “ p f ńU1 XV1 q ˆ p f ´1U2 XV2 q ‰ ∅
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.
Problem 2.30 Let f : X Ñ X be a continuous map on a compact metric space pX, dq

such that
dp f pxq, f pyqq ď dpx, yq for all x, y P X.
Show that the topological entropy of f is zero.
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
Problem 2.31 Show that if f : X Ñ X is a homeomorphism on a compact metric

space pX, dq, then
N f pn, ε q “ N f ´1 pn, ε q
for all n P N and ε ą 0.
Solution Note that

(
dn, f px, yq “ max dp f k pxq, f k pyqq : k “ 0, . . . , n ´ 1
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
and so N f pn, ε q “ N f ´1 pn, ε q.

a closed forward f -invariant set Y Ď X, show that hp f q ě hp f |Y q.
Solution Note that N f pn, ε q ě N f |Y pn, ε q. Therefore,
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.

spaces, respectively, pX, dX q and pY, dY q satisfying f ˝ H “ H ˝ g for some continu-
ous onto map H : Y Ñ X. Show that hp f q ď hpgq.
Solution It follows from the identity f ˝ H “ H ˝ g that
f n ˝ H “ H ˝ gn
for all n P N. Since H is uniformly continuous, given ε ą 0, there exists δ ą 0 such

that
dX pHpyq, Hpy1 qq ă ε whenever dY py, y1 q ă δ .
Therefore, if
dX p f m px1 q, f m px1 qq ě ε
for m “ 0, . . . , n ´ 1, then
dY pgm pH ´1 pxqq, gm pH ´1 px1 qqq ě δ
for m “ 0, . . . , n ´ 1. This implies that Ng pn, δ q ě N f pn, ε q and so

1 1
lim sup log Ng pn, δ q ě lim sup log N f pn, ε q
nÑ8 n nÑ8 n
for each ε ą 0. Letting ε Ñ 0, one can take δ Ñ 0 and thus hpgq ě hp f q.

spaces satisfying f ˝ H “ H ˝ g for some homeomorphism H : Y Ñ X. Show that
hp f q “ hpgq.
Solution It follows from Problem 2.33 that
hp f q ď hpgq and hpgq ď hp f q.
Therefore, the maps f and g have the same topological entropy.
Problem 2.35 Compute the topological entropy of the map f in Problem 1.21.
Solution Given n, k P N, consider the points
x j “ e2π ip j{3
n`k q
for j “ 0, . . . , 3n`k ´ 1.
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,
dn pxi , x j q ě 3´pk`1q for i ‰ j
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
Npn, 3´pk`1q q ď 3n`k . (II.2.5)
Combining (II.2.4) and (II.2.5) we obtain
Npn, 3´pk`1q q “ 3n`k for n, k P N,
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.
Problem 2.36 Determine whether there exists a continuous map f : X Ñ X on a

compact metric space with infinite topological entropy.
Solution Consider ś the continuous map f “ E2 ˆ E3 ˆ ¨ ¨ ¨ on the compact topolog-

ical space X “ mě2 S1 with the product topology. We note that X is metrizable.
For example, let n be the smallest integer such that xn ‰ yn , writing x “ px2 , x3 , . . .q
and y “ py2 , y3 , . . .q, and consider the distance
#
0 if x “ y,
dpx, yq “ ń
2 if x ‰ y.
For each m ě 2 the set

(
Xm “ x P X : x p “ 0 for p ‰ m
is compact and forward f -invariant. Moreover, for the homeomorphism H : Xm Ñ S1

defined by Hpxq “ xm we have
H ˝ f “ Em ˝ H on Xm
and so H is a topological conjugacy between f |Xm and Em . Hence, it follows from

Theorem 2.12 (or Problem 2.33) that
hp f q ě hp f |Xm q “ hpEm q “ log m.
Since m is arbitrary, we conclude that hp f q “ 8.
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
for any continuous map f : X Ñ X.
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
dn1 px, yq ě d 1 p f i pxq, f i pyqq ě δ pε q.
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.
Interchanging d and d 1 we also obtain hd p f q ě hd 1 pgq and so hd p f q “ hd 1 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
# +
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;
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
Mpn, δ {2q ě NpUn q;
3. if U is a finite open cover of X with diam U :“ suptdiamU : U P Uu ă ε , where
diamU “ tdpx, yq : x, y P Uu,
then Npn, ε q ď NpUn q;

4.
1
hp f q “ lim lim log NpUn q.
diam UÑ0 nÑ8 n
Solution 1. Note that the elements of Un`m are the open sets
˜ ¸
č
n`m´1 č
n´1 č
m´1
´k ´k ń ´l
f Uk “ f Uk X f f Ul`n
k“0 k“0 l“0
(
P U X f ńV : U P Un ,V P Um ,
with U0 , . . . ,Un`m´1 P U. Therefore,
NpUn`m q ď NpUn qNpUm q. (II.2.6)
In other words, the sequence cn “ log NpUn q is subadditive, that is,
cn`m ď cn ` cm for all n, m P N. (II.2.7)
We show that for any such sequence the limit

cn cn
lim “ inf (II.2.8)
nÑ8 n nPN n
exists. Given n, k P N, write n “ qk ` r with q P N Y t0u and r P t0, . . . , q ´ 1u. Then

cn cqk ` cr qck ` cr
ď ď
n qk ` r qk ` r
and so
cn ck
lim sup ď
nÑ8 n k
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
On the other hand, by the definition of dn we have
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
Bp f k ppi q, δ {2q Ď Uki
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
and so Mpn, δ {2q ě NpUn q.

3. Given n P N and ε ą 0, let An Ĺ X be a set with cardinality Npn, ε q such that
dn ppi , p j q ě ε for all pi , p j P An with i ‰ j.
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.
4. Now let U be an open cover of X with Lebesgue number δ . By items 1 and 2 we

have
1 1
lim log NpUn q ď lim sup log Mpn, δ {2q.
nÑ8 n nÑ8 n
Note that δ Ñ 0 when diam U Ñ 0. Hence,

1 1
lim sup lim log NpUn q ď lim sup lim sup log Mpn, δ {2q
diam UÑ0 nÑ8 n diam UÑ0 nÑ8 n
1 (II.2.9)
“ lim lim sup log Mpn, δ {2q
δ Ñ0 nÑ8 n
“ hp f q.
Now let U be an open cover of X with diam U ă ε . By items 1 and 3 we have

1 1
lim sup log Npn, ε q ď lim log NpUn q
nÑ8 n nÑ8 n
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.
Problem 2.39 Let f : r0, 1s Ñ r0, 1s be a homeomorphism. Show that hp f q “ 0.
Solution Let U be an open cover of r0, 1s by open intervals (more precisely, by

sets of the form pa, bq X r0, 1s with a, b P R). Since f is strictly increasing or strictly
decreasing, the cover f ´1 U is also composed of open intervals, with
card f ´1 U “ card U.
Hence, card U2 ď 2 card U (see Problem 2.38). One can show by induction that
card Un ď n card U (II.2.11)
and so
1
log card Un “ 0.
lim (II.2.12)
n
nÑ8
By Problem 2.38, given ε ą 0 and an open cover U with diam U ă ε , we have

Npn, ε q ď NpUn q ď card Un
for each n P N. Hence, it follows from (II.2.12) that

1
lim sup log Npn, ε q “ 0
nÑ8 n
for each ε (taking a cover U by open intervals of length less than ε ) and so hp f q “ 0.
Problem 2.40 Let f : S1 Ñ S1 be a homeomorphism. Show that hp f q “ 0.
Solution As in Problem 2.39, let U be an open cover of S1 by open intervals. Since

f is strictly increasing or strictly decreasing, the cover f ´1 U is again composed of
open intervals, with
card f ´1 U “ card U.
Hence, inequality (II.2.11) holds, which yields property (II.2.12). One can now show
as in Problem 2.39 that hp f q “ 0.
Chapter II.3
Low-Dimensional Dynamics
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
Rα rxs “ rx ` α s and Rβ rxs “ rx ` β s
for each rxs P S1 . Hence,
pRα ˝ Rβ qrxs “ Rα rx ` β s “ rx ` α ` β s
for each rxs P S1 and so Rα ˝ Rβ “ Rα `β .
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,
pTA ˝ TB qrxs “ TA rBxs “ rABxs
for each rxs P Tn and so TA ˝ TB “ TC with C “ AB.
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.

116 II.3 Low-Dimensional Dynamics
Solution We have ´1¯

1
f p0q “ and f “ 1.
2 2
Hence, t0, 1{2u is a periodic orbit of f with period 2. On the other hand, the function
F : R Ñ R defined by
1 1
Fpxq “ x ` ` sinp2π xq
2 4π
is a lift of f and satisfies
n
F n p0q “ for n ě 0.
2
Indeed, Fp0q “ 1{2 and if F k p0q “ k{2, then
F k`1 p0q “ FpF k p0qq “ Fpk{2q

k 1 k`1
“ ` “ .
2 2 2
Taking x “ 0 in
F n pxq ´ x
ρ pFq “ lim ,
nÑ8 n
we obtain
F n p0q n`1 1
ρ pFq “ lim “ lim “ .
nÑ8 n nÑ8 2n 2
This implies 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
f prxsq “ rx ` a sinp2π xqs
(see Figure I.3.2). Show that each orbit starting in p0, 1{2q is monotonous.
Solution Define a map F : R Ñ R by
Fpxq “ x ` a sinp2π xq.
For 0 ă x ă 1{2 we have sinp2π xq ą 0 and so Fpxq ą x. Moreover,
F 1 pxq “ 1 ` 2π a cosp2π xq ą 1 ´ 2π a ą 0.
Therefore, Fpxq ă Fp1{2q “ 1{2 and so

1
x ă Fpxq ă .
2
Proceeding inductively we obtain
1
0 ă x ă Fpxq ă F 2 pxq ă ¨ ¨ ¨ ă F n pxq ă ,
2
which implies that the orbit of rxs is monotonous.
II.3 Low-Dimensional Dynamics 117
Problem 3.5 Let f : S1 Ñ S1 be a homeomorphism with a single periodic orbit.

Show that any other orbit is asymptotic to this periodic orbit (see Figure I.3.3 for an
example).
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
f pUk q “ Uk`1 for k “ 0, . . . , p ´ 1 and f pUp q “ U0 .
Hence, f p pUk q “ Uk for each k and since f is a homeomorphism, the map f p is

monotonous on each interval Uk . It follows from the continuity of f that
(
dp f k pxq, γ px0 qq “ min dp f k pxq, f l px0 qq : l “ 0, . . . , p ´ 1 Ñ 0
when k Ñ ˘8.
Problem 3.6 Let f : S1 Ñ S1 be an orientation-reversing homeomorphism and let F

be a lift of f . Show that Fpx`1q “ Fpxq´1 for all x P R (see Figures I.3.4 and I.3.5).
Solution Define a map g : S1 Ñ S1 by gpxq “ f p´xq. Note that g is an orientation-

preserving homeomorphism since it is the composition of two orientation-reversing
homeomorphisms. Moreover, the map G : R Ñ R given by Gpxq “ Fp´xq is a lift
of g and so
Gpx ` 1q “ Gpxq ` 1 for all x P R.
Therefore, for each x P X we have
Fpx ` 1q “ Gp´x ´ 1q “ Gp´xq ´ 1 “ Fpxq ´ 1.
Problem 3.7 Let f : S1 Ñ S1 be an orientation-reserving homeomorphism. Show

that ρ p f 2 q “ 0.
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
Gn pxq “ x ` nk for all n P N.
Therefore,
Gn pxq ´ x
ρ pGq “ lim “ k,
nÑ8 n
which implies that ρ p f 2 q “ tρ pGqu “ 0.

ρ p f q “ p{q for some integers p, q P N. Show that for each lift F of f there exists
x P R such that Fpxq ´ x ´ p{q P Z.
Solution We proceed by contradiction. Assume that there exists a lift F of f such

that Fpxq ´ x ´ p{q P RzZ for all x P R. Since F is continuous, there exists k P Z
such that
p
k ă Fpxq ´ x ´ ă k ` 1 for x P R.
q
On the other hand,
Fpx ` 1q ´ px ` 1q “ Fpxq ´ x for x P R
and since r0, 1s is compact, there exists ε ą 0 such that

p
k ` ε ď Fpxq ´ x ´ ď k`1´ε
q
for all x P R. Hence, it follows from the identity
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.
Problem 3.9 Let f , g : S1 Ñ S1 be homeomorphisms. Show that if F and G are lifts,

respectively, of f and g, then
FpGn pxqq ´ Gn pxq
lim “0
nÑ8 n
for each x P R.
Solution Note that

xn :“ Gn pxq ´ tGn pxqu P r0, 1s
for each n P N. Since F is a lift, we obtain
FpGn pxqq ´ Gn pxq “ Fpxn q ´ xn .
On the other hand, since r0, 1s is compact, there exists a constant M ą 0 (indepen-
dent of x and n) such that
|FpGn pxqq ´ Gn pxq| ď |Fpxn q| ` |xn | ď M ` 1,
which readily yields the desired property.
Problem 3.10 Let f , g : S1 Ñ S1 be homeomorphisms and let F and G be lifts, re-

spectively, of f and g. Show that if f is an orientation-preserving homeomorphism,
then:
1. for each k P N we have
F n pxq ´ x F n pxn q ´ xn
lim “ lim
nÑ8 n nÑ8 n
for any x, xn P R with |x ´ xn | ď k for n P N;
2.
F n pGn pxqq ´ Gn pxq F n pxq ´ x
lim “ lim
nÑ8 n nÑ8 n
for each x P R.
Solution 1. If |x ´ xn | ď k for some k P N, then
Fpxq ď Fpxn ` kq “ Fpxn q ` k
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
for each m P N. In particular, taking m “ n yields the inequality
|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
and so it follows from item 1 that

F n pGn pxqq ´ Gn pxq F n pxn q ´ xn
lim “ lim
nÑ8 n nÑ8 n
F n pxq ´ x
“ lim .
nÑ8 n
r{s ă ρ p f q ă p{q ă 1
for some positive integers r, s, p, q. Show that:

1. any lift F of f with Fp0q P p0, 1q satisfies
F q pxq ă x ` p and F s pxq ą x ` r
for all x P R;
2. if g : S1 Ñ S1 is an orientation-preserving homeomorphism with
dp f , gq :“ max dp f pxq, gpxqq

xPS1
sufficiently small, then r{s ă ρ pgq ă p{q.
Solution 1. Assume that F q pxq ě x ` p for some x P R. Since
F q px ` pq “ F q pxq ` p
and F is increasing, we obtain
F 2q pxq “ F q pF q pxqq ě F q px ` pq
“ F q pxq ` p ě x ` 2p
and it follows by induction that
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
F q pxq ă x ` p ´ δ for all x P R.

