Regular pro jectively Anosov flows on three manifolds, , November 27, 2006

Keywords: projectively Anosov system, conformally Anosov systems, bi-contact structures MSC2000: Primary 37D30, Secondary 57R30

1 Introduction

Codimension one foliations and contact structures play important roles in the study of topology and geometry of three dimensional manifolds. In [9] , Eliashberg and Thurston combined the theories of these objects together as that of confoliations. One of the fundamental results is that any codimension one foliation on a three dimensional manifold except

ℱ = {S^{2} \times {*}}

S^{2} \times S^{1}

can be perturbed into a positive (or negative) contact structure as a plane field.

They also introduced a special class of perturbations of foliations, so called linear perturbations. Suppose a foliation generated by a plane field

ξ

. A linear perturbation of

ξ

is a one parameter family

{Ker α_{t}}_{t \in (- ε, ε)}

of plane fields defined by a family of 1-forms

{α_{t}}

with

ξ = Ker α_{0}

and

(d / d t) (α_{t} \land d α_{t}) > 0

Eliashberg and Thurston observed that if the kernel of

β = d α_{t} / d t |_{t = 0}

also generates a foliation, then

(Ker (α + t β), Ker (α - t β))

is a pair of mutually transverse positive and negative contact structures for any

t \neq 0

. Independently, Mitsumatsu [13] also studied the same deformation for invariant foliations of Anosov flows and he called such a pair of contact structures a bi-contact structure.

Mitsumatsu, and Eliashberg and Thurston observed that bi-contact structures correspond to projectively Anosov flows (or conformally Anosov flows in [9] ), which are the main objects of this paper.

A flow

Φ = {Φ^{t}}_{t \in R}

on a three dimensional manifold

M

is called a projectively Anosov flow (or

P A

flow) if it has no stationary points and admits a decomposition

T M = E^{u} + E^{s}

by continuous plane fields such that

∙ $E^{u} (z) \cap E^{s} (z) = T Φ (z)$ for any $z \in M$ , where $T Φ$ is the line field tangent to the orbits of $Φ$ ,
∙ $D Φ^{t} (E^{σ} (z)) = E^{σ} (Φ^{t} (z))$ for any $σ \in {u, s}$ , $z \in M$ , and $t \in R$ , and
∙ there exist two constants $C > 0$ and $λ \in (0, 1)$ such that
$∥ D {\hat{Φ}}^{t} |_{(E^{s} / T Φ) (z)} ∥ \cdot ∥ (D {\hat{Φ}}^{t} |_{(E^{u} / T Φ) (z)})^{- 1} ∥ \leq C λ^{t}$
for any $z \in M$ and $t \geq 0$ , where $D \hat{Φ} = {D {\hat{Φ}}^{t}}_{t \in R}$ is the induced flow on $T M / T Φ$ .

We call the decomposition

T M = E^{u} + E^{s}

P A

splitting. If it satisfies stronger inequalities

∥ D {\hat{Φ}}^{t} |_{(E^{s} / T Φ) (z)} ∥ \leq C λ^{t}, ∥ (D {\hat{Φ}}^{t} |_{(E^{u} / T Φ) (z)})^{- 1} ∥ \leq C λ^{t}

for any

z \in M

and

t \geq 0

, then the flow is called an Anosov flow and the splitting is called a weak-Anosov splitting ^$1$ . We remark that a variant of a

P A

splitting localized at a flow invariant set is called a dominated splitting, which plays important roles in the modern theory of dynamical systems. See [4] for example.

Any

P A

splitting is integrable, however, is not smooth in general. A

P A

flow (or an Anosov flow) with a

C^{r}

-smooth

P A

splitting is called

C^{r}

-regular. When a

P A

flow is

C^{\infty}

-regular, we simply say it is regular. From the viewpoint of deformations of foliations, regular

P A

flows correspond to linear deformations

{Ker α_{t}}

of a foliation such that the derivative

d α_{t} / d t |_{t = 0}

generates another foliation.

Regular Anosov flows on three dimensional manifolds are completely classified by Ghys.

Theorem 1.1 ([10] ). Up to finite covering, any regular Anosov flow on a three dimensional closed manifold is smoothly equivalent to either the suspension flow of a two dimensional hyperbolic toral automorphism or a quasi-Fuchsian flow on the unit tangent bundle of closed surface of genus greater than one.

It is natural to ask whether the similar classification exists for regular

P A

flows or not. In [16] , Noda gave a classification of regular

P A

flows with an invariant torus on a

T^{2}

-bundle over

S^{1}

. After that, he and Tuboi gave a classification for certain manifolds, which can be summarized as follows.

Theorem 1.2 ([16] ,[17] ,[18] , and [21] ). Any regular

P A

flow on a Seifert manifold or a

T^{2}

-bundle over

S^{1}

must be either an Anosov flow or represented as a finite union of

T^{2} \times I

-models.

Roughly speaking, a

T^{2} \times I

-model is a flow on

T^{2} \times [0, 1]

which is transverse to

T^{2} \times {z}

for any

z \in (0, 1)

and is equivalent to a linear flow on each boundary.

See [16] for the precise definition. The author also approached the classification from another direction. In [2] , he showed that any regular

P A

flow on any closed three dimensional manifold without non-hyperbolic periodic orbits is equivalent to one of the flows that they classified.

In [17] , Noda conjectured that there are no other regular

P A

flows. The main theorem of this paper gives a solution of this conjecture and classify three dimensional regular

P A

flows completely.

Main Theorem. Any

C^{2}

-regular

P A

flow on a closed and connected there dimensional manifold must be either an Anosov flow or represented as a finite union of

T^{2} \times I

models.

The theorem gives an answer to a conjecture posed by Mitsumatsu (Conjecture 4.3.3 in [14] ) immediately.

Corollary 1.3. Any bi-contact structure associated with a regular

P A

flow consists of tight contact structures.

The proof is divided into two parts. In Section 2 , we show a dichotomy on dynamics of regular

P A

flows. Namely, either the set of periodic orbits is dense in the manifold, or any positive or negative orbit converges to an invariant torus with rotational dynamics. We can see that the latter implies that the flow is represented by

T^{2} \times I

-models. In Sections 3 and 4 , we show the former implies that the flow is Anosov. It is done by proving the hyperbolicity of all periodic orbits.

^$1$ It is different from but equivalent to the common definition of an Anosov flow as pointed out by Doering [8,Proposition1.1] .

Acknowledgment

The author would like to thank Prof. Takashi Inaba, who pointed out that we can apply the stability theory and the level theory of Cantwell and Conlon in our proof.

2 A dichotomy on dynamics

We fix a

C^{2}

-regular

P A

flow

Φ

on a closed and connected three dimensional manifold

M

. Let

T M = E^{u} + E^{s}

be a

P A

splitting associated to

Φ

and

ℱ^{ρ}

the foliations generated by

E^{ρ}

for

ρ \in {u, s}

. Without loss of generality, we can assume that both

ℱ^{u}

and

ℱ^{s}

are transversely orientable. For a compact

Φ

-invariant set

Λ

, we define the stable set

W^{s} (Λ)

and the unstable set

W^{u} (Λ)

\begin{matrix} W^{s} (Λ) & = & {z \in M | {lim}_{t \to + \infty} d (Φ^{t} (z), Λ) \to 0} \end{matrix}

\begin{matrix} W^{u} (Λ) & = & {z \in M | {lim}_{t \to - \infty} d (Φ^{t} (z), Λ) \to 0} . \end{matrix}

For

ρ \in {u, s}

, let

Ω_{*}^{ρ} (Φ)

be the union of all closed leaves of

ℱ^{ρ}

on which the flow

Φ

is topologically conjugate to a linear flow. Remark that

Ω_{*}^{u} (Φ)

is a finite union of

Φ

-invariant tori and

W^{s} (Ω_{*}^{u} (Φ))

is an open neighborhood of

Ω_{*}^{u} (Φ)

Similarly,

Ω_{*}^{s} (Φ)

is a finite union of

Φ

-invariant tori and

W^{u} (Ω_{*}^{s} (Φ))

is an open neighborhood of

Ω_{*}^{s} (Φ)

For a foliation

G

M

, we denote the leaf of a foliation through a point

z \in M

G (z)

. We also denote the orbit

{Φ^{t} (z) | z \in R}

of a point

z \in M

O (z)

and the set of periodic points of

Φ

Per (Φ)

. We say a periodic point

z_{0}

s

-regular when there exists an embedded compact annulus

A \subset ℱ^{s} (z_{0})

such that

Φ^{t} (A) \subset Int A

for any

t > 0

and

\cap_{t > 0} Φ^{t} (A) = O (z_{0})

. Similarly, we say a periodic point

z_{0}

u

-regular when there exists an embedded compact annulus

A \subset ℱ^{u} (z_{0})

such that

Φ^{- t} (A) \subset Int A

for any

t > 0

and

\cap_{t > 0} Φ^{- t} (A) = O (z_{0})

We also say

z_{0}

ρ

-irregular if

z_{0}

is not

ρ

-regular for

ρ = u, s

The aim of this section is to prove the following proposition.

Proposition 2.1. Either one of the followings hold:

1. $M = W^{s} (Ω_{*}^{u} (Φ)) \cup Ω_{*}^{s} (Φ) = W^{u} (Ω_{*}^{s} (Φ)) \cup Ω_{*}^{u} (Φ)$ .
2. $M = \bar{Per (Φ)}$ and any periodic point of $Φ$ is $s$ and $u$ -regular.

It is not hard to show that the former implies that

Φ

is equivalent to one of known models. Namely,

Proposition 2.2. If

M = W^{s} (Ω_{*}^{u} (Φ)) \cup Ω_{*}^{s} (Φ) = W^{u} (Ω_{*}^{s} (Φ)) \cup Ω_{*}^{u} (Φ)

, then

Φ

is represented by a finite union of

T^{2} \times I

-models.

Proof. Fix a connected component $T_{0}$ of $Ω_{*}^{s} (Φ)$ and a connected component $U$ of $W^{u} (T_{0}) \ T_{0}$ . Take a subset $B$ of $W^{u} (T_{0})$ which is diffeomorphic to $T^{2} \times [0, 1]$ so that $T_{0} \subset \partial B$ and $B \ T_{0} \subset U$ . Let $T_{*}$ be the component of $\partial B$ different from $T_{0}$ .
Notice that $W^{s} (Ω_{*}^{u} (Φ))$ is the disjoint union of the stable sets of the connected components of $Ω_{*}^{u} (Φ)$ . Since $T_{*}$ is connected and contained in $W^{s} (Ω_{*}^{u} (Φ))$ , we have $T_{*} \subset W^{s} (T_{1})$ for some connected component $T_{1}$ of $Ω_{*}^{u} (Φ)$ .
Take a neighborhood $B_{*}$ of $T_{1}$ which is diffeomorphic to $T^{2} \times [0, 1]$ so that $B_{*} \subset W^{s} (T_{1})$ . Then, we have $Φ^{t_{0}} (T_{*}) \subset Int B_{*}$ for some $t_{0} > 0$ . Since $Φ^{t_{0}} (T_{*})$ separates two boundary components of $B_{*}$ in $B_{*}$ , it must be incompressible in $B_{*}$ . In particular, it is isotopic to $T_{1}$ in $B_{*}$ It implies that there exists a subset $B_{1}$ of $M$ which is diffeomorphic to $T^{2} \times [0, 1]$ , and satisfies $\partial B_{1} = T_{0} \cup T_{1}$ and $Int B_{1} = U$ . Inductively, we can take sequences $(T_{n})_{n \geq 0}$ and $(B_{n})_{n \geq 0}$ so that $T_{n}$ is a connected component of $Ω_{*}^{u} (Φ) \cup Ω_{*}^{s} (Φ)$ , $B_{n}$ is a subset of $M$ which is diffeomorphic to $T^{2} \times [0, 1]$ , $\partial B_{n} = T_{n} \cup T_{n + 1}$ , and $B_{n} \cap B_{n + 1} = T_{n + 1}$ for any $n$ . Since $Ω_{*}^{u} (Φ)$ and $Ω_{*}^{s} (Φ)$ contain only finitely many tori, we have $T_{n} = T_{m}$ for some $n \neq m$ .
It implies that $M$ is a $T^{2}$ -bundle over $S^{1}$ . By Noda's classification [16] , $Φ$ is represented by a finite union of $T^{2} \times I$ -models.

2.1 Return maps

We introduce the concept of return maps. For a finite set

Σ

, let

π_{x}

and

π_{y}

be the projections from

[- 2, 2]^{2} \times Σ

to the first and the second components respectively. We say a subset

R

[- 2, 2]^{2} \times Σ

is a rectangle if it has the form

[x_{-}, x_{+}] \times [y_{-}, y_{+}] \times σ_{0}

We call a

C^{2}

-embedding

ψ : [- 2, 2]^{2} \times Σ \to M

with a finite set

Σ

a canonical cross section if

∙ $Im ψ$ is transverse to $T Φ$ ,
∙ $ψ ([- 2, 2] \times y \times σ) \subset ℱ^{s} (ψ (x, y, σ))$ and $ψ (x \times [- 2, 2] \times σ) \subset ℱ^{u} (ψ (x, y, σ))$ for any $(x, y, σ) \in [- 2, 2]^{2} \times Σ$ , and
∙ both ${Φ^{t} (z) | t > 0}$ and ${Φ^{t} (z) | t < 0}$ intersect with $ψ ((- 1, 1)^{2} \times Σ)$ for any $z \in M$ .

It is easy to see that the flow

Φ

admits such an embedding.

