On the Dynamics of Torus

Homeomorphisms
Andres Koropecki

Instituto de Matematica Pura e Aplicada - IMPA

March, 2007

Acknowledgements
Thanks (orderless):
to my advisor Enrique Pujals, for the presence, encouragement, and good
advice.
to Alejandro Kocsard, not only for the collaboration that resulted in
most of this work, but also for the enthusiasm and friendship during these
years.
to Meysam Nassiri for many useful discussions, for his patience and
friendship, and for many corrections to this text.
to Prof. Sebastiao Firmo, for many corrections and useful suggestions.
to Profs. Jacob Palis and Marcelo Viana for the constant support.
to Lucien Guillou for the hospitality during our stay in Grenoble, and to
Francoise Beguin, Sylvain Crovisier, Bassam Fayad, and Patrice Le Calvez
for the insight and useful discussions.
to CNPq for the financial support during my doctoral studies.
to IMPA for the great environment and for making this possible.
to my colleagues Martin Andersson, Jimmy Santamaria Torrez, and
Paulo Varandas.
to all the good friends I made at IMPA.
to Eduardo Tabacman for introducing me to Dynamical Systems, and
to Liliana Forzani for the support and encouragement.
to the IMAL, in particular to Eleonor Harboure and Hugo Aimar.
to my friend Ignacio Garca.
to my family.
to Mercedes.

Contents
Introduction

0 Background

0.1

The rotation set

. . . . . . . . . . . . . . . . . . . . . . . . .

The rotation set and periodic orbits . . . . . . . . . .

10

Curves, lines . . . . . . . . . . . . . . . . . . . . . . . . . . .

11

0.2.1

Ordering lines . . . . . . . . . . . . . . . . . . . . . . .

12

0.2.2

Brouwer lines . . . . . . . . . . . . . . . . . . . . . . .

12

More about the rotation set . . . . . . . . . . . . . . . . . . .

13

0.3.1

The rotation set and Brouwer lines . . . . . . . . . . .

13

0.3.2

The rotation set and free curves . . . . . . . . . . . .

14

0.3.3

The rotation set and conjugations . . . . . . . . . . .

14

0.1.1
0.2

0.3

1 Realizing periodic orbits

16

1.1

Preliminary results . . . . . . . . . . . . . . . . . . . . . . . .

16

1.2

Proof of Theorem 1.1 . . . . . . . . . . . . . . . . . . . . . . .

17

1.3

Proof of Theorem A . . . . . . . . . . . . . . . . . . . . . . .

17

1.4

The wedge . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19

1.5

Proof of Lemma 1.2 . . . . . . . . . . . . . . . . . . . . . . .

20

1.6

Proof of Lemma 1.3 . . . . . . . . . . . . . . . . . . . . . . .

21

1.7

Proof of Lemma 1.4 . . . . . . . . . . . . . . . . . . . . . . .

24

2 Free curves and fixed points

2.1

25

Continued fractions . . . . . . . . . . . . . . . . . . . . . . . .

26

2.2

Arithmetic lemmas . . . . . . . . . . . . . . . . . . . . . . . .

27

2.3

Proof of Theorem 2.1 . . . . . . . . . . . . . . . . . . . . . . .

31

2.3.1

The case n = 1 . . . . . . . . . . . . . . . . . . . . . .

31

2.3.2

The case n > 1 . . . . . . . . . . . . . . . . . . . . . .

34

Proof of Theorem B . . . . . . . . . . . . . . . . . . . . . . .

35

2.4

3 Always free curves

3.1

37

Proof of theorem C . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1 Construction of an invariant partition . . . . . . . . .

38
40

Defining order in P . . . . . . . . . . . . . . . . . . . .

43

3.1.2
3.1.3

Construction of the semi-conjugation . . . . . . . . . .

4 Invariant foliations and minimal diffeomorphisms

45
46

4.1

Dynamic cocycles . . . . . . . . . . . . . . . . . . . . . . . . .

47

4.2

Generic minimality of the dynamic cocycle . . . . . . . . . . .

48

4.3

Generic non-existence of invariant foliations . . . . . . . . . .

53

4.4

Generic non-existence of always-free curves . . . . . . . . . .

53

4.5

C0

54

foliations with non-differentiable leaves . . . . . . . . . . .

Bibliography

60

ii

Notation
As usual, we denote the two-torus R2 /Z2 by T2 , with quotient projection
: (x, y) 7 (x, y) + Z2 . The integer translations are
T1 : (x, y) 7 (x + 1, y) and T2 : (x, y) 7 (x, y + 1),
and pri : R2 R, for i = 1, 2 are the projections onto the first and second

coordinate, respectively.

By Homeo(X) we mean the set of homeomorphisms of X to itself, and

by Homeo (X) the set of elements of Homeo(X) which are isotopic to the
identity. We remark that a torus homeomorphism is isotopic to the identity
if and only if it is homotopic to the identity [Eps66].
Given F Homeo(T2 ), a lift of F to R2 is a map f Homeo(R2 ) such

that f = F . An homeomorphism f : R2 R2 is a lift of an element of

Homeo (T2 ) if and only if f commutes with T1 and T2 . Any two lifts of a
given homeomorphism of T2 always differ by an integer translation. We will
usually denote maps of T2 to itself by uppercase letters, and lifts of such
maps to R2 by their corresponding lowercase letters.
By Z2cp we denote the set of pairs of integers (m, n) such that m and
n are coprime. We will say that (x, y) R2 is an integer point if both x
and y are integer, and a rational point if both x and y are rational. When

we write a rational number as p/q, we assume that it is in reduced form,

i.e. that p and q are coprime, except when we are talking about a rational
point (p1 /q, p2 /q), in which case we assume that p1 , p2 and q are mutually
coprime (i.e. lcd{p1 , p2 , q} = 1).
The set A will denote the open annulus R S 1 ' R2 /hT2 i. The map

T1 induces a map : A 7 A, defined by (x, v) 7 (x + 1, v). We denote

by 1 : A T2 the covering map (x, v) 7 (x + Z, v), which is the same as

the quotient projection A 7 A/h i. Similarly, A denotes the closed annulus

[0, 1] S 1 .

Introduction
The rotation set is one of the most important tools for the study of homeomorphisms of the torus in the homotopy class of the identity1 . Although
not as easily as in the case of circle homeomorphisms, one can obtain useful
a priori information about the dynamics by studying the rotation set. A
central question in this direction is under what conditions a point of rational
coordinates in the rotation set is realized as the rotation vector of a periodic
orbit. It is clear that periodic orbits have rotation vectors with rational
coordinates. Franks proved in [Fra88] that extremal points of the rotation
set are always realized by periodic orbits; but it is generally not true that
every rational point in the rotation set is realized (c.f. Example 0.1). In
this aspect, the case that is best understood is when the rotation set has
non-empty interior. For such homeomorphisms, a theorem by Llibre and
Mackay [LM91] guarantees positive topological entropy. Moreover, Franks
[Fra89] proved that every rational point in the interior of the rotation set is
realized as the rotation vector of a periodic orbit, and this is optimal in the
sense that there are examples (even area-preserving ones) where the rotation
set has nonempty interior and many rational points on the boundary, but
the only ones that are realized by periodic orbits are extremal or interior
rational points (see [MZ91, 3]).

On the other hand, when the rotation set has empty interior the situation

is more delicate. Since the rotation set is compact and convex, if it has no
interior it must be a line segment or a single point. The following simple
1

In this work, unless stated otherwise, all homeomorphisms are assumed to be in the
homotopy class of the identity. Note that by [Eps66], this is equivalent to being in the
isotopy class of the identity.

example shows that one can have homeomorphisms with the rotation set
being a segment with many rational points, but which have no periodic
orbits at all.
Example 0.1. Let P : S 1 S 1 be the north pole-south pole map on S 1 ,

e.g. the map lifted by x 7 x + 0.1 sin(2x). Define f : R2 R2 by f (x, y) =

(P (x), y + r sin(2x)) for some small irrational r. Then f is the lift of a

map F : T2 T2 isotopic to the identity, and (f ) = {0} [r, r], which
contains many rational points; but F has no periodic points.

This example shows that some additional hypothesis is required to guarantee that rational points are realized by periodic orbits. The following
theorem gives a sufficient condition.
Theorem 0.2 (Franks, [Fra95]). If an area-preserving homeomorphism of
the torus has a rotation set with empty interior, then every rational point in
its rotation set is realized as the rotation vector of a periodic orbit.
In the same article, Franks asks whether the area-preserving hypothesis
is really necessary for the conclusion of the theorem. It is natural to expect
that a weaker, more topological hypothesis should suffice to obtain the same
result. This topological substitute for the area-preserving hypothesis turns
out to be, to some extent, the curve intersection property. An essential
simple closed curve is free for F if F () = . We say that F has the

curve intersection property if F has no free curves. In Chapter 1 we prove

the following:

Theorem A. If a homeomorphism of the torus satisfying the curve intersection property has a rotation set with empty interior, then every rational
point in its rotation set is realized by a periodic orbit.
Our proof is essentially different of that of Theorem 0.2, since the latter
relies strongly on chain-recurrence properties that are guaranteed by the
area preserving hypothesis but not by the curve intersection property.
Variations of the curve intersection property are already present in some
fixed point theorems. An interesting case is a generalization of the following
classic theorem:
3

Poincar
e-Birkhoff Theorem (Birkhoff, [Bir25]). Let F : A A be an
area-preserving homeomorphism of the closed annulus, verifying the boundary twist condition. Then F has at least two fixed points.
The boundary twist condition means that F preserves boundary components and there exists a lift f : [0, 1] R [0, 1] R of F such that the

rotation numbers of f |{0}R and f |{1}R have opposite signs.

Birkhoff and Kerekj

art
o already noted that the area preserving hypoth-

esis was not really necessary, and that it could be replaced by a more topological one, and they obtained the following
Theorem ([Bir25], [Ker29]). If F : A A is a homeomorphism satisfying

the boundary twist condition and such that F () 6= for each essential

simple closed curve , then F has at least one fixed point in the interior of
the annulus.
An even more topological version of this theorem was proven by Guillou,
who substituted the twist condition by the property that every simple arc
joining the boundary components intersects its image by F :
Theorem (Guillou, [Gui94]). If F : A A is an orientation-preserving

homeomorphism such that every essential simple closed curve or simple arc
joining boundary components intersects its image by F , then F has a fixed
point.
The hypotheses of the above theorem can be regarded as the curve intersection property in the setting of the closed annulus: they say that every
interesting curve intersects its image by F . This led Guillou to ask if a
similar result holds for the torus:
Let F : T2 T2 be a homeomorphism isotopic to the identity.

Does the curve intersection property imply the existence of a fixed

point for F ?

The answer is no, as an example by Bestvina and Handel shows [BH92].

Their example relies in the fact that the existence of a free curve imposes a
4

restriction on the size of the rotation set (see Lemma 1.4). They construct
a homeomorphism such that any strip bounded by straight lines and containing the rotation set, also contains a point of integer coordinates. By the
previously mentioned lemma, and some simple considerations, this homeomorphism cannot have any free curves; but by their construction, there are
no points of integer coordinates inside the rotation set, which implies that
the homeomorphism has no fixed points.
Nevertheless, the existing examples have a rotation set with nonempty
interior, which implies that they have infinitely many periodic points of arbitrarily high periods, and positive topological entropy ([LM91], [Fra89]). The
question that arises is whether the presence of this kind of rich dynamics
is the only new obstruction to the existence of free curves in T2 . In other
words, is the answer to Guillous question affirmative if the rotation set has
empty interior? This leads to our next result, which is proved in Chapter
2:
Theorem B. Let F : T2 T2 be a homeomorphism satisfying the curve

intersection property. Then either F has a fixed point, or its rotation set
has nonempty interior.