For dp f , gq sufficiently small we have

δ
Gq pxq ă x ` p ´
2
for all x P R and some lift G of g with Gp0q P p0, 1q. Therefore,
Gqn pxq ´ x
ρ pGq “ lim
nÑ8 nq
1 ÿ
n´1
“ lim pGq pGkq pxqq ´ Gkq pxqq
nÑ8 nq
k“0
1 ´ δ¯
ď lim n p´
nÑ8 nq 2
p δ p
“ ´ ă ,
q 2q q
which implies that ρ pgq ă p{q.
Analogously, it follows from the second inequality in item 1 that there exists
μ ą 0 such that
F s pxq ą x ` r ` μ for all x P R.
Provided that dp f , gq is sufficiently small, some lift G of g with Gp0q P p0, 1q satisfies
μ
Gs pxq ą x ` r `
2
for all x P R, which implies in a similar manner that ρ pgq ą r{s.

that if ρ p f q P RzQ, then for any m, n P Z with m ‰ n we have
8
ď
S1 “ f ´k I, for I “ r f n pxq, f m pxqs.
k“0
Ť8
Solution We proceed by contradiction. Assume that S1 ‰ k“0 f ´k I. Then
8
ď 8
ď
S1 ‰ f ´kpmńq I “ r f ´km`pk`1qn pxq, f ´pk´1qm`kn pxqs.
k“1 k“1
Since the intervals f ´kpmńq I, for k P N, have adjacent endpoints, the sequence
xk “ f ´kpmńq p f n pxqq
converges to some point z P S1 when k Ñ 8. But z is a fixed point of f mń (since

f mń pxk q “ xk´1 has the same limit), thus contradicting the fact that ρ p f q P RzQ
(see Theorem 3.6).

that if ρ p f q P RzQ, then ω pxq “ ω pyq for any x, y P S1 .
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
f kn pyq P r f mn´1 pxq, f mn pxqs.
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.

that if there exists an open interval J Ĺ S1 such that the sets f k pJq, for k P Z, are
pairwise disjoint, then f is not conjugate to an irrational rotation.
Solution We proceed by contradiction. Let h : S1 Ñ S1 be a homeomorphism such

that f ˝ h “ h ˝ Rα for some irrational number α . If D is a dense subset of S1 , then
hpDq is also dense. Thus, if f is conjugate to Rα , then γ ` pxq “ S1 for every x P S1
and so ω pxq “ S1 . On the other hand, if there exists an interval J as in the statement
of the problem, then ω pxq has no interior points and so it cannot be S1 .

ρ p f q P RzQ. Show that if there exists a minimal set K ‰ S1 (see Problem 2.14 for
the definition), then f is not conjugate to a rotation.
Solution If f is conjugate to a rotation, then proceeding as in Problem 3.14 one can

show that γ ` pxq “ S1 . Hence, by Problem 2.16, there exists no minimal set K ‰ S1 .
Problem 3.16 Show that the map h : S1 Ñ S1 given by
hpxq “ sin2 pπ x{2q
is a topological conjugacy between the maps f , g : S1 Ñ S1 defined by
f pxq “ 1 ´ |2x ´ 1| and gpxq “ 4xp1 ´ xq
Solution Clearly, h is continuous. Moreover, hp0q “ 0, hp1q “ 1 and since

π
h1 pxq “ sinpπ xq ą 0 for x P p0, 1q,
2
the map h is one-to-one and onto (hence invertible). Since its domain is a compact
set, the images of closed sets are closed sets and so the inverse of h is continuous.
Therefore, h is a homeomorphism. Moreover,
´π ¯
hp f pxqq “ sin2 p1 ´ |2x ´ 1|q
´2 ¯
2 π
“ cos |2x ´ 1|
´2 ¯
2 π
“ cos p2x ´ 1q
´2 π¯
“ cos π x ´
2
“ sin2 pπ xq
2
and
gphpxqq “ 4hpxqp1 ´ hpxqq
´ π ¯´ ´ π ¯¯
“ 4 sin2 x 1 ´ sin2 x
´ π2 ¯ ´π ¯2
“ 4 sin2 x cos2 x
2 2
´ ´ π ¯ ´ π ¯¯2
“ 2 sin x cos x “ sin2 pπ xq,
2 2
which shows that h ˝ f “ g ˝ h.
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
ρ pRα q “ tα u and ρ pRα q “ t2α u,
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
and so the maps are topologically conjugate.
Problem 3.19 Show that the map f in Problem 3.16 is topologically mixing.
Solution For each n P N the graph of f n is piecewise linear with

#
´ ¯ 0 if i “ 0, 2, . . . , 2n ´ 2,
n i
f “ (II.3.3)
2n 1 if i “ 1, 3, . . . , 2n ´ 1
(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
In view of (II.3.3), for each m ě n we have f m pIq “ S1 and so also f m pUq “ S1 .

In particular,
f m pUq XV ‰ ∅
for any nonempty open set V Ĺ S1 . Now take x P f m pUq XV . Then there exists y P U
such that f m pyq “ x and so
y P f ´mV XU ‰ ∅ for all m ě n.
Hence, f is topologically mixing.
Fig. II.3.1 Graph of the map f 2.
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.
Problem 3.21 Consider an interval I “ pa, bq Ĺ S1 with 0 ă a ă b ă 1. For the map g

in Problem 3.20, show that there exists n P N such that gn pIq “ S1 , identifying I with
trxs : x P Iu Ĺ S1 .
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
S1 “ f n phpIqq “ hpgn pIqq
and so gn pIq “ S1 (recall that h is bijective).
Problem 3.22 Let f : S1 Ñ S1 be an orientation-preserving homeomorphism and let

NW p f q be the nonwandering set for f . Show that if ρ p f q P RzQ, then NW p f q is a
minimal set.
Solution By Problem 2.9, NW p f q is closed and forward f -invariant. To show that

it is a minimal set, we must verify that it has no nonempty closed f -invariant proper
subsets. We proceed by contradiction. Assume that M Ĺ NW p f q is a nonempty
closed forward f -invariant subset. Then S1 zM is a nonempty open set. Let I “ pa, bq
be a connected component of S1 zM. We claim that
f n pIq X I “ ∅ for all n P N.
Otherwise it would exist n P N such that f n pIq X I ‰ ∅ and so f ń pIq X I ‰ ∅. But

since S1 zM is backward f -invariant, this implies that f ń pIq Ď I because otherwise
I could not be a connected component. Then f n pIq Ě I and so it follows from Prob-
lem 1.5 that f has periodic points, which contradicts the fact that ρ p f q is irrational.
Therefore, I Ď S1 zNW p f q and since S1 zM is the union of its connected components,
we obtain
S1 zM Ď S1 zNW p f q.
Hence, NW p f q Ď M and so NW p f q “ M.

that if ρ p f q P RzQ, then either NW p f q “ S1 or NW p f q is a Cantor set (that is,
a closed set with empty interior and without isolated points).
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
NW p f q “ ω pxq for all x P NW p f q,

each point y P NW p f q is the limit of some sequence f nk pxq ‰ y with nk Õ 8 when

k Ñ 8. In other words, y is not an isolated point. Moreover, if NW p f q ‰ S1 and
has nonempty interior, then for each connected component U Ď NW p f q there ex-
ists n P N such that f n pUq X U ‰ ∅. But since NW p f q is forward f -invariant, this
implies that f n pUq Ď U because otherwise U could not be a connected component.
Hence, it follows from Problem 1.5 that f has periodic points (notice that U is an
interval). But this is impossible since ρ p f q is irrational. Therefore, if NW p f q ‰ S1 ,
then NW p f q has empty interior.

irrational rotation number. Show that f has a unique minimal set K Ď S1 .
Solution Take x P S1 . By Problem 2.16, ω pxq is a minimal set. It is also f -invariant.

Indeed, by Proposition 2.4, y P ω pxq if and only if there exists a sequence nk Õ 8
in N such that
y “ lim f nk pxq.
kÑ8
Since f is a homeomorphism, this is equivalent to
f pyq “ lim f nk `1 pxq,

kÑ8
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.

irrational rotation number. Show that either ω pxq is nowhere dense or ω pxq “ S1 .
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
BM “ M or BM “ ∅,
which means that either M is nowhere dense or M “ S1 .
Problem 3.26 Let f : S1 Ñ S1 be a C2 map. Show that its derivative has bounded
variation.
Solution Take K ą 0 such that
| f 2 pxq| ď K for x P S1 .
Moreover, take points x0 ă x1 ă ¨ ¨ ¨ ă xn in S1 , with xn “ x0 , for some n P N. Then
| f 1 pxi q ´ f 1 pxi´1 q| “ | f 2 pzi q| ¨ |xi ´ xi´1 |
for some zi P pxi , xi´1 q and so

ÿ
n ÿ
n
| f 1 pxi q ´ f 1 pxi´1 q| “ | f 2 pzi q| ¨ |xi ´ xi´1 |
i“1 i“1
ÿn
ď Kpxi ´ xi´1 q ď K.
i“1
Therefore, Varp f 1 q ă `8.
Problem 3.27 Let f : S1 Ñ S1 be a C1 diffeomorphism. Moreover, given n P N and

an interval I Ď S1 , let
|p f n q1 pxq|
Dn pIq “ sup log n 1 .
x,yPI |p f q pyq|
Show that
ÿ
n´1
Dn pIq ď D1 p f i pIqq.
i“0
Solution Since
p f n q1 pxq “ f 1 p f n´1 pxqq f 1 p f n´2 pxqq ¨ ¨ ¨ f 1 pxq,
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|
For x, y P I we have f i pxq, f i pyq P f i pIq and so
| 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.
Problem 3.28 Let f : S1 Ñ S1 be an orientation-preserving C1 diffeomorphism with

irrational rotation number. Show that if there exist a sequence nk P N with nk Õ 8
and a constant C ą 0 such that
|p f nk q1 pxq| ¨ |p f ńk q1 pxq| ą C
for all x P S1 and k P N, then every orbit of f is dense.
Solution We proceed by contradiction. Assume that the orbit of some point x P S1

is not dense and let X “ γ pxq. Note that X is f -invariant. Indeed, y P γ pxq if and
only if there exists a sequence nk in Z with |nk | Ñ 8 such that f nk pxq Ñ y when
k Ñ 8, but since f is a homeomorphism, this is equivalent to f nk ˘1 pxq Ñ f ˘1 pyq
when k Ñ 8. Therefore, y P X if and only if f ˘1 pyq P X and so X is f -invariant.
Now consider a connected component I of the nonempty open set S1 zX. We claim
that the intervals In “ f n pIq, for n P Z, are disjoint. Indeed, if
f i pIq X f j pIq ‰ ∅ for some i ‰ j,
then f jí pIq X I ‰ ∅ and so f jí 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 | ` |Iń | “ p|p f n q1 pxq| ` |p f ń q1 pxq|q dx
I
ż a
ě2 |p f n q1 pxq| ¨ |p f ń 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
f p1q “ 2, f p2q “ 3 and f p3q “ 1
(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.

#
2x if x P r0, 1{2s,
f pxq “
2p1 ´ xq if x P r1{2, 1s
(see Figure I.3.6) has periodic points with all periods.
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.
Solution Consider the continuous map f : X Ñ X defined by f pxq “ ´x. Clearly,

f has periodic points with period two but it has no fixed points. Hence, it does not
satisfy Sharkovsky’s ordering.
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.
Solution We proceed by contradiction. Assume that X is disconnected. Then X “

A Y B with A and B nonempty closed sets with A X B “ ∅. Take points p P A and
q P B, and define a continuous map f : X Ñ X by
#
q if x P A,
f pxq “
p if x P B.
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 ,
y1 “ x ` y ´ x2
has no periodic solutions contained in the second quadrant.
Solution In the second quadrant we have x ă 0 and y ą 0. Hence,
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).

#
r1 “ 4 ´ 3r cos θ ´ r4 ,
θ 1 “ ´1,
written in polar coordinates, has at least one periodic solution.
Solution Note that

r1 “ 4 ´ 3r cos θ ´ r4 ą 0
for any sufficiently small r and that
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)
is contained in D and so in particular it is bounded. Moreover, D contains no critical

points (the origin is the only critical point). Thus, we have the qualitative behavior
in Figure II.3.2. It follows from the Poincaré–Bendixson theorem (Theorem 3.11)
that the ω -limit set of any point in D is a periodic orbit.
Fig. II.3.2 Behavior on the boundary of the set D.

#
r1 “ rp1 ´ rq,
θ 1 “ 1 ` cos2 θ
has at least one periodic solution.
Solution Since θ 1 ą 0, it follows from the first equation that the origin is the only
critical point. Moreover,
rp1 ´ rq ą 0 for any sufficiently small r ą 0
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.
Problem 3.36 Consider the function f : R2 Ñ R2 defined by

` ˘
f px, yq “ y, ´x ` yp1 ´ 3x2 ´ 6y2 q
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 .
Solution 1. Note that

V9 px, yq “ 2xy ` 2yp´x ` yp1 ´ 3x2 ´ 6y2 qq
“ 2y2 p1 ´ 3x2 ´ 6y2 q
ď 2y2 p1 ´ 3x2 ´ 3y2 q.
Hence, if V px, yq ě 1{3, then V9 px, yq ď 0.

2. Now observe that
V9 px, yq “ 2y2 p1 ´ 3x2 ´ 6y2 q
ě 2y2 p1 ´ 6x2 ´ 6y2 q.
Hence, if V px, yq ď 1{6, then V9 px, yq ě 0.
3. For a solution pxptq, yptqq with pxp0q, yp0qq “ p we have
V9 ppq “ ∇V ppq ¨ f ppq

ˇ
“ ∇V pxptq, yptqq ¨ f pxptq, yptqqˇt“0
ˇ
“ ∇V pxptq, yptqq ¨ px1 ptq, y1 ptqqˇ
t“0
d ˇ
“ V pxptq, yptqqˇt“0 .
dt
By item 1, if p2 “ 1{3, then V9 ppq ď 0 and so
pxptq, yptqq2 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.
Solution 1. We proceed by contradiction. Assume that v : R Ñ Rn is a periodic

solution with period T . Then V pvpT qq “ V pvp0qq, but by hypothesis we also have
V pvpT qq ă V pvp0qq.
This contradiction yields the desired result.