Fix a canonical cross section

ψ : [- 2, 2]^{2} \times Σ \to M

. We call a

C^{2}

-diffeomorphism

r : R \to R^{'}

between two rectangles

R

and

R^{'}

a return associated to

(Φ, ψ)

if there exists a positive valued continuous function

τ

R

such that

Φ^{τ (w)} (ψ (w)) = ψ \circ r (w)

for any

w \in R

. The function

τ

is called the return time associated to

R

. Note that

τ

is uniquely determined since any return of a

P A

flow cannot be the identity map. For a return

r : R \to R^{'}

, we can take

C^{2}

-diffeomorphisms

r_{x} : π_{x} (R) \to π_{x} (R^{'})

and

r_{y} : π_{y} (R) \to π_{y} (R^{'})

so that

r (x, y, σ) = (r_{x} (x), r_{y} (y), σ^{'})

for any

(x, y, σ) \in R

. We call the pair

(r_{x}, r_{y})

the

x y

-decomposition of

r

. For a return

r

associated to

(Φ, ψ)

, the map

r^{- 1}

is a return associated to

(Φ^{- 1}, ψ)

For a family

ℛ = {r_{k}}_{k = 1}^{k_{*}}

of returns, we write

ℛ^{- 1}

for a family

{r_{k}^{- 1}}_{k = 1}^{k_{*}}

of returns associated to

(Φ^{- 1}, ψ)

We say a family

ℛ = {r_{k} : R_{k} \to R_{k}^{'}}_{k = 1}^{k_{*}}

of returns is full when

∙ $[- 1, 1]^{2} \times Σ$ is contained in both $\cup_{k = 1}^{k_{*}} R_{k}$ and $\cup_{k = 1}^{k_{*}} R_{k}^{'}$ , and
∙ there exists a constant $Δ > 0$ such that if $w \in R_{k} \cap ([- 1, 1]^{2} \times Σ)$ satisfies $r_{k} (w) \in [- 1, 1]^{2} \times Σ$ , then $Q_{Δ} (w) \subset R_{k}$ and $Q_{Δ} (r_{k} (w)) \subset R_{k}^{'}$ , where $Q_{Δ} (x, y, σ) = [x - Δ, x + Δ] \times [y - Δ, y + Δ] \times σ$ .

It is easy to see that any canonical cross section admits a full family of returns.

Fix a full family

ℛ = {r_{k} : R_{k} \to R_{k}^{'}}_{k = 1}^{k_{*}}

of returns associated to

(Φ, ψ)

For a subset

Λ

[- 2, 2]^{2} \times Σ

, we call a sequence

(k (n))_{n = 1}^{n_{*}}

ℛ

-admissible sequence for

Λ

r_{k (n)} \circ \dots \circ r_{k (1)} |_{Λ}

is well-defined for any

n = 1, 2, \dots, n_{*}

We say an

ℛ

-admissible sequence

(k (n))_{n = 1}^{n_{*}}

for a point

w

[- 1, 1]^{2} \times Σ

is fine when

r_{k (n)} \circ \dots \circ r_{k (1)} (w) \subset [- 1, 1]^{2} \times Σ

for any

n = 1, 2, \dots, n_{*}

For

Δ_{1} > 0

, we say an

ℛ

-admissible sequence

(k (n))_{n = 1}^{n_{*}}

for an interval

I

(ℛ, Δ_{1})

-admissible if

| r_{k (n)} \circ \dots \circ r_{k (1)} (I) | \leq Δ_{1}

, where

| J |

is the length of an interval

J

. We call a sequence

(I_{i} = [x_{i}, x_{i}^{'}] \times y_{i} \times σ_{i})_{i \geq 1}

of intervals in

[- 2, 2]^{2} \times Σ

Δ_{1}

-family if there exists a family

{(k_{i} (n))_{n = 1}^{n_{i}}}_{i \geq 1}

of sequences such that

(k_{i} (n))_{n = 1}^{n_{i}}

is an

(ℛ, Δ_{1})

-admissible sequence for

I_{i}

for any

i \geq 1

n_{i}

tends to infinity as

i \to \infty

, and

limsup | r_{k_{i} (n_{i})} \circ \dots r_{k_{i} (1)} (I_{i}) | > 0

. The following is the keystone to control the topology of the stable and unstable foliations.

Lemma 2.3. There exists a constant

Δ_{1} > 0

such that any

Δ_{1}

-family

{I_{i}}_{i = 1}^{\infty}

of intervals admits a sequence

{z_{i} \in ψ (I_{i})}_{i = 1}^{\infty}

accumulating to a point of

Ω_{*}^{u} (Φ)

or an

s

-irregular periodic point.

Proof. Notice that almost all arguments in the proof of Proposition 3.1 of [19] (or Proposition 4.2 of [1] ) work even if non-hyperbolic periodic orbits exist. They allow us to take a constant $Δ_{1} > 0$ such that if an interval $I = [x, x^{'}] \times y \times σ$ admits an $(ℛ^{- 1}, Δ_{1})$ -admissible sequence $(k (n))_{n = 1}^{\infty}$ then $ψ (I) \subset W^{u} (Ω_{*}^{s} (Φ))$ or $Int ψ (I) \cap W^{u} (O (z_{*})) \neq \emptyset$ for some $s$ -irregular periodic point $z_{*}$ .
Let $(I_{i})_{i \geq 1}$ be a $Δ_{1}$ -family of intervals and ${(k_{i} (n))_{n = 1}^{n_{i}}}_{i \geq 1}$ the corresponding family of sequences. Put $J_{i} = r_{k_{i} (n_{i})} \circ \dots \circ r_{k_{i} (1)} (I_{i})$ and $k_{i}^{'} (n) = k (n_{i} - n + 1)$ for any $n = 1, \dots, n_{i}$ . Then, $(k_{i}^{'} (n))_{n = 1}^{n_{i}}$ is an $(ℛ^{- 1}, Δ_{1})$ -admissible sequence for $J_{i}$ . By taking subsequences if it is necessary, we can assume that $J_{i}$ converges to an interval $J_{*} = [\bar{x}, {\bar{x}}^{'}] \times \bar{y} \times \bar{σ}$ and there exist sequences $(k^{'} (n))_{n = 1}^{\infty}$ and $(i_{n})_{n \geq 1}$ such that $i_{n}$ tends to infinity as $n \to \infty$ and $k_{i}^{'} (n) = k^{'} (n)$ for any $n \geq 1$ and $i = 1, \dots, i_{n}$ . It is easy to check that $(k^{'} (n))_{n = 1}^{\infty}$ is an $(ℛ^{- 1}, Δ_{1})$ -admissible sequence for $J_{*}$ By the choice of the constant $Δ_{1}$ , there exists $x_{*} \in (\bar{x}, {\bar{x}}^{'})$ such that $ψ (x_{*}, \bar{y}, \bar{σ}) \in W^{u} (Ω_{*}^{s} (Φ))$ or $ψ (x_{*}, \bar{y}, \bar{σ}) \in W^{u} (O (z_{*}))$ for some $s$ -irregular periodic point $z_{*}$ .
Hence, we can take a neighborhood $U$ of $\bar{y}$ such that $\bar{\cup_{t > T} Φ^{- t} (x_{*} \times U \times \bar{σ})}$ converges to a connected component of $Ω_{*}^{s} (Φ)$ or $O (z_{*})$ as $T \to \infty$ . It follows the lemma immediately.

2.2 Local dynamics at periodic points

Put

{Per}_{h} (Φ) = Per (Φ) \ {Ω_{*}^{u} (Φ) \cup Ω_{*}^{s} (Φ)}

. The main aim of this subsection is to show that any point of

{Per}_{h} (Φ)

u

and

s

-regular. It is a main step of the proof of Proposition 2.1 , and is done by a variant of the argument in [2] .

Fix a canonical cross section

ψ : [- 2, 2]^{2} \times Σ \to M

. For a periodic point

z_{0} = ψ (x_{0}, y_{0}, σ_{0})

, we call a return

r : R \to R^{'}

the first return of

z_{0}

(x_{0}, y_{0}, σ_{0}) \in Int R \cap Int R^{'}

r (x_{0}, y_{0}, σ_{0}) = (x_{0}, y_{0}, σ_{0})

, and the return time

τ_{R}

satisfies

τ_{R} (w) = inf {t > 0 | Φ^{t} \circ ψ (w) \in R^{'}}

for any

w \in R

We say a point

z

of a topological space

X

is accessible from a subset

A

X

when there exists a continuous map

l : [0, 1] \to X

such that

l (1) = z

and

l (t) \in A

for any

t \in [0, 1)

Lemma 2.4. Let

z_{0}

and

z_{1}

be periodic points of

Φ

and suppose that

z_{1}

is accessible from

W^{s} (O (z_{0})) \cap ℱ^{ρ} (z_{1})

for

ρ \in {u, s}

. Then, there exists an embedded closed annulus

A \subset ℱ^{ρ} (z_{0})

satisfying

\partial A = O (z_{0}) \cup O (z_{1})

and

Int A \subset W^{s} (O (z_{0})) \cap W^{u} (O (z_{1}))

. In particular, we have

ℱ^{ρ} (z_{0}) = ℱ^{ρ} (z_{1})

Proof. Without loss of generality, we can assume that $z_{0} = ψ (x_{0}, y_{0}, σ_{0})$ for some $(x_{0}, y_{0}, σ_{0}) \in [- 1, 1]^{2} \times Σ$ . We prove the lemma for the case $ρ = u$ since the proof for the other case is similar. Let $r : R \to R^{'}$ be the first return map of $z_{0}$ and $τ$ the return time associated to $r$ . Put $V = {Φ^{t} \circ ψ (w) | w \in R, t \in [0, τ (w)]}$ and let $G$ be the restriction of $ℱ^{ρ}$ on $V$ . It is easy to see that $G$ is diffeomorphic to the foliation ${x \times π_{y} (R) \times [0, 1] / \sim}_{x \in π_{x} (R)}$ on $(π_{x} (R) \times π_{y} (R) \times [0, 1]) / (x, y, 1) \sim (r_{x} (x), r_{y} (y), 0)$ . Hence, a leaf of $G$ is non-contractible if and only if it contains a periodic point of $Φ$ . See Figure 1 .
Figure 1 : A neighborhood $V$
Since $z_{1}$ is accessible from $W^{s} (O (z_{0}))$ , there exist a simple closed curve $γ$ transverse to $T Φ$ and an embedded closed annulus $A_{1} \subset ℱ^{ρ} (z_{1})$ such that $\partial A_{1} = γ \cup O (z_{1})$ and $γ \cup Int A \subset W^{s} (O (z_{0}))$ . The curve $Φ^{t} (γ)$ is contained in a leaf $L$ of $G$ for any sufficiently large $t > 0$ . By Poincaré-Bendixon's theorem, $L$ must be non-contractible, and hence, it contains a periodic point. Since $γ \in W^{s} (O (z_{0}))$ , we obtain $L = G (z_{0})$ . It implies that the existence of the required embedded annulus.

For

w = (x, y, σ) \in [- 1, 1]^{2} \times Σ

and

δ > 0

, we define intervals

I_{δ}^{s} (w)

and

I_{δ}^{u} (w)

I_{δ}^{s} (w) = [x - δ, x + δ] \times y \times σ, I_{δ}^{u} (w) = x \times [y - δ, y + δ] \times σ .

Lemma 2.5. Let

z_{0}

be an

s

-regular periodic point. Then, there exist a constant

δ_{z_{0}} > 0

such that any fine

ℛ

-admissible sequence

(k (n))_{n = 1}^{\infty}

for

w \in ([- 1, 1]^{2} \times Σ) \cap ψ^{- 1} (W^{s} (O (z_{0})))

satisfies

ψ (I_{δ}^{s} (r_{k (n)} \circ \dots \circ r_{k (1)} (w))) \subset W^{s} (O (z_{0}))

for some

n \geq 1

. In particular,

ℱ^{u} (z) \cap W^{s} (O (z_{0}))

is an open subset of

ℱ^{u} (z)

for any

z \in W^{s} (O (z_{0}))

Proof. Fix $w_{0} = (x_{0}, y_{0}, σ_{0}) \in [- 1, 1]^{2} \times Σ$ so that $ψ (w_{0}) \in O (z_{0})$ . Let $r : R \to R^{'}$ be the first return of $w_{0}$ and $(r_{x}, r_{y})$ the $x y$ -decomposition of $r$ . Since $z_{0}$ is $s$ -regular, there exists $I_{x} \subset π_{x} (R)$ such that $r_{x} (I_{x}) \subset Int I_{x}$ and $\cap_{n \geq 0} r_{x}^{n} (I_{x}) = {x_{0}}$ . Put $Λ_{y} = {y \in \cap_{n \geq 0} r_{y}^{- n} ([- 2, 2]) | {lim}_{n \to \infty} r_{y}^{n} (y) = y_{0}}$ and take $δ > 0$ so that $[x - δ, x + δ] \subset I_{x}$ for any $x \in r_{x} (I_{x})$ . It is easy to see that $ψ (I_{x} \times Λ_{y} \times σ_{0}) \subset W^{s} (O (z_{0}))$ . In particular, $ψ (I_{δ}^{s} (w)) \subset W^{s} (O (z_{0}))$ for any $w \in r_{x} (I_{x}) \times Λ_{y} \times σ_{0}$ .
There exist a neighborhood $U_{0}$ of $O (z_{0})$ and a constant $T_{0} > 0$ such that $ψ (r_{x} (I_{x}) \times Λ_{y} \times σ_{0}) \neq \emptyset$ for any $z \in W^{s} (O (z_{0})) \cap \cap_{t \geq 0} Φ^{- t} (U_{0})$ . Hence, we can take $δ_{z} > 0$ so that $I_{δ_{z}}^{s} (w) \subset W^{s} (O (z_{0}))$ for any $w \in [- 1, 1]^{2} \times Σ$ with $ψ (w) \in W^{s} (O (z_{0})) \cap \cap_{t \geq 0} Φ^{- t} (U_{0})$ . It is easy to see that the constant $δ_{z_{0}}$ satisfies the required condition.