Thus, if F has the curve intersection property, then either F has periodic
orbits of arbitrarily high periods or it has a fixed point. In the latter case,
one might expect the existence of a second fixed point (as in the case of the
annulus) but no more than that. Figure 1 shows that one cannot expect more
than two fixed points in the annulus; the time-one map of the flow sketched
there is a homeomorphism with the curve intersection property, which has
two fixed points and no other periodic points. Gluing a symmetric copy of
this homeomorphism through the boundaries of the annulus, one obtains an
example with the same properties in T2 .
In Chapter 3 we consider a situation opposite to the curve intersection
property: what can be said if there is a simple closed curve that is always
free for F (i.e. such that all iterates of the curve by F are pairwise disjoint)? In general we cannot say much, since the existence of an attractor
wrapping around the torus (e.g. an attracting homotopically non-trivial
5

Figure 1: A curve-intersecting flow with no periodic points of period > 1

closed curve) usually implies the existence of such a curve; and this is a local
property, meaning that we cannot say anything about the dynamics outside
a neighborhood of the attractor. On the other hand, if we assume that this
curve has a dense orbit (which is true, for example, if F is transitive), we
obtain the following:
Theorem C. Let F be a homeomorphism of T2 , and suppose there exists
a simple closed curve that is disjoint from all its iterates by F , whose orbit
is dense in T2 . Then F is semi-conjugate to an irrational rotation of the
circle.
Moreover, the proof of Theorem C can easily be generalized to homeomorphisms possessing an always free continuum with a dense orbit, provided
that the continuum wraps around the torus and is disjoint from some essential simple closed curve (see the precise statement in Chapter 4, Theorem
C)
The above result is related to a question arising from an article of Fathi
and Herman [FH77]. In that classical paper, they prove the existence of
minimal and uniquely ergodic diffeomorphisms in manifolds admitting a
C free action of S 1 . Their techniques are based on the fast approximation
by conjugations method, introduced by Anosov and Katok in [AK70], combining it with Baires theorem in an appropriate space (for a comprehensive
6

overview of these methods and their applications, see [FK04]). In the set

ting of T2 , this space corresponds to O (T2 ), which is the C -closure of

diffeomorphisms which are C -conjugate to rigid translations of T2 . The

following is a quotation from [FH77]:
Le second auteur montrera ailleurs que la propriete suivante:
f nadmet pas de feuilletage F (de T2 ) C 0 de codimension

1 invariant (i.e. f envoide chaque feuille de F sur une feuille

de F pas necessairement la meme)

est vraie sur un ensemble qui contient un G dense de O (T2 ).

The claim says that generic diffeomorphisms in O (T2 ) have no invariant

C 0 foliation of codimension 1 (i.e. this property holds in a residual set).

However, the proof of this fact never appeared in the subsequent publications
of Herman.
In Chapter 4, we prove some partial results regarding the above claim.
Combining Theorem C and some results of Herman, we obtain a quick proof

that generic diffeomorphisms in O (T2 ) have no invariant C 0 foliation (of

codimension 1) with a compact leaf. In fact, we have something stronger:

generic diffeomorphisms in O (T2 ) have no always free curves (4.4).

In addition, we prove that generically in O (T2 ) there are no invariant

C 0 foliations induced by a C 0 line field (4.3). This is obtained as a corollary

of the following more general result (4.2):

Theorem D. For a generic diffeomorphism f O (T2 ), its associated

dynamic cocycle

(x, v) 7

Dfx v
f (x),
kDfx vk

is minimal.

A similar result holds if we restrict our attention to Om (T2 ), the closure

of the set of area-preserving diffeomorphisms which are C -conjugated to a

rigid translation. It is worth mentioning that, while one can easily construct
SL(2, R)-cocycles over minimal diffeomorphisms of T2 such that their normal
action on T2 S 1 is minimal, it is not obvious that one can obtain dynamic
7

cocycles with that property. However, Theorem D provides a wide family

of such examples.
Finally, in 4.5 we also show, using the fast approximation method, that

there is a dense subset of O (T2 ) consisting of minimal diffeomorphisms

which have no invariant C 0 foliation of codimension 1 with C 1 leaves.

To give a complete proof of Hermans claim, it remains to show that the

subset of O (T2 ) consisting of diffeomorphisms which have a C 0 foliation

which is conjugated to the foliation induced by a minimal translation of T2 ,

is a meager set in O (T2 ). We are not able to prove this at this time. In
fact, even the following question remains open (to our knowledge):
Question 0.3. Is there a minimal homeomorphism of T2 possessing no
invariant C 0 foliation?
The results presented in this work, with the exception of Theorem C,
were obtained jointly with Alejandro Kocsard.

Chapter 0

Background
0.1

The rotation set

Throughout this section we assume that F Homeo (T2 ), and f : R2 R2

is a lift of F .

Definition 0.4 (Misiurewicz & Ziemian, [MZ89]). The rotation set of f is

defined as
(f ) =

m=1

!
 n
[
f (x) x
2
:xR
R2
cl
n
n=m

The rotation set of a point x R2 is defined by

(f, x) =

cl

m=1


f n (x) x
:n>m .
n

If the above set consists of a single point v, we say that v is the rotation
vector of x.
Remark 0.5. (f ) is just the set of all limits of convergent sequences of the
form

f nk (xk ) xk
nk

where xk R2 and nk .
9

Proposition 0.6. For all integers n, m1 , m2 ,

(T1m1 T2m2 f n ) = n (f ) + (m1 , m2 ).
Remark 0.7. In particular, the rotation set of any other lift of F is an integer
translate of (f ), and we can talk about the rotation set of F if we keep
in mind that it is defined modulo integer translations.
Theorem 0.8 ([MZ89]). The rotation set is compact and convex, and every
extremal point of (f ) is the rotation vector of some point.
By extremal point in the theorem above, we mean extremal in the usual
sense for convex sets, i.e. a point that is not in the interior of a line segment
contained in (f ).

0.1.1

The rotation set and periodic orbits

For a homeomorphism of the circle, if the rotation number is a rational p/q,

then there exists a periodic orbit of period q. This motivates the following
question in the two-dimensional case.
Question 0.9. If (p1 /q, p2 /q) (f ), when can we find a point x R2 such
that f q (x) = x + (p1 , p2 )?

Whenever f q (x) = x + (p1 , p2 ), with p1 , p2 and q mutually coprime, we

will say that (p1 /q, p2 /q) is realized as the rotation vector of a periodic orbit
of F . This is because (x) is a periodic orbit of F of period q, and its
rotation vector is (p1 /q, p2 /q) (modulo Z2 ). So the question is when can we
realize rational points of the rotation set by periodic orbits of F .
There are several results in this direction (including the already mentioned Theorem 0.2)
Theorem 0.10 (Franks, [Fra88]). If a rational point of (f ) is extremal,
then it is realized by a periodic orbit.
Theorem 0.11 (Franks, [Fra89]). If a rational point is in the interior of
(f ), then it is realized by a periodic orbit.
10

Theorem 0.12 (Jonker & Zhang, [JZ98]). If (f ) is a segment with irrational slope, and it contains a point of rational coordinates, then this point
is realized by a periodic orbit.

0.2

Curves, lines

We denote by I the interval [0, 1]. A curve on a manifold M is a continuous

map : I M . As usual, we represent by both the map and its image,

as it should be clear from the context which is the case.

We say that the curve is closed if (0) = (1), and simple if the
restriction of the map to the interior of I is injective. If is a closed
curve, we say it is essential if it is homotopically non-trivial.
Definition 0.13. A curve M is free for F if F () = . We say that
F has the curve intersection property (CIP) if there are no free essential

simple closed curves for F .

Remark 0.14. For convenience, from now on by a free curve for F we will
usually mean an essential simple closed curve that is free for F , unless stated
otherwise.
By a line we mean a proper topological embedding ` : R R2 . Again,

we use ` to represent both the function and its image.

Definition 0.15. Given (p, q) Z2cp , a (p, q)-line in R2 is a line ` that is

invariant by T1p T2q , such that `(s) = T1p T2q `(t) implies s > t, and such that

its projection to T2 by is a simple closed curve. A (p, q)-curve in T2 is

the projection by of a (p, q)-line. We will say that a simple closed curve is
vertical if it is either a (0, 1)-curve or a (0, 1)-curve. Similarly, a line will
be called vertical if it is a (0, 1)-line or a (0, 1)-line.

Remark 0.16. If ` is a line in R2 that is invariant by T1p T2q , then (`) is a

always a closed curve in T2 . We are requiring that this curve be simple to
call ` a (p, q)-line. Conversely, if is an essential simple closed curve in T2 ,
taking a lift : I R2 , we have that (p, q) = (1) (0) is an integer point
independent of the choice of the lift. The curve can be extended naturally
11

to R by (t + n) = (t) + n(p, q), if n Z and t [0, 1]; in this way we

obtain a (p, q)-line that projects to . It is not hard to see that p and q

must be coprime, from the fact that is simple and essential. We will say
that a (p, q)-line ` is a lift of (`).
Remark 0.17. Two disjoint essential simple closed curves must be either
both (p, q)-curves for some (p, q) Z2cp , or one a (p, q) curve and the other

a (p, q)-curve. Note that the difference between these two is just orientation; if we only regard the curves as sets, then (p, q)-curves and (p, q)curves are the same thing.

Remark 0.18. It is easy to see that any (p, q)-line is contained in a strip
bounded by two straight lines of slope q/p. In particular, if ` is a vertical
line, there is M > 0 such that pr1 (`) [M, M ].

0.2.1

Ordering lines

A line ` in R2 can be seen as a Jordan curve through in S 2 = R2 {}.

Thus, R2 \` has exactly two connected components, both unbounded. Using

the orientation of `, we may define the left and the right components, which
we denote by L` and R`. We also denote by L` and R` their respective
closures, which correspond to L` ` and R` `.

There is a natural partial ordering between lines in R2 , defined by `1 < `2

if `1 L`2 (and `2 R`1 ). With an abuse of notation we will write `1 `2

when `1 L`2 , which means that the lines may intersect but only from one
side. Naturally, if two lines `1 and `2 do not intersect, then either `1 < `2

or `2 < `1 . If `1 < `2 , we denote by S(`1 , `2 ) the strip L`2 R`1 , and by

S(`1 , `2 ) its closure.

Remark 0.19. If f Homeo(R2 ) preserves orientation, then f preserves

order: if `1 < `2 , then f (`1 ) < f (`2 ).

0.2.2

Brouwer lines

Definition 0.20. A Brouwer line for h Homeo(R2 ) is a line ` in R2 such

that h(`) > `.

12

The classic Brouwer Translation Theorem guarantees the existence of a

Brouwer line through any point of R2 for any fixed-point free, orientationpreserving homeomorphism (see [Bro12, Ker29, Fat87, Fra92, Gui94]; also
see [LC05] for a very useful equivariant version). The following will be much
more useful for our purposes:
Theorem 0.21 (Guillou, [Gui06]). Let f be a lift of F Homeo (T2 ), and
suppose f has no fixed points. Then f has a Brouwer (p, q)-line, for some
(p, q) Z2cp .

0.3
0.3.1

More about the rotation set

The rotation set and Brouwer lines

The following lemma is particularly useful when there is a Brouwer (p, q)line:
Lemma 0.22. Let S be a closed semiplane determined by a straight line
containing the origin, and for y R2 denote by Sy = {w + y : w S} its
translate by y. Suppose that x R2 is such that for some y,
f n (x) Sy for all n > 0.
Then (f, x) S. Moreover, if for all x R2 there is y such that the above

holds, then (f ) S.

Proof. Let : R2 R be a linear functional such that

Sy = {w R2 : (w) (y)}.
Given x such that f n (x) Sy for all n > 0, we have then

f n (x) x
n

(y) (x)
(f n (x)) (x)

0.
n
n

Thus, if z is the limit of the sequence (f ni (x) x)/ni , then (z) 0. This

implies that (f, x) S. The other claim follows from Theorem 0.8.
13

Remark 0.23. If f has a Brouwer (p, q)-line, the above lemma and Remark
0.18 imply that (f ) is contained in one of the closed semiplanes determined
by the straight line of slope q/p through the origin.