2. Assume once more that v : R Ñ Rn is a periodic solution with period T . Then
V pvpT qq “ V pvp0qq, but on the other hand
żT
d
V pvpT qq ´V pvp0qq “ V pvptqq dt
0 dt
żT
“ ∇V pvptqq ¨ v1 ptq dt
0
żT
“´ v1 ptq2 dt ă 0.
0
This contradiction implies that there are no periodic solutions.

#
x1 “ y,
y1 “ ´ f pxqy ´ gpxq
for some C1 functions f , g : R Ñ R with f ą 0. Show that the equation has no

periodic solutions.
Solution Note that

B B
divpy, ´ f pxqy ´ gpxqq “ y ` p´ f pxqy ´ gpxqq “ ´ f pxq
Bx By
is negative on R2 . Therefore, by Green’s theorem, if there exists a periodic solu-

tion px, yq : R Ñ R2 with period T and interior D (in the sense of Jordan’s curve
theorem), then ż
0ą divpy, ´ f pxqy ´ gpxqq dx dy
żD
“ p f pxqy ` gpxqq dx ` y dy
żBD
“ ý1 dx ` x1 dy
BD
żT
“ rý1 ptqx1 ptq ` x1 ptqy1 ptqs dt “ 0.
0
This contradiction shows that there are no periodic solutions.

#
x1 “ f px, yq,
y1 “ gpx, yq
on R2 . Show that if there exists a C1 function ϕ : R2 Ñ R such that

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.
Solution Without loss of generality, we assume that

Bpϕ f q Bpϕ gq
` ą0
Bx By
on some simply connected open set (the other case can be treated in a similar man-
ner). Let C be a periodic orbit of the equation with period T and let D be its interior
(in the sense of Jordan’s curve theorem). By Green’s theorem, we have
ż ˆ ˙ ż ż
Bpϕ f q Bpϕ gq
0ă ` dx dy “ p´ϕ g dx ` ϕ f dyq “ ϕ pý1 dx ` x1 dyq
D Bx By C C
żT
“ ϕ pxptq, ptqqrý1 ptqx1 ptq ` x1 ptqy1 ptqs dt “ 0.
0
This contradiction shows that there are no periodic orbits contained in U.

Problem 3.40 Show that the differential equation

#
x1 “ y,
y1 “ ´x ´ y ` x2 ` y2
has no periodic solutions.
Solution Consider the C1 function ϕ px, yq “ e´2x . Then

B ´2x B ` ´2x ˘
pe yq ` e p´x ´ y ` x2 ` y2 q “ ´2e´2x y ´ e´2x ` 2e´2x y
Bx By
“ é´2x ,
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 ń 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 ń v for x P Λ , v P E u pxq, n P N.
c
Solution We first consider the condition on the unstable space
dx f ń v ď cλ n v for x P Λ , v P E u pxq, n P N.
Take x P Λ and v P E u pxq. For all n P N we have y “ f n pxq P Λ and w “ dx f n v P

E u p f n pxqq. Therefore,
dy f ń w ď cλ n w
if and only if
d f n pxq f ń dx f n v ď cλ n dx f n v.
By the chain rule, the last inequality is equivalent to
1
dx f n v ě λ ń v.
c
One can show in an analogous manner that the condition on the stable space
dx f n v ď cλ n v for x P Λ , v P E s pxq, n P N

136 II.4 Hyperbolic Dynamics
is equivalent to
1
dx f ń 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
dx f E s pxq “ E s p f pxqq, dx f E u pxq “ E u p f pxqq, (II.4.1)
dx f n v ď d λ n v for v P E s pxq, n P N, (II.4.2)

and
dx f ń v ď d λ n v for v P E u pxq, n P N. (II.4.3)
To show that Λ is a hyperbolic set for f k with c “ 1, note first that by (II.4.1) we
have
dx f k E s pxq “ d f k´1 pxq f ¨ ¨ ¨ dx f E s pxq “ E s p f k pxqq
and
dx f k E u pxq “ d f k´1 pxq f ¨ ¨ ¨ dx f E u pxq “ E u p f k pxqq.
Moreover, it follows from (II.4.2) with n “ km that
dx f km v ď d λ km x for v P E s pxq, m P N,
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
dx p f ´k qm v ď dpλ k qm x for v P E u pxq, m P N.
Note that μ :“ d λ k ă 1 for all sufficiently large k. Since
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.
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.
Solution First note that Λ ˆ Λ is compact and g-invariant since
g´1 pΛ ˆ Λ q “ f ´1 pΛ q ˆ f pΛ q “ Λ ˆ Λ .
II.4 Hyperbolic Dynamics 137
Now we consider the splittings
Tx R p “ E u pxq ‘ E s pxq, for x P Λ ,
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
E u pzq “ E u pxq ˆ E s pyq and E s pzq “ E s pxq ˆ E u pyq,
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
w “ pu, vq P E s pxq ˆ E u pyq
we have
dx f n u ď cλ n u and dy f ń 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 ń
and so, considering the norm pu, vq “ u ` v, we obtain
dz gn w “ pdx f n u, dy f ń vq

“ dx f n u ` dy f ń v
ď cλ n u ` cλ n v “ cλ n w
for n P N. Similarly, for each
w “ pu, vq P E u pzq “ E u pxq ˆ E s pyq
we have
dx f ń u ď cλ n u and dy f n v ď cλ n v
for n P N, which gives
dz gń w ď cλ n w for n P N.
This shows that Λ ˆ Λ is a hyperbolic set for the diffeomorphism g.
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
An v ď cλ n v for v P E s , n P N
and
Ań v ď cλ n v for v P E u , n P N.
Solution Let F s Ď C p be the subspace generated by all vectors v P C p satisfying
pA ´ τ Idqk v “ 0 (II.4.5)
for some k P N and some eigenvalue τ of A with |τ | ă 1. Similarly, let F u Ď C p

be the subspace generated by all vectors v P C p satisfying (II.4.5) for some k P N
and some eigenvalue τ of A with |τ | ą 1. Since A is invertible, it follows from the
identities
pA ´ τ Idqk A “ ApA ´ τ Idqk
and
pA ´ τ Idqk A´1 “ A´1 pA ´ τ Idqk
that
AF s “ F s and AF u “ F u . (II.4.6)
Moreover, since A has no eigenvalues of modulus 1, we have Cp “ F s ‘ F u. There-
fore, letting
E s “ F s X R p and E u “ F u X R p ,
we obtain R p “ E s ‘ E u and it follows from (II.4.6) that
AE s “ E s and AE u “ E u .
Given ε ą 0, let R j “ λ j Id ` N for some eigenvalue λ j of B and let
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 ní i
Rnj “ λj N .
i“0
i
Therefore,
n´1
ÿˆ ˙
n
Bn ď S ¨ S´1 σ pBqní Ni
i“0
i
n ˆ ˙
ÿ n
ď D1 σ pBqní pD2 ε qi
i“0
i
˘n
“ D1 σ pBq ` D2 ε
`
for some constants D1 , D2 ą 0 with D2 independent of ε . Since σ pBq ă 1, one can

take ε sufficiently small such that σ pBq ` D2 ε ă 1. Then
˘n
An |E s ď Bn ď D1 σ pBq ` D2 ε .
`
A similar argument applies to Ań |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 ńVin
č
Λω “
nPZ

Solution For each ω “ p¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q and n P N, consider the set

n
f ´kVik
č
Rn pω q “
k“ń`1
(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 ńVin
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
card Λω “ 1 for each ω P t1, 2uZ .
Fig. II.4.1 Sets R1 pω q (in gray).
Problem 4.7 For the Smale horseshoe Λ , show that the map H : t1, 2uZ Ñ Λ
defined by
f ńVin
č
Hp¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q “
nPZ
is bijective (by Problem 4.6 the map is well defined).

Solution It follows from the construction of the Smale horseshoe that
f ń pV1 YV2 q
č
Λ“
nPZ
f ńVin
ď č
“
ω Pt1,2uZ nPZ
ď
“ Hpω q
ω Pt1,2uZ
and so H is onto. To show that it is one-to-one, take ω , ω 1 P t1, 2uZ with ω ‰ ω 1 .

Then there exists m P Z such that im pω q ‰ im pω 1 q and so
Vim pω q XVim pω 1 q “ ∅.
Hence,
˜ ¸ ˜ ¸
1 ń ń
č č
Hpω q X Hpω q “ f Vin pω q X f Vin pω 1 q “ ∅,
nPZ nPZ
which shows that Hpω q ‰ Hpω 1 q. Therefore, H is bijective.
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 ń`mVin
č
f mΛω “
nPZ
f ń`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.
Problem 4.9 Consider horizontal strips
Hi “ px, yq P r0, 1s2 : ϕi pxq ď y ď ψi pxq

(
and vertical strips
Vi “ px, yq P r0, 1s2 : ϕi pyq ď x ď ψi pyq ,

(
for some functions ϕi , ψi , ϕi , ψi : r0, 1s Ñ r0, 1s, for i “ 1, 2, such that
and
(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
f px, yq “ pgpxq, hpyqq for px, yq P H1 Y H2
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
is a hyperbolic set for f .
Solution Note that Λ is f -invariant. Indeed,
f ´1 Λ “ f ´1
č
f n pH1 Y H2 q
nPZ
č
“ fn´1
pH1 Y H2 q “ Λ .
nPZ
Now observe that

ˆ 1
g pxq 0
˙
dpx,yq f “ for px, yq P H1 Y H2 . (II.4.7)
0 h1 pyq
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
R2 “ E s px, yq ‘ E u px, yq.
Since the matrices in (II.4.7) are diagonal, we have
dx f E s px, yq “ E s p f px, yqq and dx f E u px, yq “ E u p f px, yqq
for all px, yq P Λ . Moreover, it follows from (II.4.7) that
dpx,yq f v ď sup |g1 | ¨ v for v P E s px, yq
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
λ “ max sup |g1 |, pinf |h1 |q´1 P p0, 1q.

(
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.
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
dim E s pyq “ dim E s pxq and dim E u pyq “ dim E u pxq
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
(
for k “ 0, 1, . . . , dim M. Note that
M “ Ak Y Bk and Ak X Bk “ ∅.
We show that Ak is closed. Take xn P Ak such that xn Ñ x P M when n Ñ 8. By the

continuity of the stable spaces, there exists N P N such that
dim E s pxn q “ dim E s pxq for n ą N. (II.4.8)
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
dim E s pxn q ‰ k for all n.
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
for any compact interval I Ĺ R.
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
for all n P N and x, y P I. Therefore,
| f n pIq| ď cλ n |I|
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
dp f n pxq, f n pyqq ď cλ n dpx, yq Ñ 0 when n Ñ 8 (II.4.10)
for all x, y P S1 . But since f is a diffeomorphism, we have f pS2 q “ S2 , which contra-

dicts to (II.4.10) because the area of any open set must eventually contract when we
apply f successively. This contradiction shows that no such diffeomorphism exists.
Problem 4.13 Given functions f , g : R Ñ R, let
dp f , gq “ sup | f pxq ´ gpxq| : x P R .

(
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
In order to show that f P X, note that
f p0q ` f p1q “ lim fn p0q ` lim fn p1q

nÑ8 nÑ8
“ lim p fn p0q ` fn p1qq “ 0
nÑ8
since fn P X for each n P N.
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
dp f , gq “ sup | f pxq ´ gpxq| : x P ra, bs .

(
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
b
fn pxq “ px ´ 1{2q2 ` 1{n for x P r´1, 1s
(see Figure II.4.2). Moreover, define f : r´1, 1s Ñ R by
f pxq “ |x ´ 1{2|.
Then dp fn , f q Ñ 0 when n Ñ 8, but the function f is not differentiable at 1{2.
f1
f 10
1 1 x
Fig. II.4.2 The functions fn and f in Problem 4.14.
Problem 4.15 Construct a sequence of bounded C1 functions fn : R Ñ R converg-

ing uniformly to a nondifferentiable function.
Solution For each n P N, let

$
&´x ´ 2n if x ď ´ 1n ,
’ 1
gn pxq “ 2 x
n 2 if ´ 1n ď x ď 1n ,
x ´ 2n1
if x ě 1n
’
%
and define a function fn : R Ñ R by

$
&´ arctanpx ` 1q ` 1 ´ 2n if x ď ´1,
’ 1
fn pxq “ gn pxq if x P r´1, 1s,

arctanpx ´ 1q ` 1 ´ 2n
1
if x ą 1
’
%
(see Figure II.4.3). Moreover, define f : R Ñ R by

$
&´ arctanpx ` 1q ` 1
’ if x ď ´1,
f pxq “ |x| if x P r´1, 1s,
arctanpx ´ 1q ` 1 if x ě 1.
’
%
One can easily verify that fn is a sequence of bounded C1 functions converging

uniformly to f . On the other hand, f is not differentiable at 0.
f1
f2
Fig. II.4.3 The functions fn and f in Problem 4.15.
Problem 4.16 Let Λ be a compact set that is invariant under a C1 diffeomorphism f

and assume that there exist a splitting Tx M “ F s pxq ‘ F u pxq for each x P Λ , an inner
product x¨, ÿ1x on Tx M varying continuously with x P Λ and a constant γ P p0, 1q such
that
dx f v1 ą v1 for x P Λ , v P Cu pxqzt0u.
Show that there exists a constant μ P p0, 1q such that
dx f v1 ě μ ´1 v1 for x P Λ , v P Cu pxq.
Solution Take x P Λ and v P Cu pxqzt0u such that v1 “ 1. Let
Sx “ Cu pxq X w P F u pxq : w1 “ 1 .