Lemma 2.6. The followings hold for any

s

-regular periodic point

z_{0}

1. $ℱ^{s} (z) \subset W^{s} (O (z_{0}))$ for any $z \in W^{s} (O (z_{0})) \ ℱ^{s} (z_{0})$ .
2. $ℱ^{s} (z_{0}) \cap W^{s} (O (z_{0}))$ is diffeomorphic to $S^{1} \times R$ .
3. If $ℱ^{s} (z_{0}) ⊄ W^{s} (O (z_{0}))$ , then there exist an $s$ -irregular periodic point $z_{1} \in ℱ^{s} (z_{0})$ and an embedded closed annulus $A \subset ℱ^{s} (z_{0})$ such that $\partial A = O (z_{0}) \cup O (z_{1})$ and $Int A \subset W^{s} (O (z_{0})) \cap W^{u} (O (z_{1})))$ .

Proof. Since $z_{0}$ is $s$ -regular, we can take an embedded closed annulus $A_{0} \subset ℱ^{s} (z_{0})$ such that $Φ^{t} (A_{0}) \subset Int A_{0}$ for any $t > 0$ and $\cap_{t > 0} Φ^{t} (A_{0}) = O (z_{0})$ .
Then, $W_{0} = \cup_{t > 0} Φ^{- t} (A_{0})$ is a connected component of $W^{s} (O (z_{0})) \cap ℱ^{s} (z_{0})$ which is diffeomorphic to $S^{1} \times R$ .
Fix a leaf $L$ of $ℱ^{s}$ with $L \cap W^{s} (O (z_{0})) \neq \emptyset$ and take a connected component $W$ of $L \cap W^{s} (O (z_{0}))$ . It is sufficient to show that if $W \neq L$ then there exists a periodic point $z_{1} \in L \ U$ which is accessible from $W$ . In fact, if such $z_{1}$ exists, then Lemma 2.4 implies that there exists an embedded closed annulus $A \subset ℱ^{s} (z_{0})$ with $\partial A = O (z_{0}) \cup O (z_{1})$ and $Int A \subset W^{s} (O (z_{0})) \cap W^{u} (O (z_{1})))$ In particular, $z_{1}$ is $s$ -irregular and $W = W_{0} \subset ℱ^{s} (z_{0})$ .
Suppose that $W \neq L$ . Then, there exists $z_{0} = ψ (x_{0}, y_{0}, σ_{0}) \in ψ ([- 1, 1]^{2} \times Σ) \cap (L \ W)$ which is accessible from $W$ . Put $I_{x} = [x_{0}, x] \times y_{0} \times σ_{0}$ for $x > x_{0}$ .
Without loss of generality, we can assume that $ψ (Int I_{x_{0}^{'}}) \subset W$ for some $x_{0}^{'} > x_{0}$ .
Fix a full family $ℛ = {r_{k} : R_{k} \to R_{k}^{'}}_{k = 1}^{k_{*}}$ of returns and let $Δ_{1} > 0$ and $δ_{z_{0}} > 0$ be the constants obtained in Lemmas 2.3 and 2.5 . Put $Δ = min {Δ_{1}, δ_{z_{0}}, x_{0}^{'} - x_{0}}$ and $C_{1} = sup {∥ D r_{k} ∥ | k = 1, \dots, k_{*}}$ . We claim that for any $x \in (x_{0}, x_{0} + Δ)$ , $I_{x} = [x_{0}, x] \times y_{0} \times σ_{0}$ admits an $(ℛ, Δ)$ -admissible sequence $(k_{x} (n))_{n = 1}^{n_{x}}$ such that $| r_{k_{x} (n_{x})} \circ r_{k_{x} (1)} (I_{x}) | \geq C_{1}^{- 1} Δ$ and $C_{1}^{n_{x}} (x - x_{0}) > Δ$ . In fact, take a fine $ℛ$ -admissible sequence $(k_{x} (n))_{n = 1}^{\infty}$ for $(x, y_{0}, σ_{0})$ . Then, $| r_{k_{x} (n^{'})} \circ r_{k_{x} (1)} (I_{x}) | \leq C_{1}^{n} (x - x_{0})$ if $(k (n))_{n = 1}^{n^{'}}$ is an $ℛ$ -admissible sequence for $I_{x}$ . Since $ψ (I_{Δ}^{s} (r_{k_{x} (n)} \circ r_{k_{x} (1)} (x))) \subset W^{s} (O (z_{0}))$ and $ψ (x) \notin W^{s} (O (z_{0}))$ , there exists $n_{x} \geq 1$ such that $(k (n))_{n = 1}^{n_{x}}$ is an $(ℛ, Δ)$ -admissible sequence for $I_{x}$ with $| r_{k_{x} (n_{x} + 1)} \circ r_{k_{x} (1)} (I_{x}) | \geq C_{1}^{- 1} Δ$ . It is easy to see the sequence $(k_{x} (n))_{n = 1}^{n_{x}}$ satisfies the required conditions.
By the above claim, ${I_{C_{1}^{- n} Δ}}_{i \geq 0}$ is a $Δ_{1}$ -family. Lemma 2.3 implies that $z_{0}$ is a point of $Ω_{*}^{s} (Φ)$ or an $s$ -irregular periodic point. If the former holds, then $ℱ^{s} (ψ (w_{0}))$ is contained in $Ω_{*}^{s} (Φ)$ . However, $Ω_{*}^{s} (Φ)$ does not intersects with $W^{s} (O (z_{0}))$ . Therefore, $ψ (w_{0})$ is an $s$ -irregular periodic point.

Recall that we say a leaf of a codimension one foliation is semi-proper when it does not accumulate to itself from at least one side. We also say a leaf is proper when it does not accumulate to itself from both sides.

Lemma 2.7. Let

G

be a

C^{2}

codimension one foliation of a closed three dimensional manifold. Then, any semi-proper leaf of

G

diffeomorphic to

S^{1} \times R

has trivial holonomy.

Proof. Let $L$ be a leaf of $G$ which is diffeomorphic to $S^{1} \times R$ . Note that the end set of $L$ consists of two points. By the level theory of Cantwell and Conlon [5] $L$ is either proper or contained in an exceptional local minimal set. However, Duminy's theorem (See [7] for the proof ) implies that the end of a semi-proper leaf in an exceptional local minimal set must be a Cantor set. Hence, the leaf $L$ is proper. By a theorem of Cantwell and Conlon [6,Theorem1] , $L$ has trivial holonomy.

Now, we show the main result of this subsection.

Proposition 2.8. Any point of

{Per}_{h} (Φ)

s

and

u

-regular.

Proof. We show that any $z_{0} \in {Per}_{h} (Φ)$ is $u$ -regular. Once it is done, then we apply it to the flow $Φ^{- 1} = {Φ^{- t}}$ and obtain that any $z_{0} \in {Per}_{h} (Φ)$ is $s$ -regular.
Suppose that $z_{0}$ is $u$ -irregular. If $ℱ^{s}$ has trivial holonomy along $O (z_{0})$ , then $Per (Φ) \cap ℱ^{u} (z_{0})$ contains a closed annulus, whose boundary consists of $u$ -irregular periodic points. Hence, we can assume that $ℱ^{s}$ has non-trivial holonomy along $O (z_{0})$ by replacing $z_{0}$ if it is necessary. Without loss of generality, we also assume $z_{0} = ψ (w_{0})$ for some $w_{0} = (x_{0}, y_{0}, σ_{0}) \in [- 1, 1]^{2} \times Σ$ .
Take the first return $r : R \to R^{'}$ of $z_{0}$ and let $(r_{x}, r_{y})$ be the $x y$ -decomposition of $r$ . Since $z_{0}$ is $u$ -irregular, it is $s$ -regular. Hence, we can assume that $I_{x} = π_{x} (R)$ satisfies $r_{x} (I_{x}) \subset Int I_{x}$ and $\cap_{n \geq 0} r^{n} (I_{x}) = {x_{0}}$ . Since $z_{0}$ is $u$ -irregular, we can take $y_{1} \in π_{y} (R) \ {y_{0}}$ such that $| r_{y} (y_{1}) - y_{0} | \leq | y_{1} - y_{0} |$ . It implies that there exists a compact interval $I_{y} \subset π_{y} (R)$ such that $r (I_{y}) \subset I_{y}$ and $y_{0} \in \partial I_{y}$ .
Put $J_{x} = ψ (I_{x} \times y_{0} \times σ_{0})$ and $J_{y} = ψ (x_{0} \times I_{y} \times σ_{0})$ .
Put $W = \cup_{t \geq 0} Φ^{- t} (J_{x})$ . Lemma 2.6 implies that $W = W^{s} (O (z_{0})) \cap ℱ^{s} (z_{0})$ and it is diffeomorphic to $S^{1} \times R$ . Since $\cup_{t > 0} Φ^{t} (Int J_{y}) \cap Im ψ \subset Int J_{y}$ , we have $W \cap Int J_{y} = \emptyset$ . If $ℱ^{s} (z_{0})$ coincide with $W$ , then it must be a semi-proper leaf of $ℱ^{s}$ . In particular, it has trivial holonomy by Lemma 2.7 . It contradicts the choice of $z_{0}$ . Therefore, we obtain $W \neq ℱ^{s} (z_{0})$ .
By Lemma 2.6 , there exist an $s$ -irregular periodic point $z_{1} \in ℱ^{s} (z_{0})$ and an embedded closed annulus $A^{s} \subset ℱ^{s} (z_{0})$ such that $\partial A^{s} = O (z_{0}) \cup O (z_{1})$ and $Int A^{s} \subset W^{s} (z_{0})$ . Let $t_{i}$ be the period of $z_{i}$ and put $λ_{i}^{ρ} = ∥ D Φ^{t_{i}} |_{(E^{ρ} / T Φ) (z_{i})} ∥$ for $i = 0, 1$ and $ρ \in {u, s}$ . Notice that the orientation of the orbits of $z_{0}$ and $z_{1}$ must be opposite since $z_{0}$ is $u$ -irregular, $z_{1}$ is $u$ -regular, and $ℱ^{s} (z_{0}) = ℱ^{s} (z_{1})$ . In particular, we have $λ_{0}^{u} \cdot λ_{1}^{u} = 1$ . See Figure 2 .
Figure 2 : Annuli $A^{s}$ and $A^{u}$
Since $z_{1}$ is $s$ -irregular and $Φ$ is a $P A$ flow, we have $1 \leq λ_{1}^{s} < λ_{1}^{u}$ , and hence, $λ_{0}^{s} < λ_{0}^{u} < 1$ . The latter implies that there exists a compact interval $I_{y}^{'} \subset π_{y} (R)$ such that $r_{y} (I_{y}^{'}) \subset Int I_{y}^{'}$ and $\cap_{n \geq 0} r_{y}^{n} (I_{y}^{'}) = {y_{0}}$ .
Put $J_{y}^{'} : = ψ (x_{0} \times (I_{y}^{'} \ {y_{0}}) \times σ_{0})$ . Then, we have $\cup_{t > 0} Φ^{t} (J_{y}^{'}) \cap Im ψ \subset J_{y}^{'}$ and $J_{y}^{'} \subset W^{s} (O (z_{0}))$ . The former implies $J_{y}^{'} \cap W = \emptyset$ , and hence, $J_{y}^{'} \subset W^{s} (O (z_{0})) \ ℱ^{s} (z_{0})$ .
Take an embedded annulus $A^{u} \subset ℱ^{u} (z_{1})$ such that $ℱ^{s} (z) \cap J_{y}^{'} \neq \emptyset$ for any $z \in A^{u} \ O (z_{1})$ . Then, Lemma 2.6 implies $A^{u} \ O (z_{1}) \subset W^{s} (O (z_{0}))$ . In particular, $z_{1}$ is accessible from $W^{s} (O (z_{0})) \cap ℱ^{u} (z_{1})$ . Applying Lemma 2.4 to $z_{0}$ , $z_{1}$ , and $ρ = u$ , we obtain an embedded annulus $A$ in $ℱ^{u} (z_{0}) = ℱ^{u} (z_{1})$ . such that $\partial A = O (z_{0}) \cap O (z_{1})$ . It implies $λ_{0}^{s} \cdot λ_{1}^{s} = 1$ since the orientation of $O (z_{0})$ and $O (z_{1}))$ in $A$ must be opposite. However, it contradicts the inequalities $λ_{0}^{u} \cdot λ_{1}^{u} = 1$ , $λ_{0}^{s} < λ_{0}^{u}$ , and $λ_{1}^{s} < λ_{1}^{u}$

Corollary 2.9. The leaf

ℱ^{ρ} (z_{0})

coincides with

W^{ρ} (O (z_{0}))

and it is diffeomorphic to

S^{1} \times R

for any

z_{0} \in {Per}_{h} (Φ)

and

ρ \in {u, s}

Proof. Since $z_{0}$ is $u$ and $s$ -regular, it is clear that $W^{s} (O (z_{0})) \subset ℱ^{s} (z_{0})$ . By Lemma 2.6 , non-existence of $s$ -irregular periodic point in ${Per}_{h} (Φ)$ implies that $ℱ^{s} (z_{0})$ is a subset of $W^{s} (O (z_{0}))$ and is diffeomorphic to $S^{1} \times R$ . The proof for $ℱ^{u} (z_{0})$ is the same.

2.3 Proof of Proposition 2.1

First, we show the Birkoff-Smale theorem in our setting.

Lemma 2.10.