0.3.2

The rotation set and free curves

Besides the existence of periodic orbits, other practical dynamical information that can be obtained from the rotation set is the existence of free
curves. Recall that an interval with rational endpoints [p/q, p0 /q 0 ] is a Farey
interval if qp0 pq 0 = 1. The following result was proved by Kwapisz for dif-

feomorphisms, and by Beguin, Crovisier, LeRoux and Patou in the general

case.

Theorem 0.24 ([Kwa02], [BCLP04]). Suppose there exists a Farey interval

[p/q, p0 /q 0 ] such that
pr1 (f )

p p0
,
q q0

Then there exists a simple closed (0, 1)-curve in T2 such that all the curves
0

, F (), F 2 (), . . ., F q+q 1 () are mutually disjoint. In particular, if

pr1 (f ) Z = , then F has a free (0, 1)-curve.

0.3.3

The rotation set and conjugations

Given A GL(2, Z), we denote by A the homeomorphism of T2 lifted by it.

If H Homeo(T2 ), there is a unique A GL(2, Z) such that for every lift h

of H, the map h A is bounded (in fact, Z2 -periodic). From this it follows

and h1 A1 is bounded. In fact, H 7 A induces
that H is isotopic to A,

an isomorphism of the isotopy group of T2 to GL(2, Z).

Lemma 0.25. If H, F Homeo(T2 ) with H isotopic to A GL(2, Z) and

F isotopic to the identity, and h, f Homeo(R2 ) are their respective lifts,

then (hf h1 ) = A (f ). In particular, (Af A1 ) = A (f ).

Proof. We can write ((hf h1 )n (x) x)/n as
(h A)(f n h1 (x))
+A
n

f n (h1 (x)) h1 (x)

n
14

+A

(h1 A1 )(x)
n

and using the fact that h A and h1 A1 are bounded, we see that the
leftmost and rightmost terms of the above expression vanish when n .
Thus if nk and xk R2 , we have
(hf h1 )nk (xk ) xk
lim
=A
k
nk

f nk (h1 (xk )) h1 (xk )

lim
k
nk

whenever the limits exist. Since h is a homeomorphism, it follows from the

definition that (hf h1 ) = A (f ).
We will use the above lemma extensively: when trying to prove some
property that is invariant by topological conjugation (like the existence of a
free curve or a periodic point for F ), it allows us to consider just the case
where the rotation set is the image of (f ) by some convenient element of
GL(2, Z).
Remark 0.26. A particular case that will often appear is when (f ) is a
segment of rational slope. In that case, there exists a map A GL(2, Z)

such that A (f ) is a vertical segment. Indeed, if (f ) is a segment of slope

p/q, then given x, y Z such that px + qy = 1, and letting
A=

p q

we have that det(A) = 1, and A(q, p) = (0, 1), so that A (f ) is vertical.

Note that the above A also maps (q, p)-curves to (0, 1)-curves.

15

Chapter 1

Realizing periodic orbits

In this chapter we present some results regarding the curve intersection
property and its relation with the rotation set, which will allow us to prove
Theorem A. Throughout this chapter, F : T2 T2 will be a homeomorphism
isotopic to the identity and f : R2 R2 a lift of F .

1.1

Preliminary results

The following result will be essential in the proof of Theorem A.

Theorem 1.1. Suppose F n has a free (p, q)-curve for some n 1. Then F

has a free (p, q)-curve.

We will also use the following lemmas, the proofs of which are postponed
to the end of this chapter.
Lemma 1.2. Suppose f n has a Brouwer (0, 1)-line, for some n N. Then

f has a Brouwer (0, 1)-line.

Lemma 1.3. Suppose that some lift f of F has a Brouwer (0, 1)-line. Then,
either F has a free (0, 1)-curve, or max(pr1 (f )) 1.
The next lemma is essentially Lemma 3 of [BH92]; we include it here for
the sake of completeness.

16

Lemma 1.4. Suppose F has a free (0, 1)-curve. Then for any lift f of F
there is k Z such that pr1 (f ) [k, k + 1].

1.2

Proof of Theorem 1.1

Conjugating the involved maps by an element of GL(2, Z), we may assume

that (p, q) = (0, 1) (see Lemma 0.25 and the remark below it).
Suppose F n has a free (0, 1)-curve for some n 2. Any lift to R2 of this

curve is a vertical Brouwer line for f n , so Lemma 1.2 implies that there is a
vertical Brouwer line for f . This holds for any lift f of F .
If pr1 (f ) Z = , then by Theorem 0.24 there is a free (0, 1)-curve for

F , and we are done.

Otherwise, let k pr1 (f ) Z, and consider the lift f0 = T1k f of F .

By Proposition 0.6, it is clear that 0 pr1 (f0 ). On the other hand, as we

already saw, f0 has a vertical Brouwer line `.

Assume ` is a Brouwer (0, 1)-line. Then by Lemma 1.3, either F has a

free (0, 1)-curve, or max(pr1 (f0 )) 1. In the latter case, since we know

that 0 (f0 ), it follows from connectedness that pr1 (f0 ) [0, 1]. But this

implies that

pr1 (f0n ) [0, n] [0, 2],

which contradicts Lemma 1.4 (since F n has a free (0, 1)-curve). Thus the
only possibility is that F has a free (0, 1)-curve.
If ` is a Brouwer (0, 1)-line, then using the previous argument with

f01

instead of f0 , we see that F 1 (and thus F ) has a free (0, 1)-curve

; and inverting the orientation of we get a free (0, 1)-curve for F . This
completes the proof.

1.3

Proof of Theorem A

Suppose F Homeo (T2 ) has the curve intersection property and (f ) has
empty interior, where f is a lift of F . We have three cases.

17

(f ) is a single point
In this case, the unique point of (f ) is extremal; and if it is rational,
Theorem 0.10 implies that it is realized by a periodic orbit of F .
(f ) is a segment of irrational slope
In this case (f ) contains at most one rational point and, by Theorem 0.12,
this point is realized by a periodic orbit.
Remark 1.5. Theorem 0.21 provides a simple way of proving this as well. In
fact, it suffices to consider the case where the unique rational point in (f )
is the origin, and to show that in this case f has a fixed point. If the origin
is an extremal point, this follows from Theorem 0.10. If the origin is strictly
inside the rotation set, then there is only one straight line through the origin
such that (f ) is contained in one of the closed semiplanes determined by
the line. This unique line is the one with the same slope as (f ), which
is irrational. If f has no fixed points, then Theorem 0.21 implies that f
has a Brouwer (p, q)-line for some (p, q) Z2cp ; but then our previous claim

contradicts Lemma 0.22.

(f ) is a segment of rational slope

Fix a rational point (p1 /q, p2 /q) (f ). Recall that this point is realized as

the rotation vector of a periodic orbit of F if and only if g = T1p1 T2p2 f q

has a fixed point. Note that (0, 0) (g), and g is a lift of F q . Moreover,
(g) = T1p1 T2p2 (q (f )),

which is a segment of rational slope containing the origin. Conjugating all

the involved maps by an element of GL(2, Z), we may assume that (g) is a
vertical segment containing the origin.
We will show by contradiction that g has a fixed point. Suppose this is
not the case. Then, by Theorem 0.21, g has a Brouwer (p, q)-line `, for some
(p, q) Z2cp . Moreover, (0, 0) must be strictly inside (g) (i.e. it cannot be

18

extremal, since otherwise g would have a fixed point by Theorem 0.10), and
by Remark 0.23 this implies that ` is a vertical Brouwer line.
Assume ` is a (0, 1)-line (if it is a (0, 1)-line, we may consider g1

instead of g and use a similar argument). Since diam(pr1 (g)) = 0, Lemma

1.3 implies that F q , the map lifted by g, has a free (0, 1)-curve; but then
by Theorem 1.1, F has a free curve, contradicting the curve intersection
property. This concludes the proof.

1.4

The wedge

We devote the rest of this chapter to the proof of Lemmas 1.2-1.4. But first
we need to introduce an operation between lines, which will be a fundamental
tool in what follows. Recall the definitions of L and R from 0.2.1.
Definition 1.6. Given two (p, q)-lines `1 and `2 in R2 , their wedge `1 `2

is the line defined as the boundary of the unique unbounded connected

component of L`1 L`2 , oriented so that this component corresponds to

L(`1 `2 ).

Figure 1.1: The wedge of two lines

This operation is called join in [BCLP04] and denoted by .

Recall that a Jordan domain is a topological disk bounded by a simple closed curve. The following theorem guarantees that the wedge is well
defined.
19

Theorem 1.7 (Kerekj

art
o, [Ker23]). If U1 and U2 are two Jordan domains
in the two-sphere, then each connected component of U1 U2 is a Jordan
domain.

In fact, identifying R2 {} with the two-sphere S 2 , we may regard lines

in R2 as simple closed curves in S 2 containing . From the fact that `1 and

`2 are (p, q)-lines, it is easy to see that the intersection of L`1 and L`2 has a

unique connected component containing in its boundary. By the above

theorem, this boundary is a simple closed curve, so that it corresponds to a

line in R2 . One can easily see that this new line is also a (p, q)-line.
We denote the wedge of multiple lines `1 , . . . , `n by
`1 `2 `n =

n
^

`i .

i=1

This is well defined because the wedge is commutative.

The following proposition resumes the interesting properties of the wedge.
Proposition 1.8. The wedge is commutative, associative and idempotent.
Furthermore,
1. The wedge of (p, q)-lines is a (p, q)-line;
2. If h Homeo(R2 ) is a lift of a torus homeomorphism, then h(`1 `2 ) =
h(`1 ) h(`2 );

3. `1 `2 `1 and `1 `2 `2 ;
4. If `1 < `2 and 1 < 2 , then `1 1 < `2 2 ;
5. The wedge of Brouwer lines is a Brouwer line.

1.5

Proof of Lemma 1.2

Let ` be a Brouwer (0, 1)-line for f n , for some n > 1. We will show that
there is a Brouwer (0, 1)-line for f n1 ; by induction, it follows that there is
a Brouwer (0, 1)-line for f .
20

We know that f n (`) > `. Let be a (0, 1)-line such that f n (`) > > `,
and define
0

` =
By Proposition 1.8,
line for

f n1 .

`0

n1
^

f i (`).

i=1

is still a (0, 1)-line. We claim that it is a Brouwer

In fact,
f

n1

(` ) = f

n1

()

= f n1 ()
=f

n1

n1
^

i=1
n1
^

f n1 (f i (`))
f i1 (f n (`))

i=1
n

() f (`)

= f n1 () f n (`)

n1
^

i=2
n2
^

f i1 (f n (`))
f i (f n (`)).

i=1

Using the facts that

f n1 () > f n1 (`),

f n (`) > , and f i (f n (`)) > f i (`),

and Proposition 1.8, we see that

f n1 (`0 ) > f n1 (`)

n2
^
i=1

f i (`) =

n1
^

f i (`) = `0 ,

i=1

so that `0 is a Brouwer (0, 1)-line for f n1 . This concludes the proof.

1.6

Proof of Lemma 1.3

Let ` be a Brouwer (0, 1)-line for f . We consider two cases.

21

Case 1. For all n > 0, f n (`) T1n (`)

In this case, for each n > 0 we can choose xn ` such that f n (xn ) R(T1n `).
From the fact that ` is a (0, 1)-line we also know that pr1 (`) [M, M ] for
some M > 0, and therefore

pr1 (T1n (`)) [M + n, M + n].

Hence,
pr1 (f n (xn )) pr1 (R(T1n (`))) [M + n, ).
It then follows that
 n

f (xn ) xn
M
(M + n) M
n
pr1
= 2
+ 1 1,

n
n
n
and by the definition of rotation set this implies that some point (x, y) (f )

satisfies x 1; i.e. max pr1 ((f )) 1.

Case 2. f n (`) < T1n ` for some n > 0

We will show that in this case F has a free (0, 1)-curve. The idea is similar
to the proof of Lemma 1.2. Let n be the smallest positive integer such that
f n (`) < T1n `. If n = 1, we are done, since ` < f (`) < T1 ` so that ` projects
to a free (0, 1)-curve for F .
Now assume n > 1. We will show how to construct a new Brouwer
(0, 1)-line for f such that f n1 () < T1n1 . Repeating this argument
n 1 times, we end up with a (0, 1)-line `0 such that `0 < f (`0 ) < T1 (`0 ), so

that `0 projects to a free curve, completing the proof.