(
By the hypothesis we have
dx f v1 ą 1 for v P Sx .
Since the set tpx, vq : x P Λ , v P Sx u is compact, there exists μ P p0, 1q such that
dx f v1 ě μ ´1 for x P Λ , v P Sx .
Now take v P Cu pxqzt0u. Then

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.
Problem 4.17 Consider the set
A “ S1 ˆ px, yq P R2 : x2 ` y2 ď 1
(
and the map f : A Ñ S1 ˆ R2 defined by

ˆ ˙
1 1 1 1
f prθ s, x, yq “ r2θ s, x ` cosp2πθ q, y ` sinp2πθ q .
5 2 5 2
Show that f pAq Ĺ A and that f |A is one-to-one.
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,
1 “ cos2 p2πθ1 q ` sin2 p2πθ1 q

1 1
“ px1 ´ x2 q2 ` py1 ´ y2 q2
25 25
and so
1
1ď rpx1 ´ x2 q2 ` py1 ´ y2 q2 s
25
1“
ď p|x1 | ` |x2 |q2 ` p|y1 | ` |y2 |q2
‰
25
8
ď ă1
25
since |xi | ď 1 and |yi | ď 1 for i “ 1, 2. This contradiction shows that one cannot have
θ1 “ θ2 ` 1{2 and so θ1 “ θ2 .
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
TzΛ “ F u pzq ‘ F s pzq “ R ˆ R2
for each z “ pθ , x, yq. Since

ˆ ˙
1 1 1 1
dprθ s,x,yq f pu, v, wq “ 2u, ´ u sinp2πθ q ` v, u cosp2πθ q ` w ,
2 5 2 5
we obtain
dz f F s pzq “ F s p f pzqq. (II.4.13)
Moreover, for each z P A and p0, v, wq P F s pzq we have
ˆ
1
˙
1 1
dz f p0, v, wq “ 0, v, w

ď 5 p0, v, wq,
5 5
which together with (II.4.13) implies that
1
dz f n p0, v, wq ď p0, v, wq for n P N. (II.4.14)
5n
Now we show that the cones
" *
1 2
C pzq “ pu, v, wq : v ` w ă u Y tp0, 0, 0qu
u 2 2
4
are d f -invariant. Assume that v2 ` w2 ď 14 u2 and let
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
dz f Cu (z)
Cu ( f ( z))
Fig. II.4.4 The cone Cu p f pzqq and its image dz fCu pzq.
d f ń pzq f nCu p f ń pzqq

č
E u pzq Ď Gu pzq :“
ně0
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 ń pzq f nCu p f ń pzqq,
č
1 , v2 P
vm m
n“0
for m P N, has a converging subsequence to some linearly independent vectors

v1 , v2 P Gu pzq. Therefore, Gu pzq contains a 2-dimensional space. We show that in
fact
E u pzq “ Gu pzq.
Indeed, if there exists pu, v, wq P Gu pzqzE u pzq, then writing
pu, v, wq “ vu ` vs with vu P F u pzq and vs P E s pzqzt0u,
it follows from (II.4.16) that

´ 16 ¯ń `
dz f ń ppu, v, wq ´ vu q ď pu, v, wq ` vu Ñ 0
˘
5
when n Ñ 8. But since
pu, v, wq ´ vu “ vs P E s pzq,
we also have
1
dz f ń 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
dz f d f ń pzq f nCu p f ń pzqq

č
dz f Gu pzq “
ně0
d f ´pn`1q p f pzqq f n`1Cu p f ´pn`1q p f pzqqq

č
“
ně0
d f ´pn`1q p f pzqq f n`1Cu p f ´pn`1q p f pzqqq

č
“
ně´1
(because the sequence is decreasing), we obtain
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).
Solution We consider a cover of Tn by dm -open balls

m´1
TA´k BpTAk ppi q, ε q
č
Bm ppi , ε q :“
k“0
(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
C´1 p|λi |´1 ´ δ qk v ď A´k v ď Cp|λi |´1 ` δ qk v
for each k P N and each vector v P Cn in the generalized eigenspace associated to

the eigenvalue λi . Hence, there exists D ą 0 (independent of m, ε and i) such that
p|λi |´1 ` δ qm ε n .
ź
the n-volume of the ball Bm ppi , ε q is at most D
|λi |ą1
Let d be the dimension of the unstable spaces and let
λ “ mint|λi | : |λi | ą 1u ą 1.
Then
Mpm, ε q ě D´1 p|λi |´1 ` δ q´m ε ń
ź
|λi |ą1
ě D pλ ´1 ` δ q´dm ε ń .
´1
Hence, it follows from Theorem 2.14 that

1
hpTA q “ lim lim inf log Mpm, ε q ě ´d logpλ ´1 ` δ q.
ε Ñ0 mÑ8 m
Letting δ Ñ 0 we conclude that
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.
Solution We assume that h is differentiable on R` with hpR` q “ R` . It follows

from the identity h ˝ f “ g ˝ h that
h ˝ f n “ gn ˝ h
for all n P N, that is, for each x ą 0 we have
hpan xq “ bn hpxq for n P N. (II.4.17)
Now consider property (II.4.17) extended to t P R, that is,
hpat xq “ bt hpxq for t P R. (II.4.18)
Taking derivatives in (II.4.18) with respect to t, we obtain
h1 pat xqat x log a “ bt log bhpxq.
Moreover, taking t “ 0 yields the identity

h1 pxqx log a “ log bhpxq
and so hpxq “ xlog b{ log a , up to a multiplicative constant. This suggests considering

the function h : R Ñ R defined by
$
&x
’ log b{ log a if x ą 0,
hpxq “ 0 if x “ 0, (II.4.19)
´|x| log b{ log a if x ă 0.
’
%
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.
Problem 4.21 Construct a topological conjugacy between the maps f , g : R2 Ñ R2

given by
f px, yq “ p2x, 3yq and gpx, yq “ p5x, 4yq
for px, yq P R2 .
Solution Consider the maps

$
&x
’ log 5{ log 2 if x ą 0,
h1 pxq “ 0 if x “ 0,
´|x|log 5{ log 2 if x ă 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
’
%
(see (II.4.19)). Moreover, let fa pxq “ ax. By Problem 4.20, we have
h1 ˝ f2 “ f5 ˝ h1 and h2 ˝ f3 “ f4 ˝ h2 ,
Now consider the map h : R2 Ñ R2 given by

hpx, yq “ ph1 pxq, h2 pyqq,
which is a homeomorphism, with inverse h´1 : R2 Ñ R2 given by
h´1 px, yq “ ph´1 ´1

1 pxq, h2 pyqq.
We have
hp f px, yqq “ p2xqlog 5{ log 2 , p3yqlog 4{ log 3
` ˘
“ elog 5 xlog 5{ log 2 , elog 4 ylog 4{ log 3

` ˘
“ 5xlog 5{ log 2 , 4ylog 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.
Therefore, h is a topological conjugacy between f and g.
Problem 4.22 Determine the 2-periodic points of the automorphism of the torus TA
21
A“ .
11
Solution The 2-periodic points of TA are obtained solving the equation

ˆ ˙2 ˆ ˙ ˆ ˙ ˆ ˙
21 x 53 x
“
11 y 32 y
ˆ ˙
x
“
y
on T2 , that is, #
5x ` 3y ” x mod 1,
3x ` 2y ” y mod 1.
This gives
y ” ´3x mod 1, 5x ” 0 mod 1
and so the 2-periodic points of TA are
ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙
x 0 1{5 2{5 3{5 4{5
“ , , , , .
y 0 2{5 4{5 1{5 3{5
Problem 4.23 Determine the 2-periodic points of the endomorphism of the torus TB
31
B“ .
11
Solution The 2-periodic points of TB are obtained solving the equation

ˆ ˙2 ˆ ˙ ˆ ˙ˆ ˙ ˆ ˙
31 x 10 4 x x
“ “
11 y 4 2 y y
on T2 , that is, #
10x ` 4y ” x mod 1,
4x ` 2y ” y mod 1.
This gives
y ” ´4x mod 1, 7x ” 0 mod 1
and so the 2-periodic points of TB are
ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙ ˆ ˙
x 0 1{7 2{7 3{7 4{7 5{7 6{7
“ , , , , , , .
y 0 3{7 6{7 2{7 5{7 1{7 4{7
Problem 4.24 Show that the vector Xpxq “ dtd ϕt pxqˇt“0 in Definition 4.9 is neither
ˇ
contracted nor expanded by dx ϕt .
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.
Problem 4.25 Let Λ be a hyperbolic set for a flow Φ on R p . Given a sequence

xm Ñ x when m Ñ 8, with xm , x P Λ , for each m P N, show that any sublimit of a
sequence vm P E s pxm q Ď R p with vm “ 1 is in E s pxq.
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
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
dim E s px p q “ dim E s pxq q and dim E u px p q “ dim E s pxq q
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
v1 , . . . , vk P E s pxq and so dim E s pxq ě k.
Proceeding analogously for the unstable spaces, we obtain dim E u pxq ě p ´ 1 ´ k.

Since
R p “ E s pxq ‘ E 0 pxq ‘ E u pxq,
this implies that
dim E s pxq “ k and dim E u pxq “ p ´ 1 ´ k. (II.4.20)
In particular, the vectors v1 , . . . , vk generate E s pxq. If zm is another subsequence of xm

such that dim E s pzm q and dim E u pzm q are independent of m, respectively, with values
l and p ´ 1 ´ l, then
dim E s pxq “ l and dim E u pxq “ p ´ 1 ´ l.
Comparing with (II.4.20), we conclude that l “ k. This establishes the desired prop-
erty.
x1 “ x,
y1 “ ý ` 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 “ ý,
has eigenvalues ˘1, the origin is a hyperbolic fixed point. The stable and unstable
spaces are, respectively,
E s “ tp0, aq : a P Ru and E u “ tpa, 0q : a P Ru
(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´ ´ “ ý ` 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
“ ý ` 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.
x1 “ xpa ´ yq,
y1 “ ý ` 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 “ ý
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,
E s “ tp0, bq : b P Ru and E u “ tpb, 0q : b P Ru.
On the other hand, by Problem 1.32, the manifolds defined by
x2
x“0 and y“
1 ` 2a
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.
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.
’
%
is a topological conjugacy between the flows determined by the differential equa-

tions
x1 “ ax and x1 “ bx.
In other words, if ϕt and ψt are the flows determined by the equations, then
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.
Solution The flows determined by the differential equations are, respectively,
ϕt pvq “ eAt v and ψt pvq “ eBt v.
Now let h : Rn Ñ Rn be a diffeomorphism such that
hpeAt vq “ eBt hpvq.
Taking derivatives with respect to v, we obtain
deAt v heAt “ eBt dv h.
Letting v “ 0 yields the identity
CeAt “ eBt C, with C “ d0 h. (II.4.22)
Finally, taking derivatives in the identity h´1 phpvqq “ v we obtain
dhp0q h´1C “ Id,
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
gpeAt vq “ eBt gpvq (II.4.23)
for every t P R and v P Rn , that is,
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
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
for px, yq ‰ p0, 0q satisfies

h ˝ ϕt “ ψt ˝ h.
Solution One can easily verify that

ˆ ´t
e´t te´t
˙ ˆ ˙
e 0
eAt “ and eBt “ .
0 e´t 0 e´t
Therefore,
hpeAt px, yqq

“ hpe´t x, e´t yq
d
e´2t x2 ` e´2t y2 e´t y e´2t x2 ` e´2t y2 ´t
ˆ ˙
´t
“ e x ´ log , e y
e´2t x2 ` e´2t xy ` 3e´2t y2 {2 2 2
d
x2 ` y2 e´t y x2 ` y2 ´t
ˆ ˙
´t ´t
“ e x ` te y ´ log ,e y
x2 ` xy ` 3y2 {2 2 2
d
x2 ` y2 x2 ` y2
ˆ ˙
y
“e Bt
x ´ log , y
x2 ` xy ` 3y2 {2 2 2
“ eBt hpx, yq.

#
x1 “ ´x,
y1 “ y ` x2 ,
show that the map h : R2 Ñ R2 defined by
x2
ˆ ˙
hpx, yq “ x, y `
3
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
#
x1 “ ´x,
y1 “ y,
whose solution with pxp0q, yp0qq “ px0 , y0 q is given by
xptq “ x0 e´t , yptq “ y0 et (II.4.25)
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)

#
x1 “ x,
y1 “ ý ` xn ,
with n ě 2, show that the map h : R2 Ñ R2 defined by

xn
ˆ ˙
hpx, yq “ x, y ´
n`1
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 “ ý,
whose solution with pxp0q, yp0qq “ px0 , y0 q is given by
xptq “ x0 et , yptq “ y0 e´t (II.4.28)
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.
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
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.
Solution Writing z “ x ` iy and B “ B1 ` iB2 , the equation in the problem becomes
Apx2 ` y2 q ` 2B1 x ´ 2B2 y `C “ 0. (II.4.29)
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` À 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.
Solution If the equation in Problem 4.36 holds, then
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
and so T maps 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.
Solution If the equation in Problem 4.36 holds, then
z “ T ´1 pwq “ ´1 ´ 1{w
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
íet
“
pcet ´ idq2
iet
“ .
pciet ` dq2
Hence,
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 ì 2 (II.4.32)
d `c e2 2t d ` c2 e2t
bd ` ace2t et
“ 2 ì 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
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.
Solution It follows from (II.4.33) and (II.4.34) that
|γ 1 ptq|2 “ pRe γ 1 ptqq2 ` pIm γ 1 ptqq2

pd 2 ` c2 e2t q2 e2t
“
pd 2 ` c2 e2t q4
e2t
“
pd 2 ` c2 e2t q2
and
e2t
pIm γ ptqq2 “ .
pd 2 ` c2 e2t q2
Hence,
|γ 1 ptq|
H1 “ “1
Im γ ptq
On the other hand, it also follows from (II.4.33) that
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 ¨¨ïn :“ ω P Σk` : i j pω q “ i j for j “ 1, . . . , n .
Solution Note that ω 1 P Bdβ pω , rq if and only if dβ pω , ω 1 q ă r. By the definition

of dβ this is equivalent to require that
1 log r
β ńpω ,ω q ă r ðñ npω , ω 1 q ą ´
log β
(again npω , ω 1 q P N is the smallest positive integer n such that in pω q ‰ in pω 1 q).

Therefore, ω 1 P Bdβ pω , rq if and only if
Z ^
1 log r
i j pω q “ i j pω q for j ď nprq :“ ´ ,
log β
which is the same as ω 1 P Ci1 pω q¨¨ïnprq pω q .
Problem 5.2 Show that any cylinder Ci1 ¨¨ïn (see Problem 5.1) is simultaneously
open and closed.
Solution By Problem 5.1, any cylinder is an open ball.