ℱ^{s} (z_{0}) \cap ℱ^{u} (z_{0}) \subset \bar{{Per}_{h} (Φ)}

for any

z_{0} \in {Per}_{h} (Φ)

Proof. Take a canonical cross section $ψ : [- 2, 2]^{2} \times Σ \to M$ . Without loss of generality, we can assume that $z_{0} = ψ (w_{0})$ for some $w_{0} = (x_{0}, y_{0}, σ_{0}) \in [- 1, 1]^{2} \times Σ$ . Let $r : R \to R^{'}$ be the first return of $z_{0}$ and $(r_{x}, r_{y})$ the $x y$ -decomposition of $r$ .
Put $I_{x} = π_{x} (R)$ and $I_{y} = π_{y} (R)$ . Since $z_{0}$ is $u$ and $s$ -regular, we can assume that $r (I_{x}) \subset Int I_{x}$ , $I_{y} \subset Int r_{y} (I_{y})$ , $\cap_{n \geq 0} r^{n} (I_{x}) = {x_{0}}$ , and $\cap_{n \geq 0} r^{- n} (I_{y}) = {y_{0}}$ .
Fix $z_{1} \in ℱ^{s} (z_{0}) \cap ℱ^{u} (z_{0})$ . By Corollary 2.9 , we have $ℱ^{ρ} (z_{0}) = W^{ρ} (O (z_{0}))$ .
Hence, there exist $t_{-} < t_{+}$ , $x_{1} \in I_{x}$ , and $y_{1} \in I_{y}$ such that $Φ^{t_{-}} (z_{1}) = ψ (x_{0}, y_{1}, σ_{0})$ and $Φ^{t_{+}} (z_{1}) = ψ (x_{1}, y_{0}, σ_{0})$ . For any neighborhood $U \subset R$ of $ψ (x_{0}, y_{1}, σ_{0})$ , we can take a return $r_{1} : R_{1} \to R_{1}^{'}$ so that $(x_{0}, y_{1}, σ_{0}) \in Int R_{1} \subset U$ , $(x_{1}, y_{0}, σ_{0}) \in Int R_{2} \subset U$ , and $r_{1} (x_{0}, y_{1}, σ_{0}) = (x_{1}, y_{0}, σ_{0})$ . Let $(r_{1, x}, r_{1, y})$ be the $x y$ -decomposition of $r_{1}$ . We can see that $r^{n} (π_{x} (R_{1}^{'})) \subset π_{x} (R_{1})$ and $r^{- n} (π_{y} (R_{1})) \subset π_{y} (R_{1}^{'})$ for some $n \geq 1$ . See Figure 3 .
Figure 3 : Proof of Lemma 2.10
Then, there exists $(x_{*}, y_{*}) \in R_{1}$ such that $r_{x}^{n} \circ r_{1, x} (x_{*}) = (x_{*})$ and $r_{y}^{n} \circ r_{1, y} (y_{*}) = (y_{*})$ . Since $ψ (x_{*}, y_{*}, σ_{0})$ is a periodic point of $Φ$ and the neighborhood $U$ can be arbitrary small, we obtain that $z_{1} \in \bar{{Per}_{h} (Φ)}$ .

Now, we show Proposition 2.1 . If

{Per}_{h} (Φ) = \emptyset

, then Theorem B of [1] implies that the non-wandering set of

Φ

coincides with

Ω_{*}^{u} (Φ) \cup Ω_{*}^{s} (Φ)

. It is clear that

M = W^{s} (Ω_{*}^{u} (Φ)) \cup Ω_{*}^{s} (Φ) = W^{u} (Ω_{*}^{s} (Φ)) \cup Ω_{*}^{u} (Φ)

in this case.

Suppose

{Per}_{h} (Φ) \neq \emptyset

. By Lemma 2.10 , we have

ℱ^{u} (z_{0}) \cap ℱ^{u} (z_{0}) \subset \bar{{Per}_{h} (Φ)}

for any

z_{0} \in {Per}_{h} (Φ)

. Since

ℱ^{u}

and

ℱ^{s}

are mutually transverse, it implies that

ℱ^{s} (z) \cap ℱ^{u} (z) \subset \bar{{Per}_{h} (Φ)}

for any

z \in \bar{{Per}_{h} (Φ)}

We claim that

ℱ^{u} (z) \subset \bar{{Per}_{h} (Φ)}

for any

z \in \bar{{Per}_{h} (Φ)}

. If it does not hold, then there exists

z_{1} \in \bar{{Per}_{h} (Φ)}

which is accessible from

ℱ^{u} (z_{1}) \ \bar{{Per}_{h} (Φ)}

. It implies that

ℱ^{s} (z_{1})

is a semi-proper leaf. However, it contradicts Lemma 2.7 since

z_{1}

u

-regular.

Applying the claim for the flow

Φ^{- 1}

, we also have

ℱ^{s} (z) \subset \bar{{Per}_{h} (Φ)}

for any

z \in \bar{{Per}_{h} (Φ)}

. It implies that

\bar{{Per}_{h} (Φ)}

is a non-empty open subset of

M

, and hence,

M = \bar{{Per}_{h} (Φ)}

3 Markov families and the redcution to one dimensional dynamics

For

ρ \in {u, s}

, let

{Per}_{*}^{ρ} (Φ)

be the set of periodic point

z_{*}

with

∥ D {\hat{Φ}}^{t_{*}} |_{(E^{ρ} / T Φ) (z_{*})} ∥ = 1

, where

t_{*}

is the period of

z_{*}

. Put

{Per}_{*} (Φ) = {Per}_{*}^{u} (Φ) \cup {Per}_{*}^{s} (Φ)

In this section, we fix a

C^{2}

-regular

P A

flow

Φ

and assume that

M = \bar{Per (Φ)}

and any periodic point of

Φ

s

and

u

-regular. In Subsection 3.1 , we show

Φ

admits a kind of Markov partitions. Such a partition allows us to reduce the family of return maps to a one-dimensional dynamical system.

In Subsection 3.2 , we apply a theorem of Mann͂é to the reduced system and estimate

∥ D {\hat{Φ}}^{t} |_{E^{u} / T Φ} (z) ∥

. One of the consequences is that

Φ_{*} (Φ)

contains only finitely many periodic orbits. The other is the flow is Anosov if

{Per}_{*} (Φ)

is empty ^$2$ .

^$2$ It also follows from Theorem B of [1] .

3.1 Markov families of returns

Fix a canonical cross-section

ψ : [- 2, 2]^{2} \times Σ \to M

. Recall that

I_{δ}^{s} (w)

and

I_{δ}^{u} (w)

be intervals

[x - δ, x + δ] \times y \times σ

and

x \times [y - δ, y + δ] \times σ

respectively.

for

δ > 0

and

w = (x, y, σ) \in [- 1, 1]^{2} \times Σ

. The following lemma asserts that we can regard

I_{δ}^{s} (w)

and

I_{δ}^{u} (w)

as the local stable and the unstable manifolds for returns if

δ

is sufficiently small.

Lemma 3.1. Let

{r_{k}}_{k = 1}^{k_{*}}

be a full family of returns and

ε_{*} > 0

a given constant. There exist a constant

Δ_{ℛ} > 0

and a sequence

(ε_{n})_{n = 1}^{\infty}

which satisfy the followings:

1. $ε_{n} \in (0, ε_{*})$ for any $n \geq 1$ and tends to $0$ as $n \to \infty$ .
2. Any fine $ℛ$ -admissible sequence $(k_{n})_{n = 1}^{\infty}$ for $w \in [- 1, 1]^{2} \times Σ$ is also $ℛ$ -admissible for the interval $I_{Δ_{ℛ}}^{s} (w)$ and it satisfies $r_{k_{n}} \circ \dots \circ r_{k_{1}} (I_{Δ_{ℛ}}^{s} (w)) \subset I_{ε_{n}}^{s} (r_{k_{n}} \circ \dots r_{k_{1}} (w))$ for any $n \geq 1$ .
3. Any fine $ℛ^{- 1}$ -admissible sequence $(k_{n})_{n = 1}^{\infty}$ for $w \in [- 1, 1]^{2} \times Σ$ is also $ℛ^{- 1}$ -admissible for the interval $I_{Δ_{ℛ}}^{u} (w)$ and it satisfies $r_{k_{n}}^{- 1} \circ \dots \circ r_{k_{1}}^{- 1} (I_{Δ_{ℛ}}^{u} (w)) \subset I_{ε_{n}}^{u} (r_{k_{n}}^{- 1} \circ \dots r_{k_{1}}^{- 1} (w))$ for any $n \geq 1$ .

Proof. It is enough to show the existence of $Δ_{ℛ}$ and $(ε_{n})_{n \geq 1}$ which satisfies the first and the second conditions.
Let $Δ_{0}$ be the constant in the definition of a full family $ℛ$ of returns. Take $Δ_{1} > 0$ in Lemma 2.3 . Remark that there is no $Δ_{1}$ -family ${I_{i}}$ of intervals since any point of ${Per}_{h} (Φ)$ is $u$ and $s$ -regular by Proposition 2.8 .
Put $Δ = min {ε_{*}, Δ_{1}}$ . We claim that there exists $Δ_{ℛ} > 0$ such that any fine $ℛ$ -admissible sequence $(k_{n})_{n \geq 1}$ for $w \in [- 1, 1]^{2} \times Σ$ is also an $(ℛ, Δ)$ -admissible sequence for $I_{Δ_{ℛ}}^{s} (w)$ . In fact, if it does not hold, then for any $δ \in (0, Δ)$ there exist $δ^{'} \in (0, δ)$ and $w \in [- 1, 1]^{2} \times Σ$ and an $(ℛ, Δ)$ admissible sequence $(k (n))_{n = 1}^{n_{*}}$ such that $| r_{k (n_{*})} \circ \dots r_{k} (1) (I_{δ^{'}}^{s} (w)) | = Δ$ . Hence, we can take sequences $(w_{i})_{i = 1}^{\infty}$ in $[- 1, 1]^{2} \times Σ$ and $(δ_{i})_{i = 1}^{\infty}$ in $(0, Δ)$ so that $δ_{i}$ tends to zero as $i \to \infty$ , and ${I_{δ_{i}}^{s} (w_{i})}_{i = 1}^{\infty}$ is a $Δ_{1}$ -family. However, it contradicts the choice of $Δ_{1}$ .
It is easy to see that if the constant $δ_{ℛ}$ that is obtained in the above claim does not satisfies the second assertion of the lemma, then we can take a $Δ_{1}$ -family of intervals. However, it contradicts the choice of $Δ_{1}$ .

For a rectangle

R = I \times J \times σ

and a point

w = (x, y, σ)

R

, we define two intervals

I^{s} (R, w)

and

I^{u} (R, w)

I^{s} (R, w) = I \times y \times σ

and

I^{u} (R, w) = x \times J \times σ

We call a family

ℛ = {r_{k} : R_{k} \to R_{k}^{'}}_{k = 1}^{k_{*}}

of returns a Markov family if there exists a

{0, 1}

-valued

(k_{*} \times k_{*})

-matrix

A_{ℛ} = (a_{i j})

such that

1. $R_{i}^{'} \subset \cup_{a_{i j} = 1} R_{j}$ ,
2. $I^{s} (R_{i}^{'}, w) \subset I^{s} (R_{j}, w)$ and $I^{u} (R_{j}, w) \subset I^{u} (R_{i}^{'}, w)$ for $w \in R_{i}^{'} \cap R_{j}$ with $a_{i j} = 1$ ,
3. $Int R_{j} \cap Int R_{j^{'}} = \emptyset$ if $a_{i j} = a_{i j^{'}} = 1$ and $j \neq j^{'}$ , and
4. there exists a sequence $(ε_{n})_{n = 1}^{\infty}$ such that $ε_{n}$ tends to zero as $n \to \infty$ and if a sequence $(k (m))_{m = 0}^{n}$ and $w \in R_{k (0)}$ satisfies $a_{k (n) k (n + 1)} = 1$ and $r_{k (n)} \circ \dots \circ r_{k (0)} (w) \in R_{k (n + 1)}$ for any $m = 0, \dots, n - 1$ , then
$| I^{u} (R_{0}, w) \cap (^{n - 1} ⋂_{m = 0} (r_{k (m)} \circ \dots \circ r_{k (1)})^{- 1} (R_{k (m + 1)})) | \leq ε_{n} .$

We call

A_{ℛ}

the transition matrix of

P

Figure 4 : Markov family

Lemma 3.2. Any canonical cross-section admits a Markov family of returns.

Proof. Fix a full family P = { p l : P l → P l ′ } l = 1 l * of returns associated to ( Φ , ψ ) . Let Δ ℛ be the constant obtained in Lemma 3.1 for some ε * > 0 . Since M = Per ( Φ ) ¯ , we can take a family ℛ = { r k : R k → R k ′ } k = 1 k * of returns such that ∪ k = 1 k * R k ⊃ [ − 1 , 1 ] 2 × Σ and
- ∙ $R_{k}^{'} \subset [- 1, 1]^{2} \times Σ$ , $R_{k} \subset P_{l_{k}}$ , and $r_{k} = p_{l_{k}} |_{R_{k}}$ for some $l_{k}$ ,
- ∙ the diameters of $R_{k}$ and $R_{k}^{'}$ is less than $Δ_{ℛ}$ , and
- ∙ every boundary segment of $\partial R_{k}$ intersects with $ψ^{- 1} (Per (Φ))$ in its interior
for any k . Lemma 3.1 allows we can apply the proof of Theorem 2 in [20,Appendix2] . In fact, we obtain a Markov family as a subdivision of ℛ .