Let be a (0, 1)-curve such that
f n (`) < < T1n `

(1.1)

We may choose such that f () > , by taking it close enough to f n (`). This
is possible because f (f n (`)) > f n (`), and these two curves are separated by a
positive distance, since they lift (0, 1)-curves in T2 . Thus is also a Brouwer

22

(0, 1)-line for f .

Define
=

n1
^

T1ni f i (`).

i=1

Let us see that f n1 () < T n1 (). Since T1 commutes with f , we have

f

n1

() = f

n1

()

= f n1 ()
=f

n1

()

n1
^

i=1
n1
^

f n1 T1ni f i (`)
f i1 T1ni f n (`)

i=1

T1n1 f n (`)

n1
^

f i1 T1ni f n (`)

i=2

By (1.1), we also have

f n1 () < f n1 (T1n `),
T1n1 f n (`) < T1n1 , and
f i1 T1ni f n (`) < f i1 T1ni T1n `.
Using these facts and Proposition 1.8 we see that
f n1 () < f n1 (T1n `) T1n1

n1
^

f i1 T1ni T1n (`)

i=2

= T1n1 (T1 f n1 (`)) T1n1 ()

= T1n1 T1 f n1 (`)
=

T1n1

T1n1 ().

n1
^

n2
^
i=1

n1
^
i=2

n(i1) i1

T1n1 (T1
!

T1ni f i (`)

T1ni f i (`)

i=1

Since is a Brouwer (0, 1)-line, this completes the proof.

23

(`))

1.7

Proof of Lemma 1.4

Suppose that N int(pr1 (f )) Z. We will show that F cannot have any

free (0, 1)-curves. Note that this also implies that F cannot have a free

(0, 1)-curve.

Considering the lift T1N f instead of f , we may assume that N = 0. If

F has a free (0, 1)-curve, then any lift of this curve is a (0, 1)-line ` such that
f (`)` = . This means that either f (`) > ` or f (`) < `, and by Lemma 0.22

this implies that (f ) is contained in one of the semiplanes {(x, y) : x 0}

or {(x, y) : x 0}. Hence 0 is an extremal point of pr1 ((f )), contradicting

our assumption that it was an interior point.

Thus we know that pr1 ((f )) has no integer points in its interior. But
pr1 ((f )) is an interval, so that it must be contained in [k, k + 1] for some
integer k.

24

Chapter 2

Free curves and fixed points

As before, we assume throughout this chapter that F is a homeomorphism
of T2 isotopic to the identity and f is a lift of F .
To prove Theorem B, we have two main cases. The first one is when the
rotation set is either a segment of rational slope or a single point; this is
dealt with Theorem A and the results stated in Chapter 0. The second case
is when the rotation set is a segment of irrational slope. In that case, the
main idea is to find A GL(2, R) such that A (f ) has no integers in the
first or second coordinate, so that we may apply directly Theorem 0.24 to

Af A1 (c.f. 0.3.3). In fact, using this argument, we obtain the following

more general result:

Theorem 2.1. Suppose (f ) is a segment of irrational slope with no rational

points. Then for each n > 0 there is an essential simple closed curve such
that , F (), . . . , F n () are pairwise disjoint.
The problem of finding the map A previously mentioned is mainly an
arithmetic one, and we consider it first. In the next section, we briefly
discuss a few facts about continued fractions that will be needed in the
proof; in 2.2 we prove the two arithmetic lemmas that allow us to find the

map A; in 2.3 we prove Theorem 2.1; finally, in 2.4 we complete the proof

of Theorem B.

25

2.1

Continued fractions

Given an integer a0 and positive integers a1 , . . . , an , we define

1

[a0 ; a1 , . . . , an ] = a0 +
a1 +

1
..

1
+

an

Given R, define {n } and {an } recursively by a0 = bc, 0 = a0 ,

and



an+1 = 1
, n+1 = 1
n
n an+1 ,

whenever n 6= 0. This gives the continued fractions representation of : If

is rational, we get a finite sequence a0 , . . . , an , and
= [a0 ; a1 , . . . , an ].
If is irrational, then the sequence is infinite and
.
= [a0 ; a1 , a2 , . . . ] = lim [a0 ; a1 , . . . , an ].
n

The rational number pn /qn = [a0 ; a1 . . . , an ] is called the n-th convergent

to . Convergents may be regarded as the best rational approximations
to , in view of the following properties (see, for instance, [HW90])
Proposition 2.2. If pn /qn are the convergents to , then
1. {qn } is an increasing sequence of positive integers, and
1
1
< (1)n (qn pn ) <
.
qn + qn+1
qn+1
2.

p2n
q2n

<

p2n+2
q2n+2

<<

p2n+3
q2n+3

<

p2n+1
q2n+1 .

3. pn+1 qn pn qn+1 = (1)n

26

A Farey interval is a closed interval with rational endpoints [p/q, p0 /q 0 ]

such that p0 q q 0 p = 1. Note that two consecutive convergents give Farey
intervals.

Proposition 2.3. Let [p/q, p0 /q 0 ] be a Farey interval. Then



p + p0 p0
,
q + q0 q0

and

p p + p0
,
q q + q0

are Farey intervals, and if p00 /q 00 (p/q, p0 /q 0 ), then q 00 > q + q 0 .

2.2

Arithmetic lemmas

Define the vertical and horizontal inverse Dehn twists by

D1 : (x, y) 7 (x y, y), D2 : (x, y) 7 (x, y x).
Let Q be the set of vectors of R2 with positive coordinates; for u = (x, y)

Q, we denote by slo(u) = y/x the slope of u. Let Q1 and Q2 be the sets of

elements of Q having slope smaller than one, and greater than one, respectively.
Remark 2.4. Note the following simple properties
1. for i = 1, 2, Di Qi = Q; and if u Q, then Dik u Qi for all k > 0;
2. for i = 1, 2, kDi uk < kuk if u Qi ;
3. slo(D2k u) = slo(u) k and slo(D1k u)1 = slo(u)1 k.
Lemma 2.5. Let u and v be elements of Q with different slopes. Then there
is A GL(2, Z) such that
1. kAuk kuk;
2. kAvk kvk;
3. Both Au and Av are in Q, and either one of these points is on the
diagonal and the other in Q1 , or one is in Q1 and the other in Q2 .
27

Proof. We first note that it suffices to consider the case where both u and
v are in Q1 . Indeed, if one of the vectors is in Qi and the other is not (for
i = 1 or 2), there is nothing to do; and if both u and v are in Q2 then we
may use Su and Sv instead, where S is the isometry (x, y) 7 (y, x).

Given u Q1 , we define a sequence of matrices An SL(2, Z) and

integers an by A0 = I, a0 = 0, and recursively (see Figure 2.1)

If slo(An u) = 1 stop the construction.
a

an+1 is the smallest integer such that Di n+1 An u

/ Qi , where i = 2 if
n is odd, 1 if n is even;
a

An+1 = Di n+1 An .
In this way we get either an infinite sequence, or a finite sequence
A1 , . . . AN such that AN u lies on the diagonal and has positive coordinates.
Furthermore, given 0 n < N if the sequence is finite, or n 0 if it is
infinite, we have

An u Qi where i = 2 if n is odd, 1 if n is even;

n

If n = slo(An u)(1) , then n = 1

n1 an
kA0 uk , kA1 uk , . . . is a decreasing sequence;
The first property is a consequence of the definition. The second follows
from Remark 2.4, since for odd n, we have
n = slo(D1an An1 u)1 = slo(An1 u)1 an = 1
n1 an
while for even n,
n = slo(D2an An1 u) = slo(An1 u) an = 1
n1 an .
The last property also follows from the construction, since if An u = Dian An1 u
then Dik An1 u Qi for all 0 k < an , so that Remark 2.4 implies that

kAn uk < kAn1 uk.

28

D1

D1

A1 u

D2
D1

D1
A2 u

Figure 2.1: The sequence Ai u

If n is odd, an is the smallest integer such that D1an An1 u
/ Q1 , or
equivalently (assuming that n < N if the sequence an is finite), the smallest
integer such that
1
1.
slo(D1an An1 u) = (slo(An1 u)1 an )1 = (1
n1 an )

Since An1 u Q1 , slo(An1 u) < 1 so that 1

n1 > 1, and an > 0. Note that

1
n cannot be an integer, since otherwise slo(An u) = 1, which contradicts
the fact that An u Q2 ; thus


an = 1
n1 .

If n is even, the above equation holds by a similar argument. One easily

sees from these facts that n coincides with the sequence obtained in the
definition of the continued fractions expression of 0 = slo(u), and thus an
coincides with the continued fractions coefficients of slo(u).
Now given v Q1 with slo(v) 6= slo(u), define in the same way as above

sequences of positive integers bn and of matrices Bn SL(2, Z) such that

B0 = I, bn is given by the continued fractions expression of slo(v), and

b

Bn+1 = Di n+1 Bn where i = 2 if n is odd, 1 if n is even. As before, we have

29

that kBn vk is a (finite or infinite) decreasing sequence, and if it is finite of

length N then slo(BN v) = 1. Also Bn v Qi where i = 2 if n is odd, 1 if n

is even (given that n < N if the sequence is finite).

D12 u

D12 v

D2 D12 u
D2 D12 v

Figure 2.2: Example

Since u and v have different slopes, their continued fractions expression
cannot coincide. Thus there exists m 0 such that a0 = b0 , . . . , am = bm

but bm+1 6= am+1 . We may further assume that am+1 < bm+1 , by swapping
u and v if necessary. This means that Ak = Bk for 0 k m; so that if m

is even, Am u Q1 and Am v Q1 , but since am+1 < bm+1 , it holds that

a

Am+1 u = D1 m+1 Am u
/ Q1 but Am+1 v = D1 m+1 Bm v Q1 ;
a

that is, D1 m+1 pushes Am u out of Q1 , while leaving Am v in Q1 (see Figure

2.2).
Moreover, since am+1 is minimal with that property, either slo(Am+1 u) =
1 or Am+1 u Q2 ; and in either case Am+1 u has positive coordinates. By

construction, it also holds that kAm+1 vk kvk and kAm+1 uk kuk.

If m is odd, a similar argument holds, and we see that Am+1 u Q2

while Am+1 v is either on the diagonal or in Q1 .

Starting from u and v in Q1 we obtained A = Am+1 GL(2, Z) such

that Au and Av have positive coordinates and either one of them is on the
30

diagonal or one is in Q1 and the other in Q2 ; furthermore, kAuk kuk and

kAvk kvk. This completes the proof.

Lemma 2.6. Let w = (x, y) be a vector with irrational slope. Then, for any
 > 0 there exists A SL(2, Z) such that kAwk < .
Proof. Let pi /qi be the convergents to y/x, and define
Ak =

(1)k pk (1)k+1 qk
qk+1

pk+1

By Proposition 2.2 we have that det A = 1, the sequence q1 , q2 , . . . is increasing, and


y 
1

pi qi  <
x
qi+1

for all i 1. Hence,

and similarly

 
y 
|x|

|pr1 Ak w| = x pk qk  <
,
x
qk+1

 

y
|x|


|pr2 Ak w| = x pk+1 qk+1  <
;
x
qk+2

Choosing k large enough so that qk+1 > 2 |x| 1 , we have kAk wk < .

2.3
2.3.1

Proof of Theorem 2.1

The case n = 1

We first assume n = 1, i.e. we prove that F has a free curve, assuming that
(f ) is a segment of irrational slope containing no rational points. The problem is reduced, by Lemma 0.25 and Theorem 0.24, to finding A GL(2, Z)
such that the projection of A (f ) to the first or the second coordinate contains no integers. Note that Lemma 2.6 allows us to assume that
1
diam((f )) <  < .
2 5
31

We may also assume that there are m1 pr1 ((f ))Z and m2 pr2 ((f ))
Z, for otherwise there is nothing to do.