To show that any cylinder is closed, note first that ω 2 R Ci1 ¨¨ïn if and only if
i j pω q ‰ i j for some j P t1, . . . , nu. If ω 1 P Ci1 ¨¨ïn , then
i j pω 1 q “ i j for j P t1, . . . , nu
and so dβ pω 2 , ω 1 q ě β ń , which shows that

170 II.5 Symbolic Dynamics
Bdβ pω 2 , β ń q XCi1 ¨¨ïn “ ∅.
Therefore, the complement of each cylinder Ci1 ¨¨ïn is open.
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
Bdβ 1 pω 1 , r1 q Ď Bdβ pω , rq. (II.5.1)
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
Bdβ pω 1 , r1 q Ď Bdβ 1 pω , rq.
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
Bdβ pω , rq “ Ci1 ¨¨ïnprq , with n “ nprq,
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
Bdβ 1 pω 1 , r1 q Ď Bdβ pω , rq for all r1 ď β ńprq .
This establishes property (II.5.1).
ÿ8
1
dpω , ω 1 q “ |i pω q ´ in pω 1 q|
n n
n“1
2
Solution Clearly, dpω , ω 1 q ě 0 and dpω , ω 1 q “ 0 if and only if in pω q “ in pω 1 q for

all n P N, that is, if and only if ω “ ω 1 . Moreover, dpω , ω 1 q “ dpω 1 , ω q and
dpω , ω 1 q ď dpω , ω 2 q ` dpω 2 , ω 1 q.
Hence, d is a distance on Σk` .

It remains to verify that the distances d and dβ generate the same topology. As in
Problem 5.3, 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
Bd pω 1 , r1 q Ď Bdβ pω , rq (II.5.2)
II.5 Symbolic Dynamics 171
and for each ω 1 P Bd pω , rq there exists r2 ą 0 such that
Bdβ pω 1 , r2 q Ď Bd pω , rq. (II.5.3)
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
in pω q “ in pω 1 q for n ď nprq. (II.5.4)
If ω 2 P Bd pω 1 , r1 q, that is, dpω 2 , ω 1 q ă r1 for some r1 ă 1{2nprq , then
in pω 2 q “ in pω 1 q for n ď nprq (II.5.5)
since otherwise we would have

1 1
dpω 2 , ω 1 q ě |in pω 2 q ´ in pω 1 q| ě n ą r1
2n 2
for some n ď nprq. This establishes property (II.5.2).
Now take ω 1 P Bd pω , rq. Again by Problem 5.1, the ball Bdβ pω 1 , r2 q is a cylinder
Ci1 ¨¨ïm with m Ñ 8 when r2 Ñ 0. Hence, for ω 2 P Bdβ pω 1 , r2 q we have
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).

#
1{npω , ω 1 q if ω ‰ ω 1 ,
dpω , ω 1 q “
0 if ω “ ω 1
Solution Clearly, dpω , ω 1 q ě 0 and dpω , ω 1 q “ 0 if and only if ω “ ω 1 . Moreover,

dpω , ω 1 q “ dpω 1 , ω q. For the triangle inequality, given ω , ω 1 , ω 2 P Σk` , let n1 , n2
and n3 be, respectively, the smallest positive integers such that
in1 pω q ‰ in1 pω 2 q, in2 pω q ‰ in2 pω 1 q, in3 pω 1 q ‰ in3 pω 2 q. (II.5.6)
Then
1 1 1
dpω , ω 2 q “ , dpω , ω 1 q “ , dpω 1 , ω 2 q “ .
n1 n2 n3
Observe that if n2 ą n1 and n3 ą n1 , then in1 pω q “ in1 pω 1 q “ in1 pω 2 q, which con-

tradicts (II.5.6). Hence, n2 ď n1 or n3 ď n1 , which gives
1 1 1 1
ď or ď .
n1 n2 n1 n3
This shows that
dpω , ω 1 q ď dpω , ω 2 q ` dpω 2 , ω 1 q
and so d is a distance on Σk` .
Now we verify that the distances d and dβ generate the same topology. As in
Problem 5.3, 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 satisfying (II.5.2) and for each ω 1 P B d pω , rq there exists r2 ą 0
satisfying (II.5.3). For the first property note that, by Problem 5.1, ω 1 P Bdβ pω , rq
if and only if there exists nprq P N satisfying (II.5.4). If ω 2 P B d pω 1 , r1 q for some
r1 ă 1{nprq, then property (II.5.5) holds since otherwise we would have
1 1
dpω 2 , ω 1 q ě ě ą r1
n nprq
for some n ď nprq. This establishes property (II.5.2).

Now take ω 1 P B d pω , rq. Again by Problem 5.1, the ball Bdβ pω 1 , r2 q is a cylinder
Ci1 ¨¨ïm with m Ñ 8 when r2 Ñ 0. Hence, for ω 2 P Bdβ pω 1 , r2 q we have
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
c´1 dpω , ωm q ď dpω , ωm q ď cdpω , ωm q for m P N
or, equivalently,
1 1 1
c´1 ď ď c m´1 for m P N.
2m´1 m 2
Multiplying by 2m´1 and letting m Ñ 8 yields a contradiction. Therefore, the dis-

tances d and d are not equivalent.
Problem 5.7 Show that pΣk` , dq is a compact metric space.
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
(again npω , ω 1 q P N is the smallest positive integer n such that in pω q ‰ in pω 1 q).

Therefore, ω 1 P B d pω , rq if and only if
Z ^
1
i j pω 1 q “ i j pω q for j ď nprq :“ .
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.
Problem 5.8 Show that pΣk` , dq is a complete metric space.
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,
dpω , ωm q ď dpω , ωkn q ` dpωkn , ωm q ď ε for k ě N
and so the sequence pωm qmPN converges.

Now we give an alternative argument without using the compactness of the space.
Again, let pωm qmPN be a Cauchy sequence in Σk` . Given ε ą 0, there exists N P N
such that
dpωi , ω j q ă ε for i, j ě N.
But dpωi , ω j q ă ε if and only if 1{npωi , ω j q ă ε . Let ε “ 1{p. Then npωi , ω j q ą p.
Hence, for each p P N there exists Nppq P N such that npωi , ω j q ą p for all i, j ě
Nppq, that is,
il pωi q “ il pω j q for all l P t1, . . . , pu.
We define a sequence ω P Σk` by
i p pω q “ i p pωNppq q for p P N.
Then i p pω q “ i p pωm q for all m ě Nppq and so dpω , ωm q ă 1{p. Letting m Ñ 8 we

obtain
1
lim sup dpω , ωm q ď
mÑ8 p
and so letting p Ñ 8 gives
lim sup dpω , ωm q “ 0.

mÑ8
This shows that ωm Ñ ω when m Ñ 8.
Problem 5.9 Show that the shift map σ |Σk is topologically transitive.
Solution First note that the sets

(
D j´m ¨¨¨ jm “ ω P Σk : in pω q “ jn for |n| ď m ,
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
ω 1 “ p¨ ¨ ¨ k´m ¨ ¨ ¨ km l1 ¨ ¨ ¨ l p´2m´1 i´m ¨ ¨ ¨ im ¨ ¨ ¨ q P Dk´m ¨¨¨km
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 ¨¨ïn´1
the pi, jq entry of An .

2. The number of n-periodic points of σ |ΣA` or σ |ΣA is the number of blocks of

length n with i “ j for some n P t1, . . . , ku. Hence, it follows from item 1 that
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 “ ∅.
Solution p1 ñ 2q. We have

ÿ
pAn qi1 in “ ai1 i2 ¨ ¨ ¨ ain´1 in “ 0
i2 ¨¨ïn´1
for all i1 , in P t1, . . . , ku and so An “ 0.

p2 ñ 3q. If An “ 0, then
ÿ
0 “ pAn qi1 in “ ai1 i2 ¨ ¨ ¨ ain´1 in
i2 ¨¨ïn´1
and since ai1 i2 ¨ ¨ ¨ ain´1 in ě 0, we conclude that
ai1 i2 ¨ ¨ ¨ ain´1 in “ 0 for all i1 , in P t1, . . . , ku.
Therefore, no block pi1 ¨ ¨ ¨ in q of length n can occur in an element of ΣA . In other

words ΣA “ ∅.
p3 ñ 1q. If ai1 i2 ¨ ¨ ¨ ain´1 in “ 1 for some n P N and i1 , . . . , in P t1, . . . , ku with i1 “ in ,
then ω “ p¨ ¨ ¨ ppp ¨ ¨ ¨ q, with p “ pi1 ¨ ¨ ¨ in q, would be an element of ΣA . But since
ΣA “ ∅ we must have
ai1 i2 ¨ ¨ ¨ ain´1 i1 “ 0
for all n P N and i1 , . . . , in P t1, . . . , ku. In particular, for blocks pi1 ¨ ¨ ¨ ik q of length k,
there exist m, n P t1, . . . , ku with m ă n such that im “ in and so
ai1 i2 ¨ ¨ ¨ aik´1 ik “ 0 because aim im`1 ¨ ¨ ¨ ain´1 in “ 0.
This establishes property 1.
odic orbits with period p prime in terms of the transition matrix A.
Solution By Problem 5.10, the number of p-periodic points of σ |ΣA` is equal to

trpA p q while the number of fixed points of σ |ΣA` is equal to tr A. Since p is prime,
the periodic points with period p are exactly the p-periodic points that are not fixed
points. Hence, the number of periodic orbits with period p is equal to
trpA p q ´ tr A
.
p
odic orbits with period p “ p1 p2 , with p1 , p2 prime, in terms of the transition
matrix A.
Solution If p “ p1 p2 , with p1 , p2 prime, then by Problem 5.10 the number of p-

periodic points is equal to trpA p q, the number of pi -periodic points is equal to trpA pi q,
and the number of fixed points is equal to tr A. Hence, the number of periodic points
with period p is equal to
trpA p q ´ ptrpA p1 q ´ tr Aq ´ ptrpA p2 q ´ tr Aq ´ tr A

“ trpA p q ´ trpA p1 q ´ trpA p2 q ` tr A,
which implies that the number of periodic orbits with period p is equal to
trpA p q ´ trpA p1 q ´ trpA p2 q ` tr A

.
p
Problem 5.14 Determine whether the matrix
¨ ˛
001
A “ ˝1 1 0‚
111
is irreducible or transitive (see Figure I.5.1 for its graph).
Solution Note that ¨ ˛

111
A2 “ ˝1 1 1‚.
222
Since all entries of A2 are positive, the matrix A is transitive and so it is also irre-
ducible (since any transitive matrix is irreducible).
Problem 5.15 Show that the matrix

¨ ˛
0 0 1 0
˚0 0 0 1‹
A“˚
˝0
‹
1 0 0‚
0 1 1 0
is not transitive (see Figure I.5.2 for its graph).

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
In particular, the matrix A is not transitive.

Alternatively, one can note that the graph in Figure I.5.2 has no arrows pointing
to 1. Hence, it follows from item 1 in Problem 5.10 that no power of A can have all
entries positive.
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
has positive topological entropy.
Solution Note that ¨ ˛

2 1 1 1
˚3 2 2 1‹
A2 “ ˚
˝2
‹.
2 3 1‚
2 3 2 2
Therefore, the matrix A is transitive and it follows from Proposition 5.9 that the
topological Markov chain has positive topological entropy.
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
1. Compute hpσ |ΣA` q and hpσ |ΣB` q.

2. Find the fixed points of σ |ΣA` and σ |ΣB` .
Solution 1. By Theorem 5.4, we have

?
1` 5
hpσ |ΣA` q “ log ρ pAq “ log
2
and
hpσ |ΣB` q “ log ρ pBq “ log 1 “ 0.
2. To determine the fixed points, note that tr A “ 1 and tr B “ 2. By Problem 5.10,

σ |ΣA` has one fixed point, which is p11 ¨ ¨ ¨ q, while σ |ΣB` has two fixed points,
namely, p11 ¨ ¨ ¨ q and p22 ¨ ¨ ¨ q.
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
hpσ |ΣA` q “ hpσ |ΣB` q ‰ 0
that are not topologically conjugate.
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.
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 ń Vin
nPZ
defines a bijective map H : t1, 2uZ Ñ Λ .
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“ń`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 |qń . Since
sup |g1 | ă 1 and inf |h1 | ą 1,
the diameter of Rn pω q tends to 0 when n Ñ 8 and so the intersection

8
č
Λω “ Rn pω q
n“1

To show that the map H is injective, take ω , ω 1 P t1, 2uZ with ω ‰ ω 1 . Writing
ω “ p¨ ¨ ¨ i´1 i0 i1 ¨ ¨ ¨ q and ω 1 “ p¨ ¨ ¨ i1´1 i10 i11 ¨ ¨ ¨ q,
we have i j ‰ i1j for some j P Z. Hence, Vi j ‰ Vi1j and so also

č č
f ń Vin ‰ f ń Vi1n ,
nPZ nPZ
that is, Hpω q ‰ Hpω 1 q. Finally, we have

č
Λ“ f ń`1 pH1 Y H2 q
nPZ
č
“ f ń pV1 Y V2 q
nPZ
ď č
“ f ń Vin
ω Pt1,2uZ nPZ
ď ` ˘
“ Λω “ H t1, 2uZ ,
ω Pt1,2uZ
which shows that the map H is onto.

Problem 5.21 Show that the bijection H : Σ2 Ñ Λ in Problem 5.20 is a topological

conjugacy between the shift map σ and the map f in Problem 4.9.
Solution To show that H is a homeomorphism, it suffices to verify that H is con-

tinuous. Indeed, since Σ2 and Λ are compact and H is a continuous bijection, it is
automatically a homeomorphism.
Note that if
ik pω q “ ik pω 1 q for k P t´m ` 1, . . . , mu,
then Rm pω q “ Rm pω 1 q (see (II.5.7)). Let
(
λ “ max sup |g1 |, pinf |h1 |q´1 ă 1.
Since Rm pω q is contained in a square of side λ m , we obtain

?
dpHpω q, Hpω 1 qq ď 2λ m .
?
Given ε ą 0, there exists m P N such that 2λ m ă ε and so if dpω , ω 1 q ă β ´m , then
Rm pω q “ Rm pω 1 q and
?
dpHpω q, Hpω 1 qq ď 2λ m ă ε .
Finally, we have
č
Hpσ pω qq “ f ń Vin`1 pω q
nPZ
č
“ f 1ń Vin pω q
nPZ
“ f pHpω qq.
Problem 5.22 Show that the set

(
X “ ω P Σk : ω is q-periodic for some even q P N
is dense in Σ2 .
Solution Given ω P Σ2 and p P N, consider the sequence

` ˘
ω p “ ¨ ¨ ¨ i´p`1 i´p`2 ¨ ¨ ¨ i p i´p`1 i´p`2 ¨ ¨ ¨ i p ¨ ¨ ¨
with i j “ i j pω q for j “ ´p ` 1, . . . , p. Clearly, ω p is 2p-periodic for the shift map

σ |Σ2 and npω , ω p q ą p. Therefore,
dpω , ω p q ă β ´p Ñ 0 when p Ñ 8.
This shows that the set X is dense in Σ2 .
Problem 5.23 Compute the zeta function of the shift map σ |Σk .
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
for |z| ă 1{k, it follows that

` ˘ 1
ζ pzq “ exp ´ logp1 ´ kzq “
1 ´ kz
for |z| ă 1{k.
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
Solution By Problem 5.10, the number of n-periodic points of σ |ΣA` is trpAn q.