3.2 The reduced one-dimensional map

Fix a canonical cross-section

ψ

and a Markov family

ℛ = {r_{k} : R_{k} \to R_{k}^{'}}_{k = 1}^{k_{*}}

of returns with the transition matrix

A_{ℛ} = (a_{i j})

. Let

π_{y}

be the projection defined by

π_{y} (x, y, σ) = y

. Put

I_{k}^{'} = π_{y} (R_{k}^{'}) \times k

for

k \in {1, \dots, k_{*}}

and

I_{*} = \cup_{k = 1}^{k_{*}}

. Take a partition

{I_{i j} | a_{i j} = 1}

I_{*}

so that

\bar{I_{i j}} = π_{y} (R_{j}) \times i

for any

i

and

j

with

a_{i j} = 1

We define a map

f : I_{*} \to I_{*}

so that

f (y, i) = ((π_{y} \circ r_{j}) (x, y, σ)) \times j

for

(x, y, σ) \in R_{j} \cap R_{i}^{'}

with

a_{i j} = 1

and

y \in I_{i j}

. For

y \in I_{*}

and

n \geq 0

, let

I (y, n)

be the set of

y^{'} \in I_{*}

satisfying

y^{'} \in I_{i_{m} j_{m}}

for any

0 \leq m \leq n

y \in I_{i_{m} j_{m}}

. It is easy to see that

f^{n} (I (y, m + n)) \subset I (y, m)

Lemma 3.3. If

I (y, m) \subset f^{n} (I (y, m))

for

y \in I_{*}

m \geq 0

, and

n \geq 1

, then

I (y, m)

contains a fixed point of

f^{n}

Proof. It is easy to see that the restriction of $f^{n}$ on $I (y, m)$ extend to a continus map ${\bar{f}}_{y, m}^{n}$ on $\bar{I (y, m)}$ uniquely for any $y$ , $m$ and $n$ . In particular, if $I (y, m) \subset f^{n} (I (y, m))$ then the map ${\bar{f}}_{y, m}^{n}$ has a fixed point.
It is easy to see that ${\bar{f}}_{y, m}^{m + 1} (\partial (I (y, m))) \subset \partial I_{*}$ for any $y \in I_{*}$ and $m \geq 0$ .
Hence, we have $\partial I (y, m) \cap Per ({\bar{f}}_{y, m}^{n}) \subset \partial I_{*}$ . On the other hand, the construction of the partition ${I_{i j}}$ implies $\partial I_{*} \cap \bar{I_{i k}} \subset I_{i k}$ .

Put

δ_{n} = sup {| I (y, n) | | y \in I_{*}}

. The condition 4 in the definition of a Markov family implies that

δ_{n}

tends to zero as

n \to \infty

. In particular, we have

\cup_{n \geq 0} I (y, n) = {y}

for any

y \in I_{*}

In the rest of the subsection, we estimate

∥ D Φ^{t} |_{E^{u} / T Φ (z)} ∥

by a variation of a theorem of Man͂é [12] to

f

. For a

C^{2}

map

g : I \to I^{'}

between intervals, we define the distortion

dist (g, I)

g

dist (g, I) = sup {log | D g (y) | - log | D g (y^{'}) | | y, y^{'} \in I} .

We define the intersection multiplicity of a family

{S_{i}}

of subsets of a set

S

{sup}_{x \in X} # {i | x \in X_{i}}

, where

# S

is the cardinality of a set

S

. Notice that

C_{1} = (\sum_{k = 1}^{k_{*}} | I_{k}^{'} |) sup {| D (log | D f |) |}

is finite. Hence, if the intersection multiplicity of a family

{Int f^{m} (I)}_{m = 0}^{n - 1}

of subintervals of

I_{*}

is at most

l

, then

dist (f^{n}, I) \leq C_{1} l

Proposition 3.4. There exists a sequence

(K_{n})_{n \geq 1}

such that

K_{n}

tends to infinity as

n \to \infty

and

| (f^{n})^{'} (y_{*}) | \geq K_{n}

for any periodic point

y_{*}

of period

n

Proof. We follow the proof of Theorem 1 of [11] . Since

M = \bar{Per (Φ)}

, there exists a periodic point

y_{i j} \in Int I_{i j}

of period

m_{i j} \geq 1

for any

i, j

with

a_{i j} = 1

. Put

J_{i j}^{n} = I (y_{i j}, n) \ I (y_{i j}, n + 1)

. Let

C_{2}

be the constant satisfying

| J | \geq C_{2}

for any connected component

J

J_{i j}^{2}

with

a_{i j} = 1

Fix a periodic point

y_{*} \in I_{*}

of period

n_{0}

. Suppose

y_{*} \in J_{i_{0} j_{0}}^{m_{0}}

. Without loss of generality, we can assume that

y_{*}

is the point closest to

y_{i_{0} j_{0}}

{f^{n} (y_{*}) | 0 \leq n \leq n_{0} - 1}

. Let

J_{*}

be the minimal compact interval that contains

y_{*}

and a connected component of

J_{i_{0} j_{0}}^{m_{0} + 1}

. For

0 \leq n \leq n_{0}

, let

J_{*}^{n}

be the connected component of

f^{- n} (J_{*})

that contains

f^{n_{0} - n} (y_{*})

. It is easy to see that

J_{*}^{n} \cap {f^{m} (y_{*}) | 0 \leq m \leq n_{0} - 1} = {f^{n_{0} - n} (y_{*})}

for any

0 \leq n \leq n_{0}

. It implies that the intersection multiplicity of

{J_{*}^{n}}_{n = 0}^{n_{0} - 1}

is at most two. Since

f^{n} (J_{*}) \subset J_{i_{0} j_{0}}^{m_{0} - n} \cup J_{i_{0} j_{0}}^{m_{0} - n + 1}

for any

1 \leq n \leq m_{0} - 1

, the intersection multiplicity of

{f^{n} (J_{*})}_{n = 0}^{m_{0} - 1}

is also at most two. Therefore, we have

\begin{matrix} dist (f^{n_{0}}, J_{*}^{n_{0} - m_{0}}) & \leq & dist (f^{n_{0} - m_{0}}, J_{*}^{n_{0} - m_{0}}) + d i s t (f^{m_{0}}, J_{*}) \end{matrix}

\begin{matrix} \leq & 2 C_{1} + 2 C_{1} = 4 C_{1} . \end{matrix}

It implies that

\begin{matrix} | (f^{n_{0}})^{'} (y_{*}) | = | (f^{n_{0}})^{'} (f^{m_{0}} (y_{*})) | & \geq & exp (- 4 C_{1}) \frac{| f^{m_{0}} (J_{*}) |}{| J_{*}^{n_{0} - m_{0}} |} \end{matrix}

\begin{matrix} \geq & exp (- 4 C_{1}) C_{2} \cdot δ_{n_{0}}^{- 1}, \end{matrix}

and the last term tends to infinity as

n_{0} \to \infty

Corollary 3.5.

{Per}_{*}^{u} (Φ)

contains only finitely many orbits.

Proof. Since any periodic point of $Φ$ is $s$ -regular, the set $ψ (I^{s} (w, R_{k}^{'}))$ contains at most one periodic point for any $w \in R_{k}^{'}$ . It implies that there exists a one-to-one correspondence $H$ between $Per (f)$ and $ψ (\cup_{k = 1}^{k_{*}} R_{k}^{'}) \cap Per (Φ)$ . It is easy to see that $| (f^{n})^{'} (y) | = 1$ if and only if $∥ D {\hat{Φ}}^{t} |_{(E^{u} / T Φ) (H (y))} ∥ = 1$ for any $y \in Per (f)$ , where $n$ is the period of $y$ and $t$ is that of $H (y)$ . Hence, the corollary follows from Proposition 3.4 .

The next is the main results of this subsection.

Proposition 3.6. For any given constant

α > 0

and any neighborhood

U_{*}

{Per}_{*}^{u} (Φ)

, there exists

T \geq 0

such that any

z \in M \ \cap_{t \geq T} (U_{*})

satisfies

sup {∥ D Φ^{t} |_{(E^{u} / T Φ) (z)} ∥ | t \geq 0} \geq α .

Corollary 3.7. If a

C^{2}

-regular

P A

flow

Φ

on a 3-dimensional manifold satisfies

\bar{Per (Φ)} = M

and

{Per}_{*}^{u} (Φ) = {Per}_{*}^{s} (Φ) = \emptyset

, the it is an Anosov flow.

Proof. Proposition 3.6 implies that any $z \in M$ satisfies $∥ D {\hat{Φ}}^{t_{z}} |_{(E^{u} / T Φ) (z)} ∥ > 2$ for some $t_{z} > 0$ . By the compactness of $M$ , there exists $T > 0$ such that $∥ D {\hat{Φ}}^{T} |_{(E^{u} / T Φ) (z)} ∥ > 2$ for any $z \in M$ . Similarly, we can also take $T^{'} > 0$ such that $∥ D {\hat{Φ}}^{- T^{'}} |_{(E^{s} / T Φ) (z)} ∥ > 2$ for any $z \in M$ .

In the proof of the proposition, we follow the argument in Theorem 5.1 of [15,ChapterIII] . Let

P_{*} = \cup_{k = 1}^{k_{*}} (π_{y} (ψ^{- 1} ({Per}_{*}^{u} (Φ)) \cap R_{k}^{'}) \times k)

. Remark that

P_{*}

coincides the set of periodic points

p_{*}

f

with

| (f^{n}) (p_{*}) | = 1

, where

n

is the period of

p_{*}

Lemma 3.8. There exists

N_{1} \geq 1

such that any

y \in I_{*} \ P_{*}

satisfies

I (f^{n_{1}} (y), N_{1}) \cap P_{*} = \emptyset

for some

n_{1} \geq 0

Proof. Since $P_{*}$ is finite, there exists $N_{1} \geq 1$ such that $I (y_{*}, N_{1} - 1)$ contains at most one point of $P_{*}$ for any $y_{*} \in I_{*}$ .
Fix $y \in I_{*}$ such that $I (f^{n} (y), N_{1}) \cap P_{*} \neq \emptyset$ for any $n \geq 0$ . Let $y_{n}$ be the unique point of $P_{*}$ in $I (f^{n} (y), N_{1})$ for $n \geq 0$ . Since both $f (y_{n})$ and $y_{n + 1}$ is contained in $I (f^{n + 1} (y), N_{1} - 1)$ , we have $y_{n + 1} = f (y_{n})$ . In particular, $f^{n} (y_{0}) \in I (f^{n} (y), N_{1})$ for any $n \geq 1$ . Since $\cap_{n \geq 0} f^{- n} (I (y, N_{1})) = {y}$ , we have $y = y_{0} \in P_{*}$ .

Lemma 3.9. There exist

N_{2} \geq 1

and

C_{2} >

such that

| (f^{n})^{'} (y) | \geq exp (- C_{2}) \frac{| I (f^{n} (y), N_{2}) |}{| I (y, N_{2} + n) |}

for any

y \in I_{*}

and

n \geq 1

with

I (f^{n} (y), N_{2}) \cap P_{*} = \emptyset

Proof. By Proposition 3.4 , there exists a sequence ( K n ) n ≥ 1 such that K n tends to infinity as n → ∞ , and | ( f n ) ′ ( y ) | ≥ K n for any periodic point y of period n
.
Take n 1 ≥ 1 so that K n ≥ 2 exp ( C 1 ) for any n ≥ n 1 . We also take λ 0 < 1 and N 2 ≥ 1 so that | ( f n ) ′ ( y ) | ≥ λ 0 for any n ≤ n 1 , y 0 ∈ Fix ( f n ) \ P * , and y ∈ I ( y , N 2 )
.
We claim that there exists $λ_{0} > 1$ such that if $y_{0} \in I_{*}$ satisfies $f^{n} (y_{0}) \in I (y_{0}, N_{2})$ and $I (y_{0}, N_{2}) \cap P_{*} = \emptyset$ , then $| (f^{n})^{'} (y) | \geq λ_{0}$ for any $y \in I (y_{0}, N_{2} + n)$ .
We can assume that $f^{m} (y_{0}) \notin I (y_{0}, N_{2})$ for $1 \leq m \leq n - 1$ without loss of generality. Then, the intersection multiplicity of ${f^{m} (I (y_{0}, N_{2} + n - m))}_{m = 0}^{n - 1}$ is one and there exists a periodic point $y_{*} \in I (y_{0}, N_{2} + n) \ P_{*}$ of period $n$ . If $n \geq n_{1}$ , then we have $| (f^{n})^{'} (y) | \geq exp (- C_{1}) | (f^{n})^{'} (y_{*}) | \geq 2$ . If $n \leq n_{1}$ , then it is clear that $| (f^{n})^{'} (y) | \geq λ_{0}$ .
We say an interval $I \subset I_{*}$ is $(λ, n)$ -compatible when

∙ $Int f^{i} (I) \cap Int f^{j} (I) = \emptyset$ or $f^{i} (I) \subset f^{j} (I)$ for any $0 \leq i < j \leq n$ , and

∙ if $f^{i} (I) \cup f^{j} (I) \subset f^{k} (I)$ for $0 \leq i < j \leq k \leq n$ , then $| f^{j} (I) | \geq λ | f^{i} (I) |$ .