Then using T1m1 T2m2 f (which also lifts F ) instead of f , we have that
the extremal points of (f ) are in opposite quadrants. By conjugating f
with a rotation by /2, we may assume that (f ) is the segment joining
u = (u1 , u2 ) and v = (v1 , v2 ) where vi 0 and ui 0, i = 1, 2. From
this, and the fact that diam((f )) < , it follows that
kuk < , and kvk < .
Case 1. One of the points has a zero coordinate
It is clear that neither u nor v can have both coordinates equal to 0. Conjugating by an appropriate isometry in GL(2, Z), we may assume the generic
case that u = (u1 , 0), with u1 > 0. Then v2 > 0: in fact if v2 = 0 then
(f ) contains the origin, which is not possible. Let k > 0 be the greatest
integer such that pr1 D1k v > u1 , i.e.



u1 + v1
k=
.
v2

Note that D1k u = u, so that D1k (f ) is the segment joining u to D1k v (see
Figure 2.3).
Moreover, D1k+1 v = (v10 , v2 ) where v10 = v1 (k + 1)v2 < u1 . Thus
max pr1 (D1k+1 (f )) = u1 < 0.
On the other hand,
min pr1 (D1k+1 (f )) u1 v1 > 2 > 1,
so that taking A = D1k+1 we have pr1 ((Af A1 )) (1, 0). By Theorem
A1 has a free (0, 1) curve. Thus F has a free curve.
0.24, it follows that AF

32

D12 v

D1 v

(f )


u = D12 u

Figure 2.3: Avoiding integers in the first coordinate

Case 2. None of the points has a zero coordinate
In this case, u Q and v Q. Since the segment joining u to v cannot

contain the origin, u and v are elements of Q with different slopes; thus

Lemma 2.5 implies that there is A GL(2, Z) such that kAuk kuk,

kAvk kvk, both Au and Av are in Q, and either one of them lies on the

diagonal, or they are in opposite sides of the diagonal. This means that Au

and Av are both contained in one of the closed semiplanes determined by

the diagonal. By using A instead of A if necessary, we may assume that

both are in the closed semiplane above the diagonal, which is mapped by
D2 to the upper semiplane H = {(x, y) : y 0}. Note that
kD2 Auk kD2 k kAuk
and similarly kD2 Avk

5 kuk ,

5 kvk. If pr2 (D2 Av) > 0 and pr2 (D2 Au) > 0,

then (see Figure 2.4) we have that

pr2 (f ) (0,

5) (0, 1),

and by Theorem 0.24 (as in Case 1) it follows that F has a free curve. On
the other hand, if either of D2 Av or D2 Au has zero second coordinate, the
argument in Case 1 implies that F has a free curve.
This completes the proof when (f ) has irrational slope and n = 1.
33


(0, 5)

Av

D2 Au
D2 Av
Au
(0, 0)
Au

Figure 2.4: Avoiding integers in the second coordinate

2.3.2

The case n > 1

Note that when (f ) has irrational slope, (f n ) = n (f ) has irrational slope

for all n.
Let N = n!. As we saw in the previous case, conjugating our maps by
some A GL(2, Z), we may assume pr1 (f N ) Z = ; thus
pr1 (f N ) (K, K + 1) for some K Z,
and by Theorem 0.8,
pr1 (f )

K K +1
,
N
N

Let [p/q, p0 /q 0 ] be the smallest Farey interval containing pr1 ((f )). We claim
that q + q 0 > n. In fact, if q + q 0 n, then [K/N, (K + 1)/N ] must be

contained in one of the smaller Farey intervals (see Proposition 2.3)



p p + p0
,
q q + q0

or

34


p + p0 p0
,
.
q + q0 q0

This is because N (p + p0 )/(q + q 0 ) is an integer, so that it cannot be in the

interior of [K, K + 1]. But we chose our Farey interval to be the smallest, so
[p/q, p0 /q 0 ] must as well be contained in one of these two intervals, which is
a contradiction. Thus q + q 0 > n, and Theorem 0.24 guarantees that there
is an essential simple closed curve such that its first n iterates by F are
pairwise disjoint. This completes the proof.

2.4

Proof of Theorem B

Assume that (f ) has empty interior. We will show that either F has a fixed
point, or it has a free curve. There are several cases:
(f ) is a segment of irrational slope which contains no rational points.
Then there is a free curve, by Theorem 2.1.
(f ) is a segment of irrational slope containing a rational non-integer

point (p1 /q, p2 /q). By Lemma 2.6 there exists A GL(2, Z) such

that diam(A (f )) < 1/q. One of the two coordinates of A(p1 /q, p2 /q)
must be non-integer. We assume p0 /q 0 = pr1 A(p1 /q, p2 /q)
/ Z (oth-

erwise, we can conjugate f with a rotation by /2, as usual). Since

A(p1 , p2 ) is an integer point, it follows that q 0 q (if we assume p0 /q 0
is irreducible). Thus,
1

pr1 (Af A

) = pr1 (A (f )) pr1

p0 1 p0 1
, 0+
q0
q q
q

A1
It is clear that the interval above contains no integers, so that AF
(and, consequently, F ) has a free curve by Theorem 0.24.
(f ) is a segment of irrational slope with an integer point. Then F
has a fixed point by Theorem 0.12 (see also Remark 1.5).

(f ) is a single point. Then either this point is integer, and F has a

fixed point by Theorem 0.10 or it is not integer, and F has a free curve

by Theorem 0.24.

35

 (f ) is a segment of rational slope. Conjugating all the maps by an

element of GL(2, Z) we may assume it is a vertical segment; and with
this assumption, if both pr1 ((f )) and pr2 ((f )) contain an integer,
it follows that (f ) contains an integer point, and by Theorem A, F
has either a fixed point or a free curve. On the other hand, if either
of the two projections contains no integer, Theorem 0.24 implies the
existence of a free curve for F as before.
This concludes the proof.

36

Chapter 3

Always free curves

Given a homeomorphism F : T2 T2 (not necessarily in the homotopy class

of the identity), we say that a simple closed curve is always free if

F n () = for all n Z, n 6= 0.

In this chapter we prove Theorem C, which says that if F has an always

free curve with a dense orbit, then F is semi-conjugate to an irrational
rotation of the circle; i.e. there exists a continuous surjection h : T2 S 1
such that hF = Rh, where R is some irrational rotation of S 1 .

The natural idea for proving the theorem is defining the map h on the
orbit of by F , mapping it to the orbit of some point x0 in S 1 by some
appropriately chosen map (which eventually turns out to be conjugated to
an irrational rotation), and then extending h to T2 . We follow this idea
in a slightly indirect way: we explicitly construct a partition of T2 which
will correspond to {h1 (x) : x S 1 } and then we show that by collapsing

elements of this partition, the map induced by F on the quotient space is

conjugated to an irrational rotation of S 1 .
With minor modifications, the proof presented here also works if is an
essential continuum (i.e. a continuum such that 1 () has an unbounded
connected component), with the additional hypothesis that there is some essential simple closed curve disjoint from . The latter condition is necessary

37

to ensure that F has a lift to the annulus. Thus we have

Theorem C. Let F Homeo(T2 ), and let T2 be an essential continuum with a dense orbit, such that all its iterates by F are pairwise disjoint.
Assume further that is disjoint from some essential simple closed curve.
Then f is semi-conjugated to an irrational rotation of the circle.
Homeomorphisms in the homotopy class of the identity satisfying the
hypothesis of Theorem C are almost fibered, in the sense that they are
similar to skew-products of the form (x, y) 7 (f (x), gx (y)) where f : S 1

S 1 has an irrational rotation number. The partition found in Theorem

C resembles the foliation by vertical circles in skew-products of this kind,

except that it could consist of general continua instead of curves.
In [Her83], Herman proved that skew-products always have a rotation
set consisting of a single point (i.e. they are pseudo-rotations). Thus, the
previous discussion motivates the following
Question 3.1. If F Homeo (T2 ) has an always free curve with a dense

orbit, must its rotation set consist of a single point?

3.1

Proof of theorem C

Let 0 be an always free curve for F with a dense orbit.

Proposition 3.2. 0 is homotopically non-trivial.
Proof. Suppose 0 is homotopic to a point, and let D be the topological disk
bounded by it in T2 . We know that 0 has a dense orbit, so that it must
hold F n (0 ) D 6= for some n which we may assume positive (otherwise
consider F 1 ). Moreover, we assume n to be the smallest positive integer

with that property. Since 0 is always free, we must have F n (D) D and
F k (D) D = when 0 < k < n. If A = D cl(F n (D)), then it follows

easily that F k (0 ) A = for all k Z. But since A is open and nonempty,

this contradicts the fact that the orbit of 0 is dense.

38

By the above proposition, 0 is not homotopic to a point. This means

that 0 is a (p, q)-curve for some (p, q) Z2 . Moreover, since 0 is simple, p

and q are coprime; hence we can find A GL(2, Z) such that A(p, q) = (0, 1)
(see Remark 0.26). Such A induces a homeomorphism A of T2 , such that

0 is an always free (0, 1)-curve for AF

A1 ; hence, it suffices to prove the
A
theorem when 0 is a (0, 1)-curve, which we will assume from now on.
Since F (0 ) is a simple closed curve which is disjoint from 0 , it must
be either a (0, 1)-curve or a (0, 1) curve (see Remark 0.17). This easily

implies (by the lifting theorem) that F can be lifted to f : A 7 A by the

covering 1 , and 0 (and any iterate of it by F) is lifted to a simple closed
curve on A.
Recall that is the translation (x, v) 7 (x + 1, v) in A. Let
= { m f n () : (n, m) Z Z}.
Note that the elements of are pairwise disjoint simple closed curves, and
is dense in A.

For any , there are exactly two connected components in A \ ,

both unbounded. We write L for the one on the left, and R for the
remaining one. We also write L and R for their respective closures. Since
all elements of are disjoint, the relation defined by < 0 if L 0

(which is the same as L L 0 ) is a linear ordering of .

Proposition 3.3. For any 0 and 1 in ,

1. f (L0 ) = Lf (0 ), f (R0 ) = Rf (0 ), and similarly for ;

2. If 0 < 1 then f (0 ) < f (1 ) and (0 ) < (1 );
3. L0 = 0 = R0 ;
4. (0 ) > 0 . In particular, there is no maximum or minimum element
of ;
5. If 0 < 1 , then there exists such that 0 < < 1 .

39

Proof. (1) The assertion for is obvious since L0 and R0 are left and
right unbounded, respectively, and is just a translation.
As it is easy to see, either f (L) = Lf () and f (R) = Rf () for all
, or f (L) = Rf () and f (R) = Lf () for all ; thus it suffices

to prove the property for the original curve 0 = . We may also assume

that f is such that < f () < (), by choosing an appropriate m and

using m f , which is also a lift of F , instead of f . Suppose that f (R) =
Lf (). Then, letting U = R Lf (), we have f (U ) U , and 1 (U )

1 (R L ()) which is a proper open subset of T2 , whose complement has

nonempty interior. Since F (1 (U )) 1 (U ), this contradicts the fact that

has a dense orbit.

(2) By item 1, Lf (0 ) = f (L0 ) f (L1 ) = Lf (1 ), and a similar

argument for .

(3) Follows from the fact that 0 is simple.

(4) Suppose (0 ) < 0 ; then, by item 2, n (0 ) < 0 for all n > 0. But
it is clear that if x 0 , then for n large enough, n (x) R0 , which is a

contradiction. Thus (0 ) > 0 .

(5) Since U = L1 R0 is open and is dense, there exists

such that U 6= . But then must be entirely contained in U .

For 1 , 2 , we define the open strip S(1 , 2 ) = R1 L2 , and the

closed strip S(1 , 2 ) = R1 L2 = S(1 , 2 ) 1 2 .

Remark 3.4. Note that if 1 < 2 , then S(1 , 2 ) is homeomorphic to A.

Also note that finite intersections of open (resp. closed) strips are open
(resp. closed) strips.