Since
¨ ˛
1´λ 0 1
det ˝ 1 1 ´ λ 1 ‚“ ´λ 3 ` 3λ 2 ´ λ “ ´λ pλ 2 ´ 3λ ` 1q,
1 1 1´λ
the eigenvalues of A are

? ?
3` 5 3´ 5
0, τ“ and τ ´1 “ .
2 2
Hence,
ÿ8
pτ n ` τ ń qzn
ζ pzq “ exp
n“1
n
˜ ¸
ÿ8 8
pτ zqn ÿ pτ ´1 zqn
“ exp ` .
n“1
n n“1
n
Using (II.5.8) with k “ τ and k “ τ ´1 , we conclude that

` ˘
ζ pzq “ exp ´ logp1 ´ τ zq ´ logp1 ´ τ ´1 zq
1
“
p1 ´ τ zqp1 ´ τ ´1 zq
1
“
1 ´ 3z ` z2
?
for |z| ă mintτ ´1 , τ u “ p3 ´ 5q{2.
Problem 5.25 Compute the zeta function of the automorphism of the torus T2
induced by the matrix ˆ ˙
51
A“ .
92
Solution First observe that the eigenvalues of A are

? ?
7`3 5 ´1 7´3 5
τ“ and τ “ .
2 2
By Problem 1.26 we have
(
an “ card x P T2 : TAn pxq “ x
“ trpAn q ´ 2
“ τ n ` τ ń ´ 2
and so
ÿ8
pτ n ` τ ń ´ 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
Using (II.5.8) with k “ τ , k “ τ ´1 and k “ 1, we conclude that

` ˘
ζ pzq “ exp ´ logp1 ´ τ zq ´ logp1 ´ τ ´1 zq ` 2 logp1 ´ zq
for |z| ă mintτ ´1 , τ , 1u. Since τ ą 1, we obtain
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 .
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 .
Problem 5.27 Show that F11 X F21 “ pF1 Y F2 q1 .
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
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 iń pω q ‰ iń 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
that if F contains only blocks of length 2, then F1 “ ΣA for some k ˆ k matrix A with
entries in t0, 1u.
Solution Consider the k ˆ k matrix A with entries ai j P t0, 1u such that ai j “ 0 if

and only if pi jq P F. Then clearly ΣA “ F1 .
Problem 5.30 Show that for k “ 2 and F “ tp22qu, the matrix A in Problem 5.29 is
ˆ ˙
11
A“ .
10
Solution The entries ai j of the matrix A satisfy ai j “ 0 if and only if pi jq P F. Hence,

the only null entry of A is a22 .
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.
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 ń .
Problem 5.32 An orientation-preserving C1 map f : S1 Ñ S1 is called an expanding

map if
λ :“ min f 1 pxq ą 1
xPS1
(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
Ii0 ...in´1 “ rp j , p j`1 s, for j “ 0, . . . , 2n ´ 1,
satisfy:
1. |Ii0 ¨¨ïn´1 | ď λ ń ;
2. Ii0 ...in´1 in Ĺ Ii0 ...in´1 ;
Solution 1. For each n P N we have
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
Ii0 ...in´1 “ rp j , p j`1 s,
with i0 ¨ ¨ ¨ in´1 as in the statement of the problem. Clearly,

f n prp j , p j`1 sq “ f n pIi0 ¨¨ïn´1 q “ S1 (II.5.9)
and so ż p j`1
1“ p f n q1 pxq dx ě λ n |Ii0 ¨¨ïn´1 |.
pj
This yields property 1.

2. Take q P f ´1 pztpu. By (II.5.9), in each interval Ii0 ¨¨ïn´1 there exists a unique
point q j such that f n pq j q “ q. Since
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
Ii0 ¨¨ïn´1 0 “ rp j , q j s and Ii0 ¨¨ïn´1 1 “ rq j , p j`1 s,
which establishes property 2.

3. For the last property, note that the image of each interval Ii0 ¨¨ïn “ rp j , p j`1 s
under f is also an interval whose endpoints are preimages of p under f n´1 (since the
endpoints of Ii0 ¨¨ïn belong to f ń p). More precisely, let p1k , for k “ 0, . . . , 2n´1 ´ 1,
be the elements of f ´pn´1q p. Then
f pIi0 ¨¨ïn q “ r f pp j q, f pp j`1 qs
(since f is orientation preserving) with
f pp j q “ p1k for k ” j mod 2n´1 .
For j “ i0 ¨ ¨ ¨ in we have
j ” i1 ¨ ¨ ¨ in mod 2n´1
and so f pIi0 ¨¨ïn q “ Ii1 ¨¨ïn . The last property follows now by induction.
Problem 5.33 Let f : S1 Ñ S1 be an expanding Ť map (see Problem 5.32) with

deg f “ 2. Show that the set of all preimages nPN f ń p is dense in S1 .
Solution By Problem 5.32, the endpoints of each interval Ii0 ¨¨ïn´1 belong to f ń p.
Moreover,
|Ii0 ¨¨ïn´1 | ď λ ń Ñ 0 when n Ñ 8
and so the distance between two consecutive
Ť elements of f ń p tends to zero when
ń
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` .
Solution We define a map H : Σ2` Ñ S1 as follows. Given ω “ p j1 j2 ¨ ¨ ¨ q P Σ2` , let

il “ jl ´ 1 for each l and define
č
Hpω q “ Ii1 ¨¨ïn ,
nPN
using the same notation as in Problem 5.32. By item 2 in that problem the inter-
vals Ii1 ¨¨ïn form a decreasing sequence of closed sets and so their intersection is
nonempty. Moreover, since
|Ii1 ¨¨ïn | Ñ 0 when n Ñ 8,
Ş
the set nPN Ii1 ¨¨ïn 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 ń . Writing
ω “ p j1 j2 ¨ ¨ ¨ q and ω 1 “ p j11 j21 ¨ ¨ ¨ q,
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
dβ pHpω q, Hpω 1 qq ď |Ii1 ¨¨ïn´1 | ď λ ń Ñ 0
when β ń Ñ 0 and so the map H is continuous. To show that it is onto, take x P S1

and for each n P N let #
1 if f n´1 pxq P rp, qq,
in “
2 otherwise,
with p and q as in Problem 5.32. Then x P Ii1 ¨¨ïn for all n P N and so x “ hpi1 i2 ¨ ¨ ¨ q.
Finally, we show that
h ˝ σ “ f ˝ H on Σ2` .
Let x “ Hpω q. Then x P Ii1 ¨¨ïn for all n P N. By Problem 5.32 we have
f pIi1 ¨¨ïn q “ Ii2 ¨¨ïn
and so č
f pxq “ Ii2 ¨¨ïn . (II.5.10)
nPN
Therefore,
f pHpω qq “ f pxq “ Hpσ pω qq. (II.5.11)
Problem 5.35 Let f : S1 Ñ S1 be an expanding map with deg f “ k ě 2. Show that

there exists a continuous onto map H : Σk` Ñ S1 such that H ˝ σ “ f ˝ H on Σk` .
Solution Let i0 ¨ ¨ ¨ in´1 be the base-k representation of j and let p1 , . . . , pkn “ p0 be

the preimages of a given fixed point p of f under f n . Proceeding in a similar manner
to that in Problem 5.32, we find that the intervals
Ii0 ...in´1 “ rp j , p j`1 s, for j “ 0, . . . , kn ´ 1,

satisfy:
1. |Ii0 ¨¨ïn´1 | ď λ ń ;
2. Ii0 ...in´1 in Ĺ Ii0 ...in´1 ;
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 ¨¨ïn .
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 ¨¨ïn and
f pIi1 ¨¨ïn q “ Ii2 ¨¨ïn
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
f n pHpω qq “ Hpσ n pω qq “ Hpω q.
That is, the image of an n-periodic point of σ under H is an n-periodic point of f .

Since the periodic points of σ are dense in Σk` and H is continuous and onto, the
periodic points of f are also dense.
Problem 5.38 Show that any expanding map f : S1 Ñ S1 is topologically transitive.
Solution First note that any map as in the statement of the problem has degree
deg f ě 2. Hence, the hypotheses of Problems 5.34 and 5.35 are satisfied. Given
nonempty open sets U,V Ĺ S1 , take intervals
Ii0 ¨¨ïn´1 Ď U and Ii1 ¨¨ï1 ĎV (II.5.12)

0 n´1
as introduced in those problems. Then
U X f nV Ě Ii0 ¨¨ïn´1 X f n pIi1 ¨¨ï1 q

0 n´1
“ Ii0 ¨¨ïn´1 X S1 ‰ ∅.
It follows from Problem 2.17 that f is topologically transitive.

Problem 5.39 Determine whether any expanding map f : S1 Ñ S1 is topologically

mixing.
Solution Given nonempty open sets U,V Ĺ S1 , take intervals Ii0 ¨¨ïn´1 and Ii1 ¨¨ï1
0 n´1
as in (II.5.12). Note that for each m ą n we have
f ´m Ii1 ¨¨ï1 Ě Ii0 ¨¨ïn´1 j1 ¨¨¨ jmń i1 ¨¨ï1

0 n´1 0 n´1
for any j1 , . . . , jmń P t1, . . . , ku. Hence,
U X f ´mV Ě Ii0 ¨¨ïn´1 X f ´m Ii1 ¨¨ï1

0 n´1
Ě Ii0 ¨¨ïn´1 j1 ¨¨¨ jmń i1 ¨¨ï1 ‰ ∅,

0 n´1
which shows that f is topologically mixing.
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
hp f q ď hpσ |Σk` q “ log k.

Chapter II.6
Ergodic Theory

1. if A, B P A and A Ď B, then μ pAq ď μ pBq;
2. if Bn P A for all n P N, then
˜ ¸
8
ď 8
ÿ
μ Bn ď μ pBn q.
n“1 n“1
Solution 1. Since
B “ A Y pBzAq and A X pBzAq “ ∅,
it follows from Definition 6.2 that
μ pBq “ μ pAq ` μ pBzAq.
Therefore, μ pBq ě μ pAq.

2. Let
ď
n´1
A1 “ B1 and An “ Bn z Bi for n ą 1.
i“1
Then An X Am “ ∅ for n ‰ m and so

˜ ¸ ˜ ¸
8
ď ď8 8
ÿ 8
ÿ
μ Bn “ μ An “ μ pAn q ď μ pBn q.
n“1 n“1 n“1 n“1

1. if A, B P B, then
` ˘ ` ˘
μ pAq ` μ pBq “ μ A Y B ` μ A X B ;

190 II.6 Ergodic Theory
2. if A, B P B and ` ˘
μ pAzBq Y pBzAq “ 0,
then μ pAq “ μ pBq.
Solution 1. Note that
A Y B “ A Y pBzAq and B “ pBzAq Y pA X Bq
are both disjoint unions. Hence,
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.
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.
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.
Solution 1. As in Problem 6.3, it suffices to note that

8 ´
č ¯
1
pa, bq “ pa, `8qz b ´ , `8
n“1
n
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 “ pń, 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.
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 “ ń, 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.
Solution Clearly, the Borel σ -algebra on R is contained in the σ -algebra C gen-

erated by the open sets in R. Hence, it suffices to show that any open set A Ď R
is a countable union of open intervals (since then any set in C is also in the Borel
σ -algebra).
Let A Ď R be an open set. For each x P A X Q, let Ix Ď A be the largest open
interval containing x (that is, the union of all open intervals I Ď A containing x).
Then ď
A“ Ix
xPAXQ
is a countable union of open intervals.
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.
Solution Clearly, if ϕ is A-measurable, then ϕ ´1 B P A for every B P C. Now

assume that (
F :“ B P B : ϕ ´1 B P A Ě C.
Then
1. ∅, R P F;
2. if B P F, then
ϕ ´1 pRzBq “ Rzϕ ´1 pBq P A
and so RzB P F;
3. given sets Bn P F, for n P N, we have
8
ď 8
ď
ϕ ´1 Bn “ ϕ ´1 Bn P A
n“1 n“1
Ť
and so nPN Bn P F.
Therefore, F is a σ -algebra containing C. Since C generates the Borel σ -algebra
on R, we conclude that F Ě B and so in particular ϕ is A-measurable.
Problem 6.7 Show that the square of an A-measurable function ϕ : X Ñ R is also

an A-measurable function.
Solution For each a ą 0 we have

( ? ? (
x P X : ϕ pxq2 ă a “ x P X : ´ a ă ϕ pxq ă a
` ? ? ˘
“ ϕ ´1 ´ a, a .
Hence, it follows from Problems 6.4 and 6.6 that the function ϕ 2 is A-measurable.
Problem 6.8 Show that any increasing function ϕ : R Ñ R is B-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.

supnPN ϕn and infnPN ϕn (here assumed to be finite everywhere) are also A-mea-
surable functions.
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
lim sup ϕn and lim inf ϕn

nÑ8 nÑ8
(here assumed to be finite everywhere) are also A-measurable functions.
Solution Note that
lim sup ϕn “ inf sup ϕm and lim inf ϕn “ sup inf ϕm .

nÑ8 ně1 měn nÑ8 ně1 měn
Hence, the desired property follows readily from Problem 6.9.
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
ϕn pxq Ñ ϕ pxq when n Ñ 8
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.
Solution For each n, q P N, let

8 !
ď 1)
Bn pqq “ x P X : |ϕk pxq ´ ϕ pxq| ě .
k“n
q
Note that for each q the sequence Bn pqq decreases with n and that
8
č
Bn pqq “ ∅.
n“1
Now let
An “ Bn pqqzBn`1 pqq for n ě 1.