By Lemma 5.7 of [15,ChapterIII] , we have $\sum_{m = 0}^{n} | f^{m} (I) | \leq λ (λ - 1)^{- 1} (\sum_{k = 1}^{k_{*}} | I_{k}^{'} |)$ for any $(λ, n)$ -compatible interval $I$ . In particular, there exists $C_{2} > 1$ such that $dist (f^{n}, I) \leq C_{2}$ for any $(λ_{0}, n)$ -compatible interval $I$ .
We show that $I (y, N_{2} + n)$ is $(λ_{0}, n)$ -compatible if $y \in I_{*}$ and $n \geq 1$ satisfy $I (f^{n} (y), N_{2}) \cap P_{*} = \emptyset$ . Once it is done, the proof is completed. First, it is clear that the first condition holds. Suppose integers $i < j \leq k \leq n$ satisfy $f^{i} (I (y, N_{2} + n)) \cup f^{j} (I (y, N_{2} + n)) \subset f^{k} (I (y, N_{2} + n))$ . Since $f^{j} (y) \in I (f^{k} (y), N_{2}) = I (f^{i} (y), N_{2})$ , we can apply the claim above to $y_{0} = f^{i} (y)$ , $n = j - i$ and $I (y_{0}, N_{2}) = I (f^{i} (y), N_{2})$ . It implies that $| f^{j} (I (y, N_{2} + n)) | \geq λ_{0} | f^{i} (I (y, N_{2} + n)) |$ . Therefore, $I (y, N_{2} + n)$ is $(λ_{0}, n)$ compatible.

Proof of Proposition 3.6 . It is enough to show that for any given $α > 0$ there exists $N \geq 1$ such that $sup {| (f^{n})^{'} (y) | | n \geq 0} \geq α$ for any $y \in I_{*}^{\infty} \ f^{- N} (P_{*})$ .
Let $N_{1}$ , $N_{2}$ and $C_{2}$ be the numbers obtained in Lemmas 3.8 and 3.9 . Fix $N \geq N_{2}$ so that $| I (y, N_{1}) | \geq α exp (C_{2}) | I (y^{'}, N_{1} + N) |$ for any $y, y^{'} \in I_{*}$ .
Take $y \in I_{*} \ f^{N} (P_{*})$ . Then, Lemma 3.8 implies $I (f^{n + N} (y), N_{1}) \cap P_{*} = \emptyset$ for some $n \geq 1$ . By Lemma 3.9 , we obtain
$| (f^{n + N})^{'} (y) | \geq exp (- C_{2}) \frac{| I (f^{n + N} (y), N_{1}) |}{| I (y, N_{1} + n + N) |} \geq exp (- C_{2}) \frac{| I (f^{n + N} (y), N_{1}) |}{| I (y, N_{1} + N) |} \geq α .$

4 Regular $P A$ flows without invariant tori

In this section, we fix a

C^{2}

-regular

P A

flow and assume that

M = \bar{Per (Φ)}

and any periodic point of

Φ

s

and

u

-regular. The goal is to show that

{Per}_{*} (Φ)

is empty. The author recommend that the readers should refer to [3] , which provides a sketch of the proof for the case that

Φ

admits a global cross-section.

First of all, we fix a good parameter change of

Φ

and a family of coordinates associated to the flow. Remark that

ℱ^{u}

ℱ^{s}

and

{Per}_{*} (Φ)

does not depend on the parameter change of

Φ

Fix a neighborhood

U_{*}

{Per}_{*} (Φ)

and a

C^{2}

-foliation

G

U_{*}

so that

T_{z} M = T_{z} G \oplus T Φ (z)

for any

z \in U_{*}

. Recall that

{Per}_{*} (Φ)

contains only finitely many periodic orbits by Corollary 3.5 . We replace

Φ

by its parameter change if it is necessary, and assume that

1. all $z_{*} \in {Per}_{*} (Φ)$ have the same period $T_{*}$ , and
2. $D Φ^{t} (T_{z} G) = T_{Φ^{t} (z)} G$ for any $t \geq 0$ and $z \in \cap_{t^{'} = 0}^{t} Φ^{- t} (U_{*})$ .

Let

X

be the vector field that generates

Φ

. For each

ρ \in {u, s}

, we fix a

C^{2}

unit vector field

Y^{ρ}

so that

{X, Y^{ρ}}

is a framing of

E^{ρ}

and

Y^{ρ} (z) \in T_{z} G

z \in U_{*}

. We replace the norm

∥ \cdot ∥

T M

so that

{X, Y^{s}, Y^{u}}

forms an orthonormal framing of

T M

. Remark that

D Φ^{t} (X (z)) = X (Φ^{t} (z))

, and hence,

∥ D Φ^{t} (X (z)) ∥ = 1

for any

z \in M

and

t \in R

Let

{e_{x} (w), e_{y} (w), e_{s} (w)}

be the natural basis of

T_{w} R^{3}

w = (x, y, s) \in R^{3}

For

w \in [- 2, 2]^{3}

and

δ > 0

, we define cones

C_{x} (w, δ)

and

C_{y} (w, δ)

T_{w} R^{3}

\begin{matrix} C_{x} (w, δ) & = & {a e_{s} (w) + b e_{x} (w) | | a | \leq δ | b |}, \end{matrix}

\begin{matrix} C_{y} (w, δ) & = & {a e_{s} (w) + b e_{y} (w) | | a | \leq δ | b |} . \end{matrix}

We call an embedding

φ_{σ} : [- 2, 2]^{3} \to M

a canonical coordinate if

\begin{matrix} D φ_{σ}^{- 1} (X (φ (w))) & \in & {a e_{s} (w) | a > 0}, \end{matrix}

\begin{matrix} D φ_{σ}^{- 1} (Y^{s} (φ (w))) & \in & C_{x} (w, 1 / 4), \end{matrix}

\begin{matrix} D φ_{σ}^{- 1} (Y^{u} (φ (w))) & \in & C_{y} (w . 1 / 4) \end{matrix}

for any

w \in [- 2, 2]^{2}

, We can take a finite family

{φ_{σ}}_{σ \in Σ}

of canonical coordinates so that

1. $\cup_{σ \in Σ} φ_{σ} ((- 1, 1)^{3}) = M$ ,
2. the map $ψ (x, y, σ) = φ_{σ} (x, y, 0)$ is a canonical cross-section associated to $Φ$ , and
3. if $Im φ_{σ} \cap {Per}_{*} (Φ) \neq \emptyset$ , then $Im φ_{σ} \cap {Per}_{*} (Φ) = φ_{σ} (0 \times 0 \times [- 2, 2])$ , $Im φ_{σ} \subset \cap_{t = - T_{*}}^{T_{*}} Φ^{- t} (U_{*})$ , and $φ_{σ} ([- 2, 2]^{2} \times s) \subset G (φ_{σ} (0, 0, s))$ for any $s \in [- 2, 2]$ .

We put

Σ_{*} = {σ \in Σ | Im φ_{σ} \cap {Per}_{*} (Φ) \neq \emptyset}

. Remark that

D φ_{σ}^{- 1} (Y^{s})

is parallel to

e_{x}

and

D φ_{σ}^{- 1} (Y^{u})

is parallel to

e_{y}

Im φ_{σ}

for

σ \in Σ_{*}

For

ρ \in {u, s}

and

t \in R

, we define a vector field

Y_{t}^{ρ}

Y_{t}^{ρ} (Φ^{t} (z)) = D Φ^{t} (Y^{ρ} (z))

. For

σ \in Σ

, we define functions

π_{σ, x}

π_{σ, y}

, and

π_{σ, s}

Im φ_{σ}

(π_{σ, x} (z), π_{σ, y} (z), π_{σ, s} (z)) = (x, y, s)

for

z = φ_{σ} (x, y, s)

In Subsection 4.1 , we show that the curves tangent to

Y_{t}^{u}

satisfy a kind of uniform continuity as graphs of functions in canonical coordinates. It allows the argument in [3] to work well. In fact, in Subsection 4.2 , we estimate the distortion of a holonomy map of

ℱ^{u}

in two ways and the comparition of them implies

{Per}_{*} (Φ) = \emptyset

. Combined with Propositions 2.1 , 2.2 , and Corollary 3.7 , it completes the proof of the main theorem.

4.1 Quasi-invariant vector fields

For

ρ \in {u, s}

z \in M

, and

δ > 0

, we define a cone

C^{ρ} (z, δ)

T_{z} M

C^{ρ} (z, δ) = {a X (z) + b Y^{ρ} (z) | | a | \leq δ | b |} .

We also define functions

\hat{a}

and

\hat{b}

M \times R

Y_{t}^{u} (z) = \hat{a} (z, t) (Φ^{t} (z)) + \hat{b} (z, t) Y^{u} (Φ^{t} (z)) .

Lemma 4.1.

D Φ^{t} (C^{u} (z, δ)) \subset C^{u} (Φ^{t} (z), α δ)

for any

z \in M

t \geq 0

α > 2 | \hat{b} (z, t) |^{- 1}

, and

δ > | \hat{a} (z, t) |

Proof. Proof is by elementary calculation. Since

D Φ^{t} (X (z))) = X (Φ^{t} (z))

, we have

D Φ^{t} (a X (z) + b Y^{u} (z)) = (a + b \cdot \hat{a} (z, t)) X (z) + b \cdot \hat{b} (z, t) Y^{u} (z) .

for

a, b \in R

. If

α > 2 | \hat{b} (z, t) |^{- 1}

δ > | \hat{a} (z, t) |

, and

| a | \leq δ | b |

, then

\begin{matrix} | a + b \cdot \hat{a} (z, t) | & \leq & | b | (δ + | \hat{a} (z, t) |) \end{matrix}

\begin{matrix} \leq & (2^{- 1} α \cdot | \hat{b} (z, t) |) \cdot | b | \cdot (2 δ) = (α δ) | b \cdot \hat{b} (z, t) | . \end{matrix}

It implies the required inclusion.

The aim of this subsection is to show the following.

Proposition 4.2. There exists

Δ_{1} \in (0, 1 / 4)

such that

1. If a curve $J \subset M$ is tangent to $Y_{t}^{u}$ for $t \geq 0$ , and satisfies $J \subset Im φ_{σ}$ and $| π_{σ, y} (J) | \leq Δ_{1}$ for $σ \in Σ$ , then $| π_{σ, s} (J) | \leq 1 / 4$ , and
2. if a curve $J \subset M$ is tangent to $Y_{- t}^{s}$ for $t \geq 0$ , and satisfies $J \subset Im φ_{σ}$ and $| π_{σ, x} (J) | \leq Δ_{1}$ for $σ \in Σ$ , then $| π_{σ, s} (J) | \leq 1 / 4$ .

We prepare two lemmas to prove the proposition. The first allows us to control the expansion of cones in a small neighborhood of

{Per}_{*} (Φ)

. The second asserts the existence of the uniform lower bound of the angle between

Y_{t}^{s}

and

X

outside any given neighborhood of

{Per}_{*} (Φ)

For any subset

V

M

, we define the escape-time function

τ_{V}^{E} : V \to {0 \leq t \leq \infty}

τ_{V}^{E} (z) = inf {t > 0 | Φ^{t} (z) \notin V} .

Lemma 4.3. Suppose that

z_{*} \in {Per}_{*} (Φ)

and a neighborhood

U

O (z_{*})

are given. There exist a neighborhood

V \subset U

O (z_{*})

and a function

T_{V}

{α > 0}

such that if

z \in V

satisfies

τ_{V}^{E} (z) \geq T_{V} (α)

then

D Φ^{τ_{V}^{E} (z)} (C^{u} (z, δ)) \subset C^{u} (Φ^{τ_{V}^{E} (z)} (z), α δ)

for any

δ > 0

Proof. Without loss of generality, we can assume that $z_{*} = φ_{σ} (0, 0, 0)$ for some $σ \in Σ_{*}$ . Let $r : R \to R^{'}$ be the first return of $z_{*}$ and $(r_{x}, r_{y})$ the $x y$ -decomposition of $r$ . We remark that $Φ^{T_{*}} (ψ (x, y, σ)) = ψ \circ r (x, y, σ)$ for any $(x, y, σ) \in R$ , Since $z_{*}$ is $u$ and $s$ -regular, we have $| r_{x} (x) | < | x |$ and $| r_{y}^{- 1} (y) | < | y |$ if $x, y \neq 0$ .
Take a subrectangle $R_{0} = I_{x} \times I_{y} \times σ$ of $R$ so that $(0, 0, 0) \in Int R_{0}$ and $V = \cup_{0 \leq t < T_{*}} Φ^{- t} (ψ (R_{0}))$ satisfies $V \cap ψ ([- 2, 2]^{2} \times σ) = ψ (R_{0})$ . Put $I_{n} = r_{y}^{- n} (I_{y})$ for $n \geq 0$ . We claim that there exists a sequence $(K_{n})_{n \geq 1}$ such that $| (r_{y}^{n})^{'} (y) | \geq K_{n}$ for any $n \geq 1$ and $y \in I_{n} \ I_{n + 1}$ . Put $I_{n}^{+} = I_{n} \cap [0, 2]$ and $I_{n}^{-} = I_{n} \cap [- 2, 0]$ . Since the intersection multiplicity of ${I_{n}^{ρ} \ I_{n + 1}^{ρ}}_{n \geq 1}$ is one for $ρ \in {+, -}$ , there exists $C_{1} > 0$ such that $dist (r_{y}^{n}, I_{n}^{ρ} \ I_{n + 1}^{ρ}) \geq C_{1}$ for any $n \geq 1$ and $ρ \in {+, -}$ . Hence, we have
$| (r_{n}^{y})^{'} (y) | \geq exp (- C_{1}) \frac{| I_{0}^{ρ} \ I_{1}^{ρ} |}{| I_{n}^{ρ} \ I_{n + 1}^{ρ} |}$
for any $y \in I_{n}^{ρ} \ I_{n + 1}^{ρ}$ . Since $I_{n}$ converges to ${0}$ , the right term tends to infinity as $n \to \infty$ .
Put $B_{n} = ψ (I_{x} \times (I_{n} \ I_{n + 1}) \times σ)$ . If $z \in V$ satisfies $τ_{V}^{E} (z) < \infty$ , then there exist $t_{*} (z) \in [0, T_{*}]$ and $n_{*} (z) \geq 0$ such that $z \in Φ^{- t_{*} (z)} (B_{n_{*} (z)})$ . Notice that $Φ^{t_{*} (z) + n T_{*} - t} (z)$ is contained in a subset $Φ^{- t} (B_{n_{*} (z) - n})$ of $V$ for any $z \in V$ , $n \leq n_{*} (z) + 1$ , and $t \in [0, T_{*})$ . Since $τ_{V}^{E} (z) = 0$ for any $z \in B_{0}$ , it implies $n_{*} (z) T_{*} + t_{*} (z) = τ_{V}^{E} (z)$ .
Since both $Y^{u}$ and $Y_{n T_{*}}^{u}$ are parallel to $D φ_{σ} \circ e_{y}$ on $ψ ([- 2, 2]^{2} \times σ)$ , there exists $C_{2} > 1$ such that
$∥ D Φ^{n T_{*}} (Y^{u} (z)) ∥ \geq C_{2}^{- 1} | (r_{y}^{n})^{'} (y) | \geq K_{n}$
for any $z = ψ (x, y, σ) \in B_{n}$ . Since $D Φ_{n T_{*}} X (z) = X (Φ^{n T_{*}} (z))$ , it implies
$D Φ^{n T_{*}} (C^{u} (z, δ)) \subset C^{u} (Φ^{n T_{*}} (z), C_{2} K_{n}^{- 1} δ)$
for any $z \in B_{n}$ and $δ > 0$ . By Lemma 4.1 , we can take $C_{3} > 0$ so that $D Φ^{t} (C^{u} (z, δ)) \subset C^{u} (Φ^{t} (z), C_{3} δ)$ for any $z \in M$ , $δ > 0$ , and $t \in [0, T_{*}]$ . Hence, if $z \in V$ satisfies $τ_{V}^{E} (z) < \infty$ , then
$D Φ^{τ_{V}^{E} (z)} (C^{u} (z, δ)) \subset C^{u} (Φ^{τ_{V}^{E} (z)} (z), C_{2} C_{3} K_{n_{*} (z)}^{- 1} δ)$
for any $δ > 0$ . Since $K_{n}$ tends to infinity as $n \to \infty$ and $n_{*} (z) = T_{*}^{- 1} (τ_{V}^{E} (z) - t_{*} (z)) \geq T_{*}^{- 1} \cdot τ_{V}^{E} (z) - 1$ , this completes the proof.