3.1.1

Construction of an invariant partition

Given x A, we will say that a strip S0 = S(1 , 2 ) is good for x if x belongs

to some closed strip which is contained in S0 . Define

Px = {S 0 : S0 is good for x }

and let P = {Px : x A}. This will be our partition.

40

Remark 3.5. By proposition 3.3, for any x we can find 10 , 20 such that x
R10 R20 . Thus, if we have 1 < 2 such that x L1 , then necessarily

Px L2 ; moreover, since there exists some such that 1 < < 2 , we

have Px L2 . An analogous property holds in the other direction. As a

consequence, we may take the intersection of open strips instead of closed

ones in the definition of Px .

Proposition 3.6. The following properties hold:
1. P is a partition;
2. Each element of the partition is compact, connected, has empty interior, and splits A into exactly two connected components.
3. P is f -invariant, i.e. f (Px ) = Pf (x) for all x A;
4. P is -invariant, and consequently {1 (P ) : P P} is an F -invariant
partition of T2 ;

5. For any x, y A, if Px 6= Py then there exists that separates Px

and Py , so that Px L and Py R or vice-versa;

6. For each x A, at most one element of intersects Px , and when it

does it is contained in Px .

Proof. (1): It is clear that x P for all x A; thus it suffices to show that

whenever Px 6= Py , it holds that Px Py = . Fix x, y A, and suppose

there are 1 < 2 such that x L1 and y R2 . Then by Proposition

3.3 we may find such that 1 < < 2 , so that x L1 L and

y R2 R. By Remark 3.5, Px L and Py R; thus, Px Py =

and separates Px from Py .

Now suppose that no element of separates Px from Py . Then whenever

1 < 2 and x L1 , we must have y L2 ; by a symmetric argument,
also if 1 < 2 and x R2 , we have y R1 . This implies that y belongs

to every good strip of x, so that x Py . Analogously, y Px . We will

show that every good strip for x is a good strip for y; that will imply that
Px Py , and by symmetry that Py Px finishing the proof of this item.
41

Let S0 = S(10 , 20 ) be a good strip for x. By definition, there is a closed

strip S 1 = S(1 , 2 ) such that x S 1 S0 . We know that y Px , so
that y S0 . If y S 1 , then S0 is a good strip for y and we are done.

Assume that y
/ S 1 . Then either y L1 or y R2 . Suppose without

loss of generality that y L1 . By Proposition 3.3 there is such that

10 < < 1 , and we must have y R. In fact if this is not the case,

y L and x R1 , which by the argument in the beginning of the proof

implies that Px and Py are disjoint, contradicting our assumption. Hence,

y S(, 1 ) S0 which means that S0 is a good strip for y, as we wanted.

(3 and 4): This is clear since Proposition 3.3 implies that the image by f
(resp. f 1 , , 1 ) of a good strip for x is a good strip for f (x) (resp. f 1 (x),
(x), 1 (x)).
(5) Follows from the proof of item 1.
(6) Let , x . Then it is clear that any good strip for x contains

, so Px . Given 0 6= , suppose without loss of generality that 0 > .

Then x L L 0 , so Px L 0 by remark 3.5. Hence Px does not

intersect 0 .

(2) Since is dense and is a union of simple closed curves, any open

set intersects infinitely many elements of ; by part 6, Px intersects at most

one such element, so it has empty interior.

It is clear that x Px for all x A, and it is easy to see that finite

intersections of good strips for x are good strips for x. Any family of compact

connected nonempty sets which is closed under finite intersections has a

compact, connected intersection; thus Px is compact and connected.
It remains to show that A Px has exactly two connected components.

Suppose first that A Px is connected. Since it is open, it is arc-connected.

Take 1 < 2 such that Px S(1 , 2 ), let N be large enough so that

both 1 and 2 are contained in [N, N ] S 1 , and let be a compact

arc joining (N 1, 0) to (N + 1, 0), contained in A Px . Then any

such that 1 < < 2 intersects , so that any good strip for x intersects
. Hence, the family {good strips for x} {} has the finite intersection

property, and therefore its intersection is nonempty. This means that Px

intersects , which contradicts our choice of . Thus A Px has more than

42

one connected component.

Now suppose A Px has more than two components. Then, since Px is

bounded, there is some connected component U of A Px that is bounded.

Since U is open, infinitely many elements of intersect it; each one of these

elements must then be contained in U , or intersect U Px . Since we

already saw that at most one element of intersects Px , there must exist

such that U . Since Px is connected, either Px L or Px R.

But U intersects both L and R, so that it has boundary points on both

sets, which is a contradiction. This shows that A Px has exactly two
connected components.

3.1.2

Defining order in P

Define an order in P by P < Q if there is such that P L and

Q R. In view of the previous propositions, it is clear that this defines a

total order. For P P, we will denote by L(P ) and R(P ) the two connected

components of A P , being L(P ) the one on the left; and we will endow

P with the topology induced by this order. Note that for P, Q P, P < Q
means that P L(Q) and Q R(P ).

Proposition 3.7. (P, <) is separable and has the property of the supremum.
Proof. From Propositions 3.3 and 3.6 it easily follows that
D = {P P : P contains some element of }
is dense; and it is countable because is countable.
Now let P0 P and let S P be such that P < P0 for all P S. Let
L=

L(P ).

P S

Then L is open and L L(P0 ). Since A L 6= , we may choose x L.

We claim that L Px is empty. In fact, if y L Px then y L(P ) for

some P S, so that Px = Py < P , which implies that x L(P ) L

and this contradicts the choice of x. Thus we have P < Px for all P S.
43

Furthermore, if Q P is such that P < Q for all P S, then L L(Q),

so that L cl(L(Q)). Since Px contains a boundary point of L, we have

Px cl(L(Q)) 6= , which implies that Px Q. Hence Px is the lowest upper

bound of S.

Let p : A P be the quotient projection that maps x A to the unique

element of P that contains x. The quotient topology on P is given by



A P : p1 (A) is open in A .

Note that p1 (A) = A.

Proposition 3.8. The order topology of P coincides with the quotient topology induced by p.
Proof. We use the interval notation (P1 , P2 ) = {P P : P1 < P < P2 }.

Such intervals form a basis of open sets for the order topology. Given I =
(P1 , P2 ), we see that p1 (I) = L(P2 ) R(P1 ) which is open in A, so that I
is open in the quotient topology.

Now let A P be such that U = p1 (A) is open in A. We will find, for

every P A, an interval I = (P1 , P2 ) such that P I A, thus showing

that A is open in the order topology. Suppose that this is not the case for
some P A. Since L(P ) P and R(P ) P , we can choose x, y P

such that any neighborhood of x intersects L(P ) and any neighborhood of y

intersects R(P ). Since U is open and contains P , we may find  small enough
so that B (x) U and B (y) U . Let x0 B (x)L(P ), y 0 B (y)R(P ),

and let be an arc contained in U and joining x to x0 . Then any Q P

such that Px0 < Q < Px must intersect , because connects L(Q) and
R(Q), and Q separates those sets. Since U and U = A, every element

of P that intersects U must be contained in U ; thus Q U . This shows

that (Px0 , Px ) A. An analogous argument with y and y 0 shows that

(Py , Py0 ) A. Since Py = Px = P , it follows that (Px0 , Py0 ) A as we

wanted.

44

3.1.3

Construction of the semi-conjugation

We now know that the order topology in (P, <) coincides with the quotient
topology induced by p. Moreover, (P, <) is a separable totally ordered space
which has no maximum or minimum element (i.e. it is unbounded), and has
the property of the supremum (i.e. it is complete). It is well known that
any such space is homeomorphic to R by an order-isomorphism (see, for
example, [Ros82, 4.2]).

By Proposition 3.6, the maps f and induce homeomorphisms f and

of P which commute. Moreover, {

n : n Z} is a properly discontinuous
.
action of Z on P. Thus, S = P/h
i is homeomorphic to R/Z ' S 1 . The
quotient projection p also induces a projection p: T2 S, and since f
commutes with , it induces a map F : S S such that the following

diagrams commute:

T2

py

A A

py
py
f

T2

py

S S

P P

It is clear that F is semi-conjugated to F (by means of p). We show that

F is transitive. If I, J are nonempty intervals in P, then U = p1 (I) and

V = p1 (J) are open subsets of A, and both U and V must contain an

element of . If m1 f n1 U and m2 f n2 V , then
m2 m1 f n2n1 (U ) V 6= .

This implies that m2 m1 fn2 n1 (I) J 6= . Since this holds for any pair of
intervals I, J in P, it follows easily that F is transitive.

By the theory of Poincare, any transitive homeomorphism of S 1 is conjugated to an irrational rotation. Since F is semi-conjugated to F , we conclude
that F is semi-conjugated to an irrational rotation of S 1 . This completes
the proof of Theorem C.

45

Chapter 4

Invariant foliations and

minimal diffeomorphisms
In this chapter, we study the diffeomorphisms in the closure of the conjugacy
class of the set of rigid translations of T2 . This is motivated by the work of
Fathi and Herman [FH77], where they prove the existence of strictly ergodic
(i.e. uniquely ergodic and minimal) diffeomorphisms on any manifold M
admitting a free C action of S 1 . In their article, they prove that the set

of such diffeomorphisms is residual in the space O (), the C -closure of

the set of diffeomorphisms C -conjugated to elements of the action.

We denote by R(1 ,2 ) either the translation (x, y) 7 (x + 1 , y + 2 ), or

the diffeomorphism of T2 lifted by it (with a slight abuse of notation). Let



O(T2 ) = hR h1 : h Diff (T2 ), T2 .

If is the action t 7 R(0,t) of S 1 on T2 , it is easy to see that O () =

O (T2 ), where the closure is taken in the C -topology.

Recall that Diff (T2 ) with the C -topology is a Polish space, so that

any closed subset of it (in particular, O (T2 )) is a Baire space. A property

is said to be generic in a Baire space if it holds in a residual subset (i.e. a

dense G ) of the space.
A homeomorphism f : X X of the compact metric space X is minimal
46

if the f -orbit of every point is dense in X or, equivalently, if the only compact
f -invariant sets are the whole space X and the empty set.
By a C r foliation of codimension 1 of T2 , we mean a partition F of the

space into 1-dimensional, connected topological sub-manifolds (the leaves)

such that there is a C r atlas of T2 formed by charts (i , Ui ) such that i 1
j

has the form (u(x), v(x, y)) on its domain, and if S is a vertical segment in
the domain of i , then 1
i (S) F1 (p) for any p S (where Fx denotes

the unique leaf of the partition that contains x).

The foliation F is said to be invariant for a homeomorphism f : T2 T2

if f (Fx ) = Ff (x) for each x T2 . A foliation of T2 always has a lift to R2 ,

i.e. a foliation F 0 such that (Fz0 ) = F(z) for any z R2 . This foliation is
invariant for any lift of f to R2 if F is invariant for f .

We say that a C 0 foliation is induced by a C 0 line field if its leaves are

C 1 , and if the tangent spaces to the leaves form a continuous sub-bundle

of the tangent space of T2 . Note that this is weaker than saying that the
foliation is of class C 1 , but stronger than just saying that the leaves are of
class C 1 .

4.1

Dynamic cocycles

Since the tangent space of T2 is parallelizable, we identify it with T2 R2 .

Definition 4.1. Given f Diff(T2 ), we define its associated dynamic co-

cycle by

f: T2 S 1 T2 S 1 ,

(x, v) 7

Dfx v
f (x),
kDfx vk

which is of class C if f is C .
It is useful to note that fcg = fg. This is true because
Dfg(x)

Dgx v
kDgx vk

Dfg(x) Dgx v
D(f g)x v




=
Dfg(x) Dgx v =

Dg v
kD(f g)x vk
Dfg(x) kDgxx vk
47

 fh
1 .
Hence, if g = hf h1 , where h Diff (T2 ), then g = h

We will denote by d2 the flat metric on T2 , by d1 the one on S 1 , and we

endow T2 S 1 with the metric

d ((x, v), (y, w)) = max {d2 (x, y), d1 (v, w)} .

4.2

Generic minimality of the dynamic cocycle

To prove Theorem D, we define an appropriate family Un of open subsets

of O (T2 ) such that every element of n Un satisfies the required property.