Note that An X Am “ ∅ for n ‰ m and that
8
ď
Bn pqq “ Am
m“n
for n P N. Since μ pXq ă `8, we obtain

˜ ¸
8
ÿ 8
ď
μ pAm q “ μ Am “ μ pBn pqqq ă `8
m“n m“n
for each n P N. Therefore,

8
ÿ
lim μ pBn pqqq “ lim μ pAm q “ 0.
nÑ8 nÑ8
m“n
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
By Problem 6.1 we obtain
ÿ8 8
` ˘ ÿ ε
μ pBq ď μ Bnq pqq ă q
“ ε.
q“1 q“1
2
On the other hand, for n ą nq and x P XzB we have
1
|ϕn pxq ´ ϕ pxq| ă .
q
Therefore, 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
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.
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
Since Bi X B j “ ∅ for i ‰ j, we obtain
ÿ
m
μ pEm´1 Bq “ λ pBi q
i“1
ÿm
λ pBq
“
i“1
m
“ λ pBq “ μ pBq
and so the measure μ is Em -invariant.
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.
Solution For each a P r0, 1s we have

ď
f ´1 p0, aq “ pai , ai ` apbi ´ ai qq
iPI
and the union is disjoint. Hence,

` ˘ ÿ ` ˘
λ f ´1 p0, aq “ λ pai , ai ` apbi ´ ai qq
iPI
ÿ
“ apbi ´ ai q
iPI
` ˘
“ a “ λ p0, aq .
Since the intervals p0, aq generate the Borel σ -algebra on r0, 1s, it follows that the
measure λ is f -invariant.

#
x{p1 ´ xq if x P r0, 1{2s,
f pxq “
p1 ´ xq{x if x P r1{2, 1s
(see Figure I.6.2) preserves the measure μ on r0, 1s defined by

ż
dx
μ pAq “ .
A x
Solution Note that żb

dx
μ pra, bsq “ “ log b ´ log a.
a x
Since ” a b ı ” 1 1 ı
f ´1 ra, bs “ , Y ,
1à 1`b 1`b 1à
(and the intervals on the right-hand side are disjoint), we obtain
ˆ” ˙ ˆ” ˙
` ˘ a b ı 1 1 ı
μ f ´1 ra, bs “ μ , `μ ,
1à 1`b 1`b 1à
b a 1 1
“ log ´ log ` log ´ log
1`b 1à 1à 1`b
“ log b ´ log a “ μ pra, bsq.
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.
Problem 6.16 Consider the Gauss map f : r0, 1s Ñ r0, 1s defined by

#
t1{xu if x ‰ 0,
f pxq “
0 if x “ 0
(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
nà n

` ˘ ÿ 8 ´” 1 1 ¯¯
μ f ´1 r0, aq “ μ ,
n“1
nà n
ÿ8 ż 1{n
dx
“
n“1 1{pnàq
1 `x
ˆ ´ ˙
1¯ ´ 1 ¯
ÿ8
“ log 1 ` ´ log 1 `
n“1
n nà
ÿ8 ˆ ˙
n`1 n
“ log ´ log
n“1
n`1à nà
ża
1 dx
“ ´ log “ “ μ pr0, aqq
1à 0 1 `x
and so the measure μ is f -invariant.
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
Solution Note that

´ ´ x ¯¯1 1{a 1
arcsin “a “?
a 1 ´ x2 {a2 a ´ x2
2
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
”1 1? 1 1? ı ”1 1? 1 1? ı
g´1 ra, bs “ ´ 1 ´ a, ´ 1´b Y ` 1 ´ b, ` 1á
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, ` 1á
2 2 2 2
? ?
“ arcsinp´ 1 ´ bq ´ arcsinp´ 1 ´ aq
? ?
` arcsinp 1 ´ aq ´ arcsinp 1 ´ bq
? ?
“ 2 arcsinp 1 ´ aq ´ 2 arcsinp 1 ´ bq.
Now observe that

?
? p 1 ´ xq1 1
parcsinp 1 ´ xqq1 “ a “´ a .
1 ´ p1 ´ xq 2 xp1 ´ xq
Comparing with (II.6.1) and taking x “ 1, we conclude that

? π
arcsinp2x ´ 1q ` 2 arcsinp 1 ´ xq “
2
for all x P r0, 1s. Therefore,
` ˘ ? ?
μ g´1 ra, bs “ 2 arcsinp 1 ´ aq ´ 2 arcsinp 1 ´ bq
“ arcsinp2b ´ 1q ´ arcsinp2a ´ 1q
“ μ pra, bsq.
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.
Problem 6.18 Given α P R, show that the map f : T2 Ñ T2 defined by
px, yq ÞÑ px ` α , x ` yq
preserves the measure m on T2 obtained from the Lebesgue measure on r0, 1s2 .
Solution Let
h : px, yq ÞÑ px, x ` yq and g : px, yq ÞÑ px ` α , yq.
Then f “ g ˝ h. Since the map Hpx, yq “ px, x ` yq on R2 has Jacobian

ˆ ˙
10
det dpx,yq H “ det “ 1,
11
for each B Ď r0, 1s2 in B2 we have

ż ż ż
mphpBqq “ 1“ |det dpx,yq H| dxdy “ 1 “ mpBq
hpBq B B
and so the measure m is h-invariant. On the other hand, since g is a translation, m is

also g-invariant. Hence, for each set A Ď r0, 1s2 in B2 (identifying T2 with r0, 1s2 )
we have ` ˘ ` ˘
m f ´1 A “ m pg ˝ hq´1 A
` ˘
“ m h´1 pg´1 Aq
“ mpg´1 Aq
“ mpAq
and so the measure m is f -invariant.
Problem 6.19 Let f be the Gauss map in Problem 6.16. Show that if
1
x “ ra1 , a2 , . . .s “
a1 ` a2 `¨¨¨
1
is the continued fraction of a number x P p0, 1q, then:

1. f pxq “ ra2 , a3 , . . .s;
2. letting
p´1 “ 0, p0 “ 1, pm “ am pm´1 ` pm´2
and
q´1 “ 1, q0 “ 0, qm “ am qm´1 ` qm´2
for m P N, we have
pm qm´1 ´ pm´1 qm “ p´1qm´1 for m ě 0;
3. for m P N we have
1 pm
ra1 , a2 , . . . , am s :“ “ .
a1 ` 1
1
qm
a2 `
¨¨¨` a1m
Solution 1. It is clear that f pxq “ ra2 , a3 , . . .s.

2. The identity is immediate for m “ 0. Now assume that it holds with m replaced
by m ´ 1. Then
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 ,
which establishes the identity.

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
By the formulas in item 2 with m replaced by m ´ 1 and with am´1 replaced by

am´1 ` 1{am , we obtain
` ˘
am´1 ` a1m pm´2 ` pm´3
ra1 , . . . , am s “ ` ˘
am´1 ` a1m qm´2 ` qm´3
am pam´1 pm´2 ` pm´3 q ` pm´2
“
am pam´1 qm´2 ` qm´3 q ` qm´2
am pm´1 ` pm´2
“
am qm´1 ` am´2
pm
“ .
qm
Problem 6.20 Let P “ ppi j q be a k ˆ k matrix with entries pi j ě 0 for i, j “ 1, . . . , k

such that
ÿk
pi j “ 1 for j “ 1, . . . , k.
j“1
Moreover, take p1 , . . . , pk P p0, 1q such that
ÿ
k ÿ
k
pi “ 1 and pi pi j “ p j
i“1 i“1
for j “ 1, . . . , k. We define a measure μ on the σ -algebra on Σk` generated by the

cylinders (see Problem 5.1) by
μ pCi1 ¨¨ïn q “ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in
for each n P N and i1 , . . . , in P t1, . . . , ku. Show that:

ř
1. μ pCi1 ¨¨ïn q “ kj“1 μ pCi1 ¨¨ïn j q;
2. μ is σ -invariant; and
3. μ pΣk` q “ 1.
Solution 1. We have
ÿ
k ÿ
k
μ pCi1 ¨¨ïn j q “ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in pin j
j“1 j“1
“ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in

“ μ pCi1 ¨¨ïn q.
2. Note that
ď
k
σ ´1Ci1 ¨¨ïn “ C ji1 ¨¨ïn
j“1
` ˘ ÿk
μ σ ´1Ci1 ¨¨ïn “ μ pC ji1 ¨¨ïn q
j“1
ÿ
k
“ p j p ji1 ¨ ¨ ¨ pin´1 in
j“1
“ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in

“ μ pCi1 ¨¨ïn q.
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.
Show that supp μ “ ΣA` .
Solution Given pi1 i2 ¨ ¨ ¨ q P ΣA` and n P N, we have
μ pCi1 ¨¨ïn q “ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in ą 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 ¨¨ïn`1 q “ 0.
Since pi1 i2 ¨ ¨ ¨ q P Ci1 ¨¨ïn`1 , we conclude that pi1 i2 ¨ ¨ ¨ q R supp μ and so supp μ Ď ΣA` .

on X. Given a set A P A with μ pAq ą 0, show that there exist integers m, n ě 0 with
m ‰ n such that
f ´m A X f ń A ‰ ∅.
Solution We proceed by contradiction. Assume that
f ´m A X f ń A “ ∅
for any integers m, n ě 0 with m ‰ n. Since μ is f -invariant, we obtain

˜ ¸
8
ď ÿ8
í
` ˘
μ f A “ μ f í A
i“1 i“1
ÿ8
“ μ pAq “ `8,
i“1
but this is impossible because the measure μ is finite. This contradiction yields the
desired property.

on X. Given a set A P A with μ pAq ą 0, show that the map τA : X Ñ N given by
τA pxq “ mintn P N : f n pxq P Au
is well defined for almost every x P A.
Solution By Poincaré’s recurrence theorem (Theorem 6.9), we have

` (˘
μ x P A : f n pxq P A for infinitely many integers n P N “ μ pAq.
In particular, the set of points in A that eventually return to A applying f successively

has the same measure as the set A. This shows that the number τA pxq is well defined
for almost every x P A.
Problem 6.24 Show that a function ϕ : X Ñ R satisfies ϕ ˝ f “ ϕ for some map

f : X Ñ X if and only if the set ϕ ´1 α is f -invariant for every α P R.
Solution The set ϕ ´1 α is f -invariant if and only if
ϕ ´1 α “ f ´1 pϕ ´1 α q “ pϕ ˝ f q´1 α .
This is equivalent to require that
x P ϕ ´1 α if and only if x P 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
for every x P X, that is, if and only if ϕ is f -invariant.

Problem 6.25 Given α P Q, consider the rotation Rα : S1 Ñ S1 . Let U “ ra, bs be

an interval in S1 with 0 ď a ă b ď 1 and define
(
Tn pxq “ card 1 ď k ď n : Rkα pxq P U .
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
2. the former property holds for all continuous functions ϕ : S1 Ñ R; and

3. for each x P S1 we have
Tn pxq
lim “ b ´ a.
nÑ8 n
Solution 1. Let am “ e2π imα . For the function ϕ pxq “ e2π imx we have
ϕ pRα pxqq “ e2π impx`α q

“ e2π imα e2π imx
“ am ϕ pxq.
It follows by induction that
ϕ pRkα pxqq “ akm ϕ pxq for k P N.
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
On the other hand, for m “ 0 we have
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.
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 .
Hence, given ε ą 0, there exists such a linear combination ψ with ϕ ´ ψ ă ε {2.

Letting
ε ε
ψ1 pxq “ ψ pxq ´ and ψ2 pxq “ ψ pxq `
2 2
for x P S1 , we have
ψ1 pxq ď ϕ pxq ď ψ2 pxq
and ż
pψ2 ´ ψ1 q d μ “ ε .
S1
Since ϕ ď ψ2 , we obtain
ż ż
pϕ ´ ψ1 q d μ ď pψ2 ´ ψ1 q d μ “ ε
S1 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
Given ε ą 0, there exist continuous functions ψ1 , ψ2 : S1 Ñ R with
ψ1 ď χU ď ψ2
and
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

on X is said to be ergodic (with respect to f ) if any f -invariant set has either mea-
sure 0 or measure 1. Show that if
1 ÿ ` ´k ˘
n´1
lim μ f A X B “ μ pAqμ pBq
nÑ8 n
k“0
for any sets A, B P A, then the measure μ is ergodic.

Solution Let A Ď X be an f -invariant set. By the hypothesis we obtain
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
because f ´k A “ A for every k ě 0. Thus, either μ pAq “ 0 or μ pXzAq “ 0 and so the

measure μ is ergodic.
Problem 6.27 Let f : X Ñ X be an A-measurable map. Show that an f -invariant
measure μ on X is ergodic (see Problem 6.26) if and only if
˜ ¸
8
ď
ń
μ f A “1
n“1
for any set A P A with μ pAq ą 0.