Lemma 4.4. For any given neighborhood

U

{Per}_{*} (Φ)

, there exist

δ_{*} > 0

such that if

z \in M

satisfies

Φ^{t} (z) \notin U

for

t \geq 0

, then

Y_{t}^{u} (z) \subset C^{u} (z, δ_{*})

Proof. To prove the lemma, it is sufficient to show that there exist

δ_{*}, δ_{*}^{'} > 0

such that if

z \in M

satisfies

Φ^{t} (z) \notin U

for

t \geq 0

, then

D Φ^{t} (C^{u} (z, δ_{*}^{'})) \subset C^{u} (Φ^{t} (z, δ_{*}))

By Lemma 4.3 , there exists an open neighborhood

V_{0}

{Per}_{*} (Φ)

and a function

T_{V_{0}}

such that

\bar{V_{0}} \subset U

and

\begin{matrix} D Φ^{τ_{V_{0}}^{E}} (z) (C^{u} (z, δ)) \subset C^{u} (Φ^{τ_{V_{0}}^{E} (z)} (z), α δ) \end{matrix}

(1)

for any

α > 0

δ > 0

, and

z \in V_{0}

with

τ_{V_{0}}^{E} \geq T_{V_{0}} (α)

. Put

V_{*} = \cap_{t \geq 0} Φ^{- t} (V_{0})

By Proposition 3.6 , there exists

T_{1} > 0

such that

sup {\hat{b} (z, t) | t \geq 0} > 2

for any

z \in M \ Φ^{- T_{1}} (V_{*})

. By Lemma 4.1 , we can take

δ_{1} > 0

and

α_{1} > 0

so that

D Φ^{T_{1}} (C^{u} (z, δ)) \subset C^{u} (Φ^{T_{1}} (z), α_{1} δ_{1})

for any

z \in M

and

δ \geq δ_{1}

We define a function

τ_{1}^{*}

Φ^{- T_{1}} (V_{0})

τ_{1}^{*} (z) = τ_{V_{0}}^{E} (Φ_{1}^{T} (z)) + T_{1}

. Put

T_{2} = T_{V_{0}} (α_{1}^{- 1}) + T_{1}

and take an open set

V_{1} = {z \in Φ^{- T_{1}} (V_{0}) | τ_{1}^{*} (z) > T_{2}}

By the inclusion ( 1 ), we have

\begin{matrix} D Φ^{τ_{1}^{*} (z)} (C^{u} (z, δ)) \subset C^{u} (Φ^{τ_{1}^{*} (z)} (z), δ) \end{matrix}

(2)

for any

δ \geq δ_{1}

and

z \in V_{1}

Notice that

sup {\hat{b} (z, t) | t \geq 0} > 2

for any

z \in M \ V_{1}

since

Φ^{- T} (V_{*}) \subset V_{1}

By Lemma 4.4 and the compactness of

M \ V_{1}

, there exist

0 < τ_{1} < τ_{2}

δ_{*}^{'} > δ_{1}

, and a function

τ_{2}^{*} : M \ V_{1} \to [τ_{1}, τ_{2}]

such that

\begin{matrix} D Φ^{τ_{2}^{*} (z)} (C^{u} (z, δ)) \subset C^{u} (Φ^{τ_{2}^{*} (z)} (z), δ) \end{matrix}

(3)

for any

δ \geq δ_{*}^{'}

and

z \in M \ V_{1}

. By Lemma 4.1 , we can take

δ_{*}

so that

D Φ^{t} (C^{u} (z, δ_{*}^{'})) \subset C^{u} (Φ^{t} (z), δ_{*})

for any

z \in M

and

t \in [0, max {T_{1}, τ_{2}}]

Fix

t > 0

and

z_{0} \in M \ Φ^{- t} (U)

. By the inclusions ( 2 ) and ( 3 ), there exists a sequence

(t_{i})_{i = 0}^{i_{*} + 1}

such that

t_{0} = 0

t_{i_{*}} < t \leq t_{i_{*} + 1}

and each

t_{i}

satisfies

1. $Φ^{t_{i}} (z_{0}) \in V_{1}$ and $t_{i + 1} = t_{i} + τ_{1}^{*} (Φ^{t_{i}} (z_{0})) \geq t_{i} + T_{2}$ , or
2. $Φ^{t_{i}} (z_{0}) \in M \ V_{1}$ and $t_{i + 1} = t_{i} + τ_{2}^{*} (Φ^{t_{i}} (z_{0})) \geq t_{i} + τ_{1}$ .

Remark that

D Φ^{t_{i_{*}}} (C^{u} (z, δ_{*}^{'})) \subset C^{u} (Φ^{t_{i_{*}}} (z_{0}), δ_{*}^{'})

by the inclusions ( 2 ) and ( 3 ).

Figure 5 : Proof of Lemma 4.4

We see

t - t_{i_{*}} \leq max {T_{1}, τ_{2}}

. In fact, if

Φ^{t_{i_{*}}} (z_{0}) \in V_{1}

, then

Φ^{t^{'}} (z_{0}) \in V_{0}

for

t^{'} \in [t_{i_{*}} + T_{1}, t_{i_{*} + 1}]

. Since

Φ^{t} (z_{0}) \notin U

, it implies

t - t_{i_{*}} \leq T_{1}

. If

Φ^{t_{i_{*} - 1}} (z_{0}) \in V_{1}

, then

t - τ_{i_{*}} \leq τ_{2}^{*} (Φ^{t_{i_{*}}} (z_{0})) \leq τ_{2}

. Therefore, we obtain

D Φ^{t} (C^{u} (z, δ_{*}^{'})) \subset C^{u} (Φ^{t} (z_{0}), δ_{*})

by the choice of

δ_{*}

Proof of Proposition 4.2 . We only prove the former assertion. The latter is given by applying the former to the flow

Φ^{- 1}

Take

δ_{0} > 0

so that

\begin{matrix} D φ_{σ}^{- 1} (Y_{t}^{u} (φ_{σ} (w))) \subset C_{y} (w, δ_{0} / 4) \end{matrix}

(4)

for any

w \in [- 2, 2]^{3}

σ \in Σ

, and

t \in [0, T_{*}]

For

σ \in Σ_{*}

, let

r_{σ} : R_{σ} \to R_{σ}^{'}

be the first return of

(0, 0, σ)

and

(r_{σ, x}, r_{σ, y})

the

x y

-decomposition of

r_{σ}

. Remark that

Φ^{T_{*}} (φ_{σ} (x, y, s)) = φ_{σ} (r_{σ, x} (x), r_{σ, y} (y), s)

for any

(x, y, σ) \in R

and

s \in [- 2, 2]

, and

| r_{σ, x} (x) | < | x |

and

| r_{σ, y}^{- 1} (y) | < | y |

x, y \neq 0

. Fix intervals

I_{x}

and

I_{y}

so that

(0, 0) \in Int I_{x} \times I_{y}

and

I_{x} \times I_{y} \times σ \subset R_{σ}

for any

σ \in Σ_{*}

Take a neighborhood

U_{1}

{Per}_{*} (Φ)

so that

U_{1} \cap Im φ_{σ} = \emptyset

for

σ \in Σ \ Σ_{*}

and

U_{1} \cap Im φ_{σ_{*}} \subset φ_{σ_{*}} (r_{σ, x} (I_{x}) \times I_{y} \times [- 2, 2])

for

σ_{*} \in Σ_{*}

. By Lemma 4.4 , there exists

δ_{1} > 0

such that if

w \in [- 2, 2]^{3}

and

σ \in Σ

satisfy

φ_{σ} (w) \notin U_{1}

then

\begin{matrix} D φ_{σ}^{- 1} (Y^{u} (φ_{σ} (w))) \subset C_{y} (w, δ_{1} / 4) \end{matrix}

(5)

for any

t \geq 0

Fix

ε_{*} \in (0, min {δ_{0}^{- 1}, δ_{1}^{- 1}} / 4)

so that

[- 2 ε_{*}, 2 ε_{*}] \subset I_{y}

and put

B_{σ}^{n} = φ_{σ} (r_{σ, x}^{n} (I_{x}) \times [- 2 ε_{*}, 2 ε_{*}] \times [- 2, 2])

for

σ \in Σ_{*}

and

n \geq 0

. Take a neighborhood

U_{2} \subset U_{1}

{Per}_{*} (Φ)

so that

U_{2} \cap Im φ_{σ} \subset B_{σ}^{1}

for any

σ \in Σ_{*}

. By Lemma 4.4 again, there exists

δ_{2} > 0

such that if

w \in [- 2, 2]^{3}

and

σ \in Σ

satisfy

φ_{σ} (w) \notin U_{2}

, then

\begin{matrix} D φ_{σ} (Y_{t}^{u} (φ_{σ} (w))) \subset C_{y} (w, δ_{2} / 4) \end{matrix}

(6)

for any

t \geq 0

. Put

Δ_{1} = min {δ_{2}^{- 1}, ε_{*}}

Suppose a curve

J \subset M

is tangent to

Y_{t}^{u}

for

t \geq 0

, and satisfies

J \subset Im φ_{σ}

and

| π_{σ, y} (J) | \leq Δ_{1}

for

σ \in Σ

. Then, there exists a function

h_{J}

such that

φ_{σ}^{- 1} (J) = {(x_{0}, y, h_{J} (y)) | y \in π_{σ, y} (J)}

, where

π_{σ, x} (J) = {x_{0}}

J \cap U_{2} = \emptyset

, then the inclusion ( 6 ) implies that

| π_{σ, s} (J) | \leq (δ_{2} / 4) | π_{σ, y} (J) | \leq 1 / 4

Suppose

J \cap U_{2} \neq \emptyset

. Then,

σ

is an element of

Σ_{*}

and

J \subset B_{σ}^{n} \ B_{σ}^{n + 1}

for some

n \geq 1

. It implies

Φ^{- m T_{*}} (J) = φ_{σ} ({(r_{σ, x}^{- m} (x_{0}), r_{σ, y}^{- m} (y), h_{J} (y)) | y \in π_{σ, y} (J)})

for any

0 \leq m \leq n

. In particular, we have

J \subset B_{σ}^{n - m} \ B_{σ}^{n - m + 1}

, and

| π_{σ, s} (Φ^{- m T_{*}} (J)) | = | π_{σ, s} (J) |, | π_{σ, y} (Φ^{- m T_{*}} (J)) | \leq | π_{σ, s} (J) | \leq 4 ε_{*} .

Take

N \geq 1

so that

N T_{*} \leq t \leq (N + 1) T_{*}

. If

n \geq N

, then

Φ^{- N T_{*}} (J)

is tangent to

Y_{t - N T *}^{u}

and is contained in

B_{σ}^{0} \subset Im φ_{σ}

. Hence, we have

\begin{matrix} | π_{σ, s} (J) | & = & | π_{σ, s} (Φ^{- N T_{*}} (J)) | \end{matrix}

\begin{matrix} \leq & (δ_{0} / 4) | π_{σ, y} (Φ^{- N T_{*}} (J)) | \leq δ_{0} ε_{*} \leq 1 / 4 \end{matrix}

by the inclusion ( 4 ) and

ε_{*} \leq δ_{0}^{- 1} / 4

n < N

, then

Φ^{- n T_{*}} (J) \subset B_{σ}^{0} \ B_{σ}^{1}

and hence,

Φ^{- n T_{*}} (J) \cap U_{1} = \emptyset

. Since

Φ^{- n T_{*}} (J)

is tangent to

Y_{t - n T_{*}}^{u}

, we have

\begin{matrix} | π_{σ, s} (J) | & = & | π_{σ, s} (Φ^{- n T_{*}} (J)) | \end{matrix}

\begin{matrix} \leq & (δ_{1} / 4) | π_{σ, y} (Φ^{- n T_{*}} (J)) | \leq δ_{1} ε_{*} \leq 1 / 4 \end{matrix}

by the inclusion ( 5 ) and the inequality

ε_{*} \leq δ_{1}^{- 1} / 4

4.2 Hyperbolicity of periodic orbits

The goal is the following proposition, which completes the proof of the main theorem combining with Propositions 2.1 , 2.2 and Corollary 3.7 .