Then, following the ideas of [FH77], we prove that R(0,p/q) is in the closure of
Un , for each p/q Q and each n. Using this fact, we show that hR(0,p/q) h1

is also in the closure of Un for all h Diff (T2 ). Since such diffeomorphisms

are dense in O (T2 ), we conclude that Un is dense in O (T2 ) for each n.

It then follows that n Un is a residual set with the required properties.

Remark 4.2. With almost no modifications, the proof we present below is

valid if we restrict it to the space Om (T2 ) of elements of O (T2 ) preserving

Lebesgue measure. This is because the conjugations used in the proof are
measure preserving.
If (X, ) is a compact metric space, we say that a map f : (X, ) (X, )

is -minimal if every point has an -dense orbit, i.e. for each x, y X there

is n Z such that (f n (x), y) < .

Remark 4.3. By compactness, if f is -minimal then there is N N, de-

pending only on f , such that {f k (x) : |k| < N } is -dense.

Let
Un =



1

f O (T2 ) : f is -minimal .
n

Note that by the previous remark, and by continuity of f 7 f k in the

C 0 -topology, Un is C 0 -open for each n N.

Claim 1. Let = (0, p/q). Then for each N > 0, there exists h Diff (T2 )

such that hR = R h, and for each integer k with 0 k 2N there is

an open 1/N -dense set Uk T2 such that Dfz is a rigid rotation with angle
48

k/N for each z Uk . Moreover, h is 1/N -close to the identity in the C 0

topology.

Proof. Let us briefly sketch the idea of the construction. We will define
: R2 R2 with the desired properties, which commutes with R (i.e. with
h

(x, y) 7 (x, y + p/q)) and with the translations T1 , T2 (so that it lifts a
is the map we are looking for. We
map of T2 ); then the map h lifted by h
as follows: first we choose a disjoint family of 1/N -dense sets
construct h
Sk , 0 k < 2N , each consisting of equally spaced points and such that

each Sk is invarant by R (and by integer translations). Then we define h

in a way that it coincides with a rigid rotation by angle k/N in a small disk
around each point z Sk , and with the identity outside a neighborhood of

k Sk . Essentially, the action of h near any point of Sk is a copy of its

commutes with
action near any other such point, which guarantees that h
R . We now proceed to the explicit construction.
Let : R2 R R2 be defined by
(x, y, t) = (x cos(t) + y sin(t), x sin(t) y cos(t)),

i.e. (, , t) is a rotation with angle t around the origin. Let b : R R be a

C bump function such that:

b(t) = 1 if |t| 1/6;
b(t) = 0 if |t| 1/2;
|b(t)| 1 for all t R.

p
Define t : R2 R2 by t (x, y) = (x, y, tb( x2 + y 2 )). It is clear that

t is a C map. Moreover, it is a diffeomorphism: in fact it has an inverse

given by


p

(x, y) 7 x, y, tb
x2 + y 2 ,

which is also C . Also note that Dt (z) is a rigid rotation with angle t
when kzk < 1/6, and the support of t is contained in the disk of radius

1/2 centered at the origin.

49

Now let M Z be large enough so that 1/(qM ) < 1/N , let K = 7qM N ,

and for 0 k < 2N , define

Sk =



iN + k jN + k
,
K
K


: (i, j) Z2 .

Note that each Sk is an 1/N -dense subset of R2 , and R Sk = Sk because

K is a multiple of qN . Moreover, the sets Sk , 0 k < 28N are pairwise

disjoint, and any pair of different points in k Sk are at least a distance of

1/K apart. Define
k,z (x, y) = Rz
h



1
2KRz (x, y)

,
2K k/N

k,z (x, y) is a diffeomorphism with support in the disk of radius 1/(2K)

i.e. h
around z, and which is a rotation with angle k/N in a small disk around z.
k,z and
Note that if z Sk and z 0 Sk0 , with z 6= z 0 , then the supports of h
k0 ,z 0 are disjoint; thus they commute. Moreover, if z Sk , then
h
k,z R = R hk,R z .
h

(4.1)

Let
k,z ,
k = h
h
zSk

k,z with z Sk . The order does not matter

i.e. the composition of all h

because any pair of such diffeomorphisms commute. From the fact that
k R = R h
k . Also, the
R Sk = Sk and from (4.1), it easily follows that h
i and h
j are disjoint if i 6= j, so that h
ih
j = h
j h
i.
supports of h
=
h

0k<2N

k.
h

k commutes with R , so does h.

By a similar argument, h

Since each h
commutes with any integer translation, so that it is a lift of a torus diffeo coincides with a rotation by k/N around z
morphism. If z Sk , then h
k is an 1/(12K)in a small disk (of radius 1/(12K)) around z. Thus if U
50

neighborhood of Sk we have that D hz is a rotation with angle k/N around

k . Since U
k is 1/N -dense (because Sk is 1/N -dense),
the origin for each z U

k ) gives an 1/N -dense family of open sets with the required propUk = (U
The fact that h is 1/N -close
erties for the map h Diff (T2 ) lifted by h.
to the identity in the C 0 topology follows from the fact that this is true for
k,z and their supports are disjoint. This completes the proof of the
each h

claim.

Claim 2. R U n for every n N and = (0, p/q).

Proof. Fix n N and = (0, p/q) with p/q Q in reduced form, and let h
be as in the previous claim with N = 4n. If k is a sequence of rationally

independent pairs (so that Rk is minimal) such that k , then

C

hRk h1 hR h1 = R .
k

Thus, it suffices to show that hRk h1 Un for each k. This is possible due

to the minimality of Rk .

Let {Ui : 0 < i < 2N } be the open sets given by the previous claim for
1 (z 0 , v 0 ) = (h1 (z 0 ), w)
h, and fix (z, v) T2 S 1 . If (z 0 , v 0 ) T2 S 1 , then h

for some w S 1 . If the angle (measured in [0, 2)) between w and v is ,

we can find m such that

| m/N | < 1/N < 1/n,

and there is z0 Um such that d2 (z0 , h1 (z)) < 1/N . By the properties of

Um , there is a small neighborhood U of z0 such that Dhz1 is a rotation with

angle m/N for each z1 U , and we may assume that d2 (z1 , z0 ) < 1/N for

each z1 U by taking a smaller neighborhood, if necessary. Since Rk is

minimal, there exists j such that Rj k (h1 (z 0 )) U . Since DRk = Id, it

follows that

1 (z 0 , v 0 ) = (Rj h1 (z 0 ), w).
j h
R
k
k

51

Hence, letting z1 = Rj k h1 (z 0 ), we have z1 U and

1 (z 0 , v 0 ) = (h(z1 ), Dhz w) = h(z
1 , w).
R
j h
h
1
k
But since d2 (z1 , z0 ) < 1/N and d2 (z0 , h1 (z)) < 1/N , and since h is 1/N close to the identity, it follows that d(h(z1 ), z) < 4/N = 1/n. We also
know that d1 (Dhz1 w, v) < 1/n (because Dhz1 is a rotation by m/N and
| m/N | < 1/N < 1/n); thus



1 j 0 0

\
d (hR
k h ) (z , v ), (z, v) = d(h(z1 , w), (z, v)) < 1/n.

Since (z, v) and k were arbitrary, this shows that hRk h1 Un for each k.

This completes the proof of the claim.

Claim 3. O (T2 ) = U n .
Proof. First note that {hR(0,p/q) h1 : p/q Q, h Diff (T2 )} is C -dense

in O (T2 ). In fact, it suffices to show that every element of O(T2 ) is the

limit of a sequence in that set. If hR h1 O(T2 ), we can choose a sequence

n Q2 such that n ; by Remark 0.26, there is An GL(2, Z) and

pn /qn Q such that An Rn A1
n = R(0,pn /qn ) , so that

C
(hAn )R(0,pn /qn ) (hAn )1 = hRn h1 hR h1 .
n

Un

From this fact, to prove the claim it suffices to show that hR(0,p/q) h1
for each n and p/q Q.

Fix n N, p/q Q, and let > 0 be such that d(

x, y) < implies

1 (
1 (
d(h
x), h
y ))

< 1/n. Let k > 1 . Given f Uk , and x

, y T2 S 1 ,

we know from the 1/k-minimality that there is i Z such that



1 (
1 (
d fi (h
x)), h
y ) < 1/k < .

Hence,





1
fh
1 )i (
fi h1 (
h
1 (
d (h
x), y = d h
x), h(
y )) < .
n

fh
1 is 1/n-minimal, i.e.
Since x
and y are arbitrary, it follows that h
52

hf h1 Un . Since this is true for any f Uk , and by the previous claim

we can choose a sequence fi Uk such that fi R(0,p/q) , we have a se-

quence hfi h1 of elements of Un which converges to hR(0,p/q) h1 . Thus

hR(0,p/q) h1 U n .

This completes the proof of the claim.

We proved that each Un is C 0 -open (thus C -open) and C -dense in

O (T2 ). Thus,

R=

nN

Un

is a residual subset of O (T2 ). The fact f is minimal for each f R is

obvious from our definition of the sets Un . This proves Theorem D.

4.3

Generic non-existence of invariant foliations

The following is an easy consequence of Theorem D:

Corollary 4.4. If R is the residual set from Theorem D, and f R, there

is no Df -invariant continuous sub-bundle of T (T2 ) of codimension 1. In

particular, f has no invariant foliation induced by a C 0 line field.

Proof. If z 7 Ez is a continuous map assigning to each z T2 a 1-

dimensional subspace of R2 , then whenever f n (z) is close to z, it must be the

case that Df n (z)Ez = Ef n (z) is close to Ez . But if v Ez is a unit vector,
and w
/ Ez is another, by minimality of f we can find a sequence nk

such that fnk (z, v) (z, w). This clearly implies that Ef nk (z) Rw 6= Ez ,

which is a contradiction.

4.4

Generic non-existence of always-free curves

Combining Theorem C and a result of Herman, we have:

Corollary 4.5. There is a residual set in O (T2 ) such that given any f

in the set and any simple closed curve in T2 , there is n 6= 0 such that
53

f n () 6= . In particular, f cannot have an invariant C 0 foliation with

a compact leaf.

Proof. By [Her92, 9.8], there exists a residual set GP8 O (T2 ) such

that if f GP8 , then f is not semi-conjugated (not even metrically semi-

conjugated) to a translation on a compact Abelian group. This follows from

the fact that all elements of GP8 are minimal and uniquely ergodic, and they
are weak-mixing with respect to its unique invariant measure.
Let f GP8 and let be a simple closed curve. Suppose is always

free. Then, by the minimality of f , the orbit of is dense. Thus, Theorem

C implies that f is semi-conjugated to an irrational rotation of S 1 . But this
contradicts the aforementioned property of GP8 . Hence there exists n 6= 0

such that f n () 6= .

If f has an invariant foliation with a compact leaf , then is a simple

closed curve, so that by the previous argument there exists n 6= 0 such that

f n () 6= . But since is the leaf of an invariant foliation, this implies

that f n () = , so that is a periodic curve for f . In particular, the orbit

of is a proper compact f -invariant subset of T2 . This contradicts the
minimality of f ; hence there is no such foliation.

4.5

C 0 foliations with non-differentiable leaves

We will now show how to construct diffeomorphisms in O (T2 ) such that

any invariant foliation has a leaf which is not C 1 .

For simplicity, we will work in the spaces H0 of all homeomorphisms

h : R2 R2 which are lifts of torus homeomorphisms in the isotopy class

of the identity, and H of all C diffeomorphisms which are lifts of torus

diffeomorphisms in the isotopy class of the identity. We endow H0 with the

C 0 -metric

d0 (f, g) = max

sup kf (z) g(z)k , sup f 1 (z) g1 (z)

zR2

zR2

54

and H with the C -metric

d (f, g) = min 1,

sup
zR2 , k0

)


k

Dz (f g) .

These metrics make both spaces complete, and if a sequence fn H0 (resp.

H ) converges to F , then the corresponding sequence Fn of maps lifted by

fn converges in the C 0 (resp. C ) topology to F (the map lifted by F ).