Solution First assume that μ is ergodic and consider the set
8
ď
B“ f ń A. (II.6.6)
n“1
Clearly, f ń B Ď B for all n P N. Since the measure μ is f -invariant, we conclude

that μ pBz f ń Bq “ 0 for all n P N. Now we consider the set
8 ď
č 8
C“ f ´k B Ď B.
n“1 k“n
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
and so μ pBzCq “ 0. Moreover, since
μ p f ń Aq “ μ pAq ą 0
and B Ě f ń A, we conclude that
μ pCq “ μ pBq ą 0,
which implies that μ pCq “ 1. Therefore,
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 ń A “ A for all n P N.
Then B “ A, which implies that
μ pAq “ μ pBq “ 1.
Problem 6.28 Let f : X Ñ X be an A-measurable map preserving an ergodic prob-

ability measure μ on X. Given a set A P A with μ pAq ą 0, show that
ż
τA d μ “ 1
A
with the function τA as in Problem 6.23.
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 “ ∅
for any n, m P N with n ‰ m. We have

8
ď
f ´k Az Bn “ ∅ for all k P N.
n“1
Ť
On the other hand, by Problem 6.27, μ p 8
n“1 f
ń Aq “ 1 and so also
˜ ¸
ď8
μ Bn “ 1.
n“1
Now observe that f ´1 pBn zAn q “ Bn`1 and
μ pBn q “ μ pAn q ` μ pBn zAn q

“ μ pAn q ` μ p f ´1 pBn zAn qq
“ μ pAn q ` μ pBn`1 q.
Proceeding inductively, we obtain

ÿ
μ pBn q “ μ pAk q
kěn
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

on X is said to be mixing (with respect to f ) if
` ˘
lim μ f ń A X B “ μ pAqμ pBq
nÑ8
for any sets A, B P B and is said to be weakly mixing (with respect to f ) if
1 ÿ ` í ˘
n´1
lim μ f A X B ´ μ pAqμ pBq “ 0
nÑ8 n
i“0
for any sets A, B P B. Show that if μ is mixing, then it is weakly mixing.
Solution Assume that μ is mixing. Given ε ą 0, there exists p P N such that

` í ˘
μ f A X B ´ μ pAqμ pBq ă ε
for all i ě p. Therefore,
1 ÿ ` í ˘ 2p n ´ p
n´1
μ f A X B ´ μ pAqμ pBq ď ` ε
n i“0 n n
(separating the terms with i ă p and i ě p). Letting n Ñ 8 we obtain
1 ÿ ` í ˘
n´1
lim sup μ f A X B ´ μ pAqμ pBq ď ε
nÑ8 n i“0
and since ε is arbitrary, it follows that μ is weakly mixing.
Problem 6.30 Let f : X Ñ X be an A-measurable map. Show that if a probability

measure μ on X is weakly mixing (see Problem 6.29), then it is ergodic.
Solution Assume that μ is weakly mixing and let A P B be an f -invariant set. It

follows from the definition of weakly mixing measure with A “ B that
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.
Problem 6.31 Let f : X Ñ X be an A-measurable map preserving a mixing proba-

bility measure μ on X (see Problem 6.29). Show that f is topologically mixing on
the support supp μ of μ .
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 ńU XV “ μ pUqμ pV q ą 0.
nÑ8
Hence,
μ p f ńU XV q ą 0
for all sufficiently large n, which implies that
f ńU XV ‰ ∅
also for all sufficiently large n. Therefore, f is topologically mixing on supp μ .
Problem 6.32 Let μ be a probability measure on X. Show that if ξ is a partition

of X, then
Hμ pξ q ď log card ξ .
Solution Let ψ : X Ñ R be the continuous function defined by

#
x log x if x ‰ 0,
ψ pxq “ (II.6.7)
0 if x “ 0.
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
Problem 6.33 Let μ be a probability measure on X and let ξ and η be partitions

of X. Show that if η is a refinement of ξ (that is, if for each D P η there exists C P ξ
such that μ pDzCq “ 0), then
Hμ pξ q ď Hμ pη q.
Solution Note that ÿ

Hμ pη q “ ´ ψ pμ pDqq
DPη
ÿ ÿ (II.6.9)
“´ ψ pμ pDqq
CPξ DĎC
with ψ as in (II.6.7). For each C P ξ , we have

ÿ
μ pDq “ μ pCq
DĎC
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
Therefore, it follows from (II.6.9) that

ÿ
Hμ pξ q “ ´ ψ pμ pCqq
CPξ
ÿ ÿ
ď´ ψ pμ pDqq “ Hμ pη q.
CPξ 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
˜ ¸
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.

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.
Solution Since the measure μ is f -invariant, we obtain

ÿ
Hμ p f ´1 ξ q “ ´ μ p f ´1Cq log μ p f ´1Cq
CPξ
ÿ
“´ μ pCq log μ pCq “ Hμ pξ q.
CPξ

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 í ξ “ lim Hμ f í ξ .
nPN n nÑ8 n
i“0 i“0
Solution First note that

ł
n´1
ξn “ f í ξ for each n P N.
i“0
Hence, by Problems 6.34 and 6.35, for each n, m P N we have
Hμ pξn`m q “ Hμ pξn _ f ń ξm q
ď Hμ pξn q ` Hμ p f ń 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.
Problem 6.37 Let μ be a probability measure on X and let ξ , ζ , and η be partitions

of X. Show that if ζ is a refinement of η (see Problem 6.33) and
ÿ μ pC X Dq
Hμ pξ |ζ q “ ´ μ pC X Dq log ,
μ pDq
CPξ ,DPζ
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
It follows from the convexity of ψ (see Problem 6.32) that

˜ ¸
ÿ ÿ μ pD X Eq μ pC X Dq
Hμ pξ |ζ q ď ´ μ pEqψ ¨ .
μ pEq μ pDq
CPξ ,EPη DPζ
Finally, since ζ is a refinement of η , we obtain

ÿ μ pD X Eq μ pC X Dq ÿ μ pC X Dq μ pC X Eq
¨ “ “
μ pEq μ pDq μ pEq μ pEq
DPζ DPζ ,DĎE
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.
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.
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.
Problem 6.39 For the measure μ in Problem 6.20, compute hμ pσ q.
Solution Consider the partition ξ “ tC1 , . . . ,Ck u. Then

ÿ
Hμ pξn q “ ´ μ pCi1 ¨¨ïn q log μ pCi1 ¨¨ïn q
i1 ¨¨ïn
ÿ ` ˘
“´ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in log pi1 pi1 i2 ¨ ¨ ¨ pin´1 in
i1 ¨¨ïn
ÿ
“´ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in log pi1
i1 ¨¨ïn
ÿ ÿ
n´1
´ pi1 pi1 i2 ¨ ¨ ¨ pin´1 in log pi j i j`1
i1 ¨¨ïn 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
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.
Solution It follows from (II.6.10) that
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
1. R. Abraham, J. Marsden, Foundations of Mechanics (Benjamin/Cummings, MA, 1978)

2. L. Alsedà, J. Llibre, M. Misiurewicz, Combinatorial Dynamics and Entropy in Dimension
One. Advanced Series in Nonlinear Dynamics, vol. 5 (World Scientific, NJ, 2000)
3. H. Amann, Ordinary Differential Equations: An Introduction to Nonlinear Analysis. de
Gruyter Studies in Mathematics, vol. 13 (Walter de Gruyter, Berlin, 1990)
4. J. Anderson, Hyperbolic Geometry. Springer Undergraduate Mathematics Series (Springer,
London, 2005)
5. V. Arnol’d, Geometrical Methods in the Theory of Ordinary Differential Equations.
Grundlehren der Mathematischen Wissenschaften, vol. 250 (Springer, NY, 1988)
6. V. Arnol’d, Mathematical Methods of Classical Mechanics. Graduate Texts in Mathematics,
vol. 60 (Springer, NY, 1989)
7. L. Barreira, Dimension and Recurrence in Hyperbolic Dynamics. Progress in Mathematics,
vol. 272 (Birkhäuser, Basel, 2008)
8. L. Barreira, Ergodic Theory, Hyperbolic Dynamics and Dimension Theory. Universitext
(Springer, Berlin, 2012)
9. L. Barreira, Ya. Pesin, Nonuniform Hyperbolicity: Dynamics of Systems with Nonzero Lya-
punov Exponents, Encyclopedia of Mathematics and its Applications, vol. 115 (Cambridge
University Press, Cambridge, 2007)
10. L. Barreira, C. Valls, Ordinary Differential Equations: Qualitative Theory. Graduate Studies
in Mathematics, vol. 137 (American Mathematical Society, RI, 2012)
11. L. Barreira, C. Valls, Dynamical Systems: An Introduction. Universitext (Springer, London,
2013)
12. A. Beardon, The Geometry of Discrete Groups. Graduate Texts in Mathematics, vol. 91
(Springer, NY, 1983)
13. A. Beardon, Iteration of Rational Functions: Complex Analytic Dynamical Systems. Graduate
Texts in Mathematics, vol. 132 (Springer, Berlin, 1991)
14. C. Bonatti, L. Dı́az, M. Viana, Dynamics Beyond Uniform Hyperbolicity: A Global Geometric
and Probabilistic Perspective. Encyclopaedia of Mathematical Sciences, vol. 102 (Springer,
Berlin, 2005)
15. R. Bowen, Equilibrium States and Ergodic Theory of Anosov Diffeomorphisms. Lecture Notes
in Mathematics, vol. 470 (Springer, Berlin, 1975)
16. M. Brin, G. Stuck, Introduction to Dynamical Systems (Cambridge University Press, Cam-
bridge, 2002)
17. J. Brown, Ergodic Theory and Topological Dynamics. Pure and Applied Mathematics, vol.
70 (Academic Press, NY, 1976)
18. L. Carleson, T. Gamelin, Complex Dynamics. Universitext (Springer, NY, 1993)
19. N. Chernov, R. Markarian, Chaotic Billiards. Mathematical Surveys and Monographs, vol.
127 (American Mathematical Society, RI, 2006)

in Mathematics, https://doi.org/10.1007/978-3-030-15915-3
218 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
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
diffeomorphism, 24 Splitting, 35, 36

homeomorphism, 23, 24 Stable
Orientation-reversing homeomorphism, 23 manifold, 39
space, 35, 36, 40
P Strip
Partition horizontal, 36
of numbers, 24 vertical, 36
of sets, 59 Symbolic dynamics, 49, 169
Period, 4
Periodic point, 4, 24, 25 T
Poincaré TA , 5
map, 6 Tn , 4
recurrence theorem, 59 Theorem
section, 6 Birkhoff, 59
Poincaré–Bendixson theorem, 25 Denjoy, 24
Point Grobman–Hartman, 38
(forward) recurrent, 16 Poincaré, 59
critical, 5 Poincaré–Bendixson, 25
fixed, 4 Sharkovsky, 25
nonwandering, 16 Topological
periodic, 4, 24, 25 conjugacy, 17, 24, 38
Positive semiorbit, 4, 5, 25 dynamical system, 15
R dynamics, 15, 95
Rp f q, 16 entropy, 17, 18, 49–51
Rα , 4 flow, 15
Recurrent point, 16 Markov chain, 50
Refinement, 67, 210 semiflow, 15
Rotation, 24 Topologically
circle, 4 mixing, 17, 50
number, 24 transitive, 17, 50
Torus, 4
S automorphism, 5
S1 , 4, 23 endomorphism, 5
Section, 6 Transition matrix, 50
Semiflow, 5, 6 Transitive matrix, 50
topological, 15
Semiorbit U
negative, 4, 6 Unit tangent bundle, 40
positive, 4, 5, 25 Unstable
Set manifold, 39
backward invariant, 3 space, 35, 36, 40
forward invariant, 3 Upper half plane, 40
hyperbolic, 35–37, 39
invariant, 3, 5 V
minimal, 19, 101 V s pxq, 38
Sharkovsky’s V u pxq, 38
ordering, 24 Var ϕ , 24
theorem, 25 Vertical strip, 36
Shift map, 49, 50
Simple function, 58 W
Smale horseshoe, 36 Weakly mixing measure, 66, 208
Space
stable, 35, 36, 40 Z
unstable, 35, 36, 40 Zeta function, 51

2-Luís Barreira, Claudia Valls - Dynamical Systems by Example-Springer (2019)

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2-Luís Barreira, Claudia Valls - Dynamical Systems by Example-Springer (2019)

Uploaded by

Copyright:

Available Formats

Problem Books in Mathematics

ISSN 0941-3502 ISSN 2197-8506 (electronic)

Mathematics Subject Classiﬁcation (2010): 37-01, 37Dxx, 37Bxx, 37Exx, 37Axx

© Springer Nature Switzerland AG 2019

solutions, on the basic theory, topological dynamics, low-dimensional dynamics,

Lisbon, Portugal Luís Barreira

Part I Theory and Problems

Part II Problems and Solutions

Notions and Results

Definition 1.2 Given a map f : X Ñ X, a set A Ď X is said to be:

© Springer Nature Switzerland AG 2019 3

Definition 1.3 Given a map f : X Ñ X and a point x P X, the set

is called the positive semiorbit of x. Moreover, when f is invertible, the sets

is called a q-periodic point of f . Moreover, x P X is called a periodic point if it is a

Moreover, given an integer m ą 1, the expanding map Em : S1 Ñ S1 is defined by

Definition 1.6 Given n P N, let „ be the equivalence relation on Rn given by

Given an n ˆ n matrix A with entries in Z, the endomorphism of the torus induced

TA rxs “ rAxs for rxs P Tn .

When TA is invertible, it is also called the automorphism of the torus induced by A.

is called a flow on X. A family Φ of maps is called a dynamical system with contin-

Moreover, given a flow Φ on X, a set A Ď X is said to be Φ -invariant if

Definition 1.10 Given a semiflow Φ on X and a point x P X, the set

γ ` pxq “ tϕt pxq : t ě 0u

is called the positive semiorbit of x. Moreover, for a flow Φ on X, the sets

γ ´ pxq “ tϕ´t pxq : t ě 0u

Fig. I.1.1 The first return time τ pxq.

f p1q “ 3, f p2q “ 5, f p3q “ 4, f p4q “ 2 and f p5q “ 1

Fig. I.1.2 The map f in Problem 1.6.

Problem 1.7 Let f : I Ñ I be a strictly increasing map on an interval I Ď R. Show

Problem 1.8 Let f : I Ñ I be a strictly decreasing map on an interval I Ď R. Show

Problem 1.19 Show that given x P S1 and δ ą 0, there exist y P px ´ δ , x ` δ q Ď S1

Problem 1.21 Consider the map f : R Ñ R defined by f pzq “ z3 on the set

Show that the set of periodic points of f is dense in R.

Fig. I.1.3 The expanding map E2 .

Fig. I.1.4 The map E22 “ E4 .

Problem 1.22 Consider the map f : T2 Ñ T2 defined by

f px, yq “ pEm pxq, Em pyqq.

Show that the set of periodic points of f is dense in T2 .

card TA´1 x “ |det A| for all x P Tn ,

card f ´1 x “ q for all x P S1 .

(The number q coincides with the degree deg f of f .)

on R2 . Show that for each set I Ď R` 0 the union

is invariant under the flow determined by the differential equation.

on R2 for some a ą 0. Show that the solutions starting in the sets

remain in these sets for all time.

Problem 1.33 Find the flow determined by the equation x2 ` x “ 0.

Show that the set (

hpϕτ px,tq pxqq “ ψt phpxqq for x,t ą 0.

Fig. I.1.6 Phase portrait of the equation x1 “ ´x.

Fig. I.1.7 Phase portrait of the equation x1 “ ´x2 .

Problem 1.38 Let f , g : R2 Ñ R be C1 functions with f ą 0 such that

f px ` k, y ` lq “ f px, yq and gpx ` k, y ` lq “ gpx, yq

for all x, y P R and k, l P Z. Then the differential equation

x1 “ f px, yq, y1 “ gpx, yq

ϕt p0, zq “ pxptq, yptqq “ pxpt, zq, ypt, zqq

(see Figure I.1.8) is invertible.

Fig. I.1.8 The map h in Problem 1.38.

Problem 1.39 Consider the differential equation px1 , y1 q “ pα , β q on T2 “ R2 {Z2

2. has at least one point in each of the sets

3. intersects Aϕ transversally at each point in the sets

Notions and Results

We first recall some notions and results from topological dynamics.

Moreover, when f is invertible, the α -limit set of a point x P X is defined by

© Springer Nature Switzerland AG 2019 15