Proposition 4.5.

\bar{{Per}_{*} (Φ)} = \emptyset

We need some preparation to prove the proposition. Suppose

{Per}_{*}^{s} (Φ) \neq \emptyset

and fix

z_{*} = φ_{σ_{*}} (0, 0, 0) \in {Per}_{*}^{s} (Φ)

. Let

Δ_{1} \in (0, 1 / 4)

be the constant obtained in Proposition 4.2 . For

t \geq 0

, we define a map

H_{t} : [0, Δ_{1}]^{2} \to M

so that

H_{t} (0, y) = φ_{σ_{*}} (0, y, 0)

π_{σ_{*}, x} (H_{t} (x, y)) = x

, and

H_{t} ([0, Δ_{1}] \times y)

is a curve tangent to

Y_{- t}^{s}

for any

(x, y) \in [0, Δ_{1}]^{2}

Lemma 4.6.

H_{t}

is well-defined and satisfies the followings:

1. $H_{t} ([0, Δ_{1}] \times y) \subset φ_{σ_{*}} ([0, Δ_{1}] \times [- 1 / 4, 1 / 4] \times 0)$ for any $y$ .
2. $H_{t} (0 \times [0, Δ_{1}]) = φ_{σ_{*}} (0 \times [0, Δ_{1}] \times 0)$ is tangent to $Y_{0}^{u}$ .
3. $H_{n T_{*}} (x, 0) = φ_{σ_{*}} (x, 0, 0)$ for any $n \geq 0$ .

Proof. The first assertion is a consequence of the choice of $Δ_{1}$ The second follows from the fact that $Y_{0}^{u}$ is parallel to $D φ_{σ_{*}} \circ e_{s}$ on $Im φ_{σ_{*}}$ .
Put $J = φ_{σ_{*}} ([0, Δ_{1}] \times 0 \times 0)$ . Then, we have $Φ^{n T_{*}} (J) \subset J$ . Since $Y_{0}^{s}$ is parallel to $D φ_{σ_{*}} \circ e_{x}$ on $Im φ_{σ_{*}}$ , the interval $Φ^{n T_{*}} (J)$ is tangent to $Y_{0}^{s}$ . It implies the last assertion of the lemma.

Put

C_{Σ} = sup {∥ D φ_{σ} ∥, ∥ D φ_{σ}^{- 1} ∥ | σ \in Σ} .

It is easy to see that

∙ $C_{Σ}^{- 1} | J | \leq | φ_{σ} (J) | \leq C_{Σ} | J |$ for any interval $J \subset Im φ_{σ}$ , and
∙ if an interval $J \subset M$ satisfies $J \cap φ_{σ} ([- 3 / 2, 3 / 2]^{2}) \neq \emptyset$ and $| J | \leq (4 C_{Σ})^{- 1}$ , then $J \subset φ_{σ} ((- 2, 2)^{3})$ .

For

t \geq 0

y \in [0, Δ_{1}]

, and

Δ \in (0, Δ_{1})

, we define an interval

J_{t} (y; Δ)

tangent to

Y_{0}^{s}

J_{t} (y; Δ) = Φ^{t} \circ H_{t} ([0, Δ] \times y)

Lemma 4.7. There exists

Δ_{2} \in (0, Δ_{1})

such that

| J_{t} (y; Δ_{2}) | \leq (4 C_{Σ})^{- 1}

for any

t \geq 0

and

y \in [0, Δ_{1}]

Proof. Fix a full family ${r_{k} : R_{k} \to R_{k}^{'}}_{k = 1}^{k_{*}}$ of returns associated to $(Φ, ψ)$ . Let $τ_{k}$ be the return time of $r_{k}$ . Put $T_{+} = sup {τ_{k} (w) | k = 1, \dots, k_{*}, w \in R_{k}}$ and $C_{1} = sup {∥ D Φ^{t} ∥ | t \in [0, T_{+}]}$ . Since $D φ_{σ}^{- 1} (Y_{- t}^{s} (φ_{σ} (w)))$ is transverse to the $x s$ -plane, there exists $Δ > 0$ such that if an interval $J$ is tangent to $Y_{- t}^{s}$ for $t \in [0, T_{+}]$ and satisfies $J \cap φ_{σ} ([- 1, 1]^{3}) \neq \emptyset$ and $| π_{σ, x} (J) | \leq Δ$ for $σ \in Σ$ , then $| J | \leq (4 C_{1} C_{Σ})^{- 1}$ . Suppose $(0, 0, σ_{*}) \in Int R_{k_{0}}$ . By Lemma 3.1 , we can take $Δ_{2} > 0$ so that any fine $ℛ$ -admissible sequence for $(0, y, σ_{*})$ with $y \in [0, Δ_{1}]$ is also an $(ℛ, Δ)$ -admissible sequence for $[0, Δ_{2}] \times y \times k_{0}$ .
Fix $t \geq 0$ and $y \in [0, Δ_{1}]$ . Take a fine $ℛ$ -admissible sequence $(k (n))_{n \geq 1}$ for $(0, y, σ_{*})$ . Put $w_{0} = (0, y, σ_{*})$ , $t_{0} = 0$ $w_{n} = r_{k (n)} \circ \dots \circ r_{k (1)} (w_{0})$ and $t_{n} = \sum_{m = 1}^{n} τ_{k (m)} (w_{m - 1})$ for any $n \geq 1$ . Take $n_{*} \geq 0$ so that $t_{n_{*}} \leq t < t_{n_{*} + 1}$ .
It is easy to see $Φ^{t_{n_{*}}} (φ_{σ_{*}} (0, y, 0)) = ψ (w_{n_{*}})$ and $0 \leq t - t_{n_{*}} \leq T_{+}$ . It implies that
$| π_{σ^{'}, x} \circ Φ^{t_{n_{*}}} \circ H_{T} ([0, Δ_{2}] \times y) | \leq Δ,$
where $R_{k (n_{*})}^{'} \subset [- 2, 2]^{2} \times σ^{'}$ . Since $Φ^{t_{n_{*}}} \circ H_{T} ([0, Δ_{2}] \times y)$ is tangent to $Y_{- t + t_{n_{*}}}^{s}$ , we have
$| Φ^{t_{n_{*}}} \circ H_{t} ([0, Δ_{2}] \times y) | \leq (4 C_{1} C_{Σ})^{- 1} .$
It is easy to see that the lemma follows from the choice of $C_{1}$ .

Since

M = \bar{Per (Φ)}

and

{Per}_{*} (Φ)

contains only finitely many periodic orbits, there exists

(x_{h}, y_{h}) \in [0, Δ_{2}] \times [0, Δ_{1}]

such that

z_{h} = H_{t} (x_{h}, y_{h})

is a point of

Per (Φ) \ {Per}_{*} (Φ)

. Put

J_{0} = φ_{σ_{*}} ([0, x_{h}] \times 0 \times 0)

and

γ_{t} = Φ^{t} \circ φ_{σ_{*}} (0 \times [0, y_{h}] \times 0)

We define a map

h_{t} : J_{0} \to J_{t} (y_{h}; x_{h})

h_{t} (φ_{σ} (x, 0, 0)) = Φ^{t} (H_{t} (x, y_{h})) .

Remark that

Φ^{- t} \circ h_{t}

and

h_{t} \circ Φ^{- t}

are the holonomy maps of

ℱ^{u}

between

J_{0}

and

Φ^{- t} (J_{t} (y_{h}; x_{h}))

along

γ_{0}

, and between

Φ^{t} (J_{0})

and

J_{t} (y_{h}; x_{h})

along

γ_{t}

. See figure 6 .

Figure 6 : The holonomy maps

Φ^{- t} \circ h_{t}

and

h_{t} \circ Φ^{- t}

Lemma 4.8. There exists

T_{h} \geq 0

such that

(dist (h_{n T_{h}}, J_{0}))_{n \geq 0}

is a bounded sequence.

Proof. Suppose

H_{t} (x_{h}, y_{h}) = φ_{σ_{*} (x_{h}, y_{h}, s_{h})}

. Since

z_{h} = φ_{σ_{*}} (x_{h}, y_{h}, s_{h})

is a hyperbolic periodic point, there exist maps

g : [0, x_{h}] \to [0, x_{h}]

τ : [0, x_{h}] \to {t > 0}

, and a constant

λ \in (0, 1)

such that

g (x_{h}) = x_{h}

Φ^{τ (x)} (φ_{σ_{*}} (x, y_{h}, s_{h})) = φ_{σ_{*}} (g (x), y_{h}, s_{h})

, and

0 < g^{'} (x) < λ

for any

x \in [0, x_{h}]

. Put

T_{h} = τ (x_{h})

. Then, we see that

Φ^{n T_{h}} (z_{h}) = z_{h}

and

J_{n T_{h}} (y_{h}; x_{h}) = φ_{σ_{*}} (g^{n} ([0, x_{h}]) \times y_{h} \times s_{h})

. In fact, the former is clear, and the latter follows from the fact that

Y_{0}^{s}

is parallel to

D φ_{σ_{*}} (e_{x})

Im φ_{σ_{*}}

, and hence,

J_{n T_{h}} (y_{h}; x_{h}) \subset φ_{σ_{*}} ([- 2, 2] \times y_{h} \times s_{h})

Put

C_{h} = sup {| D (log | D g (x) |) | | x \in [0, x_{h}]}

. Then, we have

\begin{matrix} dist (g^{n}, [0, x_{h}]) & \leq & \sum_{m = 0}^{n - 1} dist (g, g^{m} ([0, x_{h}])) \end{matrix}

\begin{matrix} \leq & \sum_{m = 0}^{n - 1} C_{h} | g^{m} ([0, x_{h}]) | \leq \sum_{m = 0}^{n - 1} C_{h} λ^{m} x_{h} < C_{h} (1 - λ)^{- 1} x_{h} . \end{matrix}

We define maps

l_{0}

and

l_{1}

from

[0, x_{h}]

M

l_{0} (x) = φ_{σ_{*}} (x, 0, 0)

and

l_{1} (x) = φ_{σ_{*}} (x, y_{h}, s_{h})

. It is easy to see

h_{n T_{h}} (z) = Φ^{n T_{h}} \circ H_{n T_{1}} (l_{0}^{- 1} (x), y_{h}) = l_{1} \circ g^{n} \circ l_{0}^{- 1} (z)

for any

z \in J_{0} = φ_{σ_{*}} ([0, x_{h}] \times 0 \times 0)

. Since

(dist (g^{n}, [0, x_{h}]))_{n \geq 1}

is a bounded sequence, the sequence

(dist (h_{n T_{h}}, J_{0}))_{n \geq 0}

also is.

Lemma 4.9. The family

{dist (h_{t} \circ Φ^{- t}, Φ^{t} (J_{0}))}_{t \geq 0}

is bounded.

Proof. For

t > 0

, there exist sequences

(y_{n} (t))_{n = 0}^{n (t)}

[0, Δ_{1}]

and

(σ_{n} (t))_{n = 0}^{n (t)}

Σ

such that

∙ $y_{0} (t) = 0$ , $σ_{0} (t) = σ_{*}$ , $y_{n (t)} (t) = Δ_{1}$ ,
∙ $y_{n} (t) < y_{n + 1} (t)$ and $σ_{n} (t) \neq σ_{n + 1} (t)$ for any $n$ , and
∙ $Φ^{t} \circ H_{t} (0 \times [y_{n} (t), y_{n + 1} (t)] \times 0)$ is contained in $φ_{σ_{n} (t)} ([- 1, 1]^{3})$ for any $n$ .

Let

L_{n} (t)

be the connected component of

{y \in [0, Δ_{1}] | Φ^{t} \circ H_{t} (0, y) \in φ_{σ_{n} (t)} ([- 3 / 2, 3 / 2]^{3})}

that contains

y_{n} (t)

. Put

I_{n} (t) = Φ^{t} \circ H_{t} (0 \times L_{n} (t))

and

B_{n} (t) = \cup_{y \in L_{n} (t)} J_{t} (y; x_{h})

. Note that

L_{n} (t)

contains

[y_{n} (t), y_{n + 1} (t)]

and

I_{n} (t)

is the connected component of

γ_{t} \cap φ_{σ_{n} (t)} ([- 3 / 2, 3 / 2]^{3})

that contains

Φ^{t} \circ H_{t} (0, y_{n} (t))

. Since

| J_{t} (y; x_{h}) | \leq (4 C_{Σ})^{- 1}

and

J_{t} (y; x_{h}) \cap φ_{σ_{n} (t)} ([- 3 / 2, 3 / 2]^{3}) \neq \emptyset

for any

y \in L_{n} (t)

, we have

B_{n} (t) \subset Im φ_{σ_{n} (t)}

. It implies

π_{σ_{n} (t)} (B_{n} (t)) = π_{σ_{n} (t), x} (J_{t} (y_{n} (t); x_{h})) \times π_{σ_{n} (t), y} (I_{n} (t))

where

π_{σ} (z) = (π_{σ, x} (z), π_{σ, y} (z))

Since

Y_