Theorem 4.6. There is a dense subset S of O (T2 ) such that every f S

is conjugated to a minimal translation but has no invariant foliation with
C 1 leaves.
To prove this theorem, it suffices to show that any translation R is
by elements of S. In fact, since the property of having

C -approximated

no invariant foliation with C 1 leaves is invariant by C conjugation, this

automatically implies that S

= O (T2 ).

Let /2 be as in the proof of Theorem D (Claim 1). That is, /2 is

a rotation by /2 in the disk of radius 1/6 centered at the origin, and the
identity outside the disk of radius 1/2. Note that d0 (, Id) 1.

Now let 0 : R2 R2 be defined by requiring that 0 = on [1/2, 1/2]2

and that 0 (x + m, y + n) = 0 (x, y) + (m, n) for all (m, n) Z2 , (x, y) R2 .

That is the same as

0 =

(m,n)Z2

/2 (x m, y n) + (m, n).

Finally, define n by n (x, y) = 3n 0 (3n x, 3n y). The following properties

are easily verified:
n H and it preserves Lebesgue measure;
if p = 3n (i, j) for some (i, j) Z2 , n is a rotation by /2 around p
on B(p, 3(n+1) /2) and the identity on B(p, 3n /2);

n commutes with R whenever = 3n (i, j) for some (i, j) Z2 ;

d0 (n , id) 3n .
55

Fix R2 and > 0. We now show how to find f H which is the

lift of a map with the required properties, and such that d (f, R ) < .

Let 0 = (p0 /3n0 , r0 /3m0 ) be such that |0 | < /2, where m0 , n0 , p0 , r0

are nonzero integers with m0 > n0 > 0, and let h0 = H0 = id. We define re-

cursively sequences {pi /3ni }nN and {ri /3mi }iN of rational numbers (pi 6= 0

(mod 3), ri 6= 0 (mod 3)) and {ui }iN of natural numbers such that for i 1,

if hi = ui , i = (pi /3ni , ri /3mi ), Hi = h0 hi , and fi = Hi Ri Hi1 , then

1. Ri1 hi = hi Ri1 ;
2. i = i1 + (3ni , 3mi );
3. d0 (Hi1 , Hi ) < 1/2i ;
4. d (fi1 , fi ) < /2i+1 ;
5. mi 2i > ni > mi1 + i;
6. ui ui1 + 3.

Assume that we have constructed such sequences up to step i = k. Since

d0 (n , id) 0 as n , there is N such that d0 (Hk n , Hk ) < 2(k+1) for
all n N . Letting uk+1 = max{mk , uk + 3, N }, properties (1), (3) and (6)

hold.

Now note that for any R2 ,

1
1
1
1
Hk+1 Rk + Hk+1
= Hk hk+1 R Rk h1
k+1 Hk = Hk (hk+1 R hk+1 )Rk Hk

so that there is  > 0 such that if || < , then

1
d (Hk+1 Rk + Hk+1
, fk ) < /2k+1 .

Let M be such that 3M < , and define

nk+1 = max{mk + (k + 1), M, uk } + 1,
mk+1 = nk+1 + i, and k+1 as in (2). It is easy to see that the remaining
properties hold, and the recursion step is complete.
56

Claim 1. There exist f H , h H0 and R2 such that fi f in

the C topology, Hi h in the C 0 topology, and i . Moreover, R is

minimal, f = hR h1 , and d (f, R ) < .

Proof. The existence of f , h and follows by completeness of the respective spaces, since the corresponding sequences are Cauchy by construcBy continuity of composition in the C 0 topology, f = hR h1 .
P
(i+1) = /2 and d (f , R ) < /2, we have
Since d (f0 , fi )
0

i=1 2
tion.

d (R , f ) < . Finally, R is minimal because, by the conditions we re-

quired on the sequences mi and ni , Lemma 4.7 implies that the two coordinates of are rationally independent.

Image of a vertical line by H1

Image of the vertical foliation by h

Figure 4.1: The conjugation h

Claim 2. Let F O (T2 ) be the map lifted by f . Then, any F -invariant

codimension 1 foliation of T2 has leaves which are not C 1 .

Proof. We first note the simple fact that any R -invariant codimension 1
foliation of T2 lifts to a foliation of the plane by straight lines. Thus, in
order to prove the claim, it suffices to show that there is a point p such
that any straight line containing p is mapped by h to a curve which is not
differentiable. We will see that this is the case for p = (0, 0).
57

Since the origin is fixed by n for all n, it is fixed by h as well. Let

rk = 3(ui +1) /2 and Bi = B(0, rk ), and let L R2 be a straight line

containing the origin. We will find two sequences of points in L, {xk }kN

and {yk }kN , both converging to the origin, such that (h(xk ), h(yk )) is

uniformly away from 0 or . This will imply that h(L) does not have a well-

defined tangent direction at the origin. Let L0 be one of the components of

L {(0, 0)}.

Note that Hk1 h = limi Hk1 Hi = limi hk+1 hk+2 hi , so that

d0 (Hk1 h, id) = lim d0 (Hk1 Hi , id)
i

3(uk+1 1)
2

j=k+1

3(uk +2)
2

d0 (hj , id)

3uj

j=k+1

= rk /3.

Thus, if zk = L0 Bk , we have that Hk1 h(zk ) zk rk /3. This implies

that (Hk1 h(zk ), zk ) 1/3. For 1 i k, hi is a rotation by /2 on

Bi Bk , so that Hk is a rotation by k/2 on Bk ; thus when k = 1 (mod

4), we have
|(h(zk ), zk ) /2| < 1/3.
By the same argument, if k = 0 (mod 4),
|(h(zk ), zk )| 1/3.
Letting xi = z4i+1 and yi = z4i , observing that |(xi , yi )| = 0, we have two
sequences of points of L converging to the origin, such that
|(h(xi ), h(yi )) /2| 2/3 < /2.
This means that the above angle is uniformly away from 0 or . If h(L)
were differentiable, we would have |(h(xi ), h(xi ))| 0 (or ). Thus, we

just showed that h(L) is not differentiable. This completes the proof of the

theorem.

58

Lemma 4.7. Let {mi }iN and {ni }iN be sequences of natural numbers such

that ni < mi < ni+1 for all i and

lim mi ni = lim ni+1 mi = .

Then, for any integer q > 1, the numbers =

are rationally independent.

i=0 q

ni

and =

i=0 q

mi

P
Proof. Suppose x + y = z for integers x, y, z. Let k = ki=0 q ni and
P
k = ki=0 q mi . Note that k = pk /q nk for some pk relatively prime to

q and similarly k = rk /q mk . We may write y = q b y 0 /a for some natural

numbers b, a and some integer y 0 coprime with q. Then,

aq mk b (xk + yk z) = q mk nk b (axpk aq nk z) + y 0 rk .
If k is large enough we have mk nk > b, and since y 0 rk is coprime with q,

it follows from the above that aq mk b (xk + yk z) is an integer coprime

with q, so that

a|xk + yk z| q mk +b .
On the other hand,
|xk + yk z| = |x(k ) + y(k )|
|x|
|x|

ni

i=k+1
q nk+1 +1

q1

+ |y|

+ |y|

q mi

i=k+1
q mk+1 +1

q1

2(|x| + |y|)q nk+1 .

Thus if k is large enough we have

2a(|x| + |y|)q nk+1 a|xk + yk z| q mk +b ,
so that
q nk+1 mk 2a(|x| + |y|)q b
which is a contradiction.
59

Bibliography
[AK70]

D. Anosov and A. Katok, New examples in smooth ergodic theory.

Ergodic diffeomorphisms., Transactions of the Moscow Mathematical Society 23 (1970), 135.

[BCLP04] F. Beguin, S. Crovisier, F. LeRoux, and A. Patou, Pseudorotations of the closed annulus: variation on a theorem of J.
Kwapisz, Nonlinearity 17 (2004), no. 4, 14271453.
[BH92]

M. Bestvina and M. Handel, An area preserving homeomorphism

of T2 that is fixed point free but does not move any essential simple closed curve off itself, Ergodic Theory & Dynamical Systems
12 (1992), 673676.

[Bir25]

G. Birkhoff, An extension of Poincares last geometric theorem,

Acta Mathematica 47 (1925), 297311.

[Bro12]

L. E. J. Brouwer, Beweis des Ebenen Translationssatzes, Mathematische Annalen 72 (1912), 3754.

[Eps66]

D. B. A. Epstein, Curves on 2-manifolds and isotopies, Acta

Mathematica 115 (1966), 83107.

[Fat87]

A. Fathi, An orbit closing proof of Brouwers lemma on translation arcs, LEnseignement Mathematique 33 (1987), 315322.

[FH77]

A. Fathi and M. Herman, Existence de diffeomorphismes minimaux, Asterisque 49 (1977), 3759.

60

[FK04]

B. Fayad and A. Katok, Constructions in elliptic dynamics, Ergodic Theory & Dynamical Systems 24 (2004), 14771520.

[Fra88]

J. Franks, Recurrence and fixed points of surface homeomorphisms, Ergodic Theory & Dynamical Systems 8* (1988), 99
107.

[Fra89]

, Realizing rotation vectors for torus homeomorphisms,

Transactions of the American Mathematical Society 311 (1989),
no. 1, 107115.

[Fra92]

, A new proof of the Brouwer plane translation theorem,

Ergodic Theory & Dynamical Systems 12 (1992), 217226.

[Fra95]

, The rotation set and periodic points for torus homeomorphisms, Dynamical Systems and Chaos (Aoki, Shiraiwa, and
Takahashi, eds.), World Scientific, Singapore, 1995, pp. 4148.

[Gui94]

L. Guillou, Theor`eme de translation plane de Brouwer et

generalisations du theor`eme de Poincare-Birkhoff, Topology 33
(1994), 331351.

[Gui06]

, Free lines for homeomorphisms of the open annulus,

Preprint, 2006.

[Her83]

M. R. Herman, Une methode pour minorer les exposants de

Lyapunov et quelques exemples montrant le caract`ere local dun
theor`eme dArnold et de Moser sur le tore de dimension 2, Commentarii Mathematici Helvetici 58 (1983), 453502.

[Her92]

, On the dynamics of Lagrangian tori invariant by symplectic diffeomorphisms, Progress in Variational Methods in
Hamiltonian Systems and Elliptic Equations (LAquila, 1990),
Longman Science and Technology, Harlow, 1992, (Pitman Research Notes Mathematical Series, 243), pp. 92112.

[HW90]

G. Hardy and E. Wright, An introduction to the theory of numbers, fifth ed., Oxford Science Publications, 1990.
61

[JZ98]

L Jonker and L Zhang, Torus homeomorphisms whose rotation

sets have empty interior, Ergodic Theory & Dynamical Systems
18 (1998), 11731185.

[Ker23]

B. de Kerekj
art
o, Vorlesungen u
ber Topologie (I), Berlin:
Springer, 1923.

[Ker29]

, The plane translation theorem of Brouwer and the last

geometric theorem of poincare, Acta Sci. Math. Szeged 4 (192829), 86102.

[Kwa02]

J. Kwapisz, A priori degenracy of one-dimensional rotation sets

for periodic point free torus maps, Transactions of the American
Mathematical Society 354 (2002), no. 7, 28652895.

[LC05]

P. Le Calvez, Une version feuilletee equivariante du theor`eme de

translation de Brouwer, Publications Mathematiques de LIHES

102 (2005), no. 1, 198.

[LM91]

J Llibre and R.S Mackay, Rotation vectors and entropy for homeomorphisms of the torus isotopic to the identity, Ergodic Theory
& Dynamical Systems 11 (1991), 115128.

[MZ89]

M. Misiurewicz and K. Ziemian, Rotation sets for maps of tori,

Journal of the London Mathematical Society 40 (1989), no. 2,
490506.

[MZ91]

, Rotation sets and ergodic measures for torus homeomorphisms, Fundamenta Mathematicae 137 (1991), 4452.

[Ros82]

J. G. Rosenstein, Linear Orderings, Academic Press, New York,

1982.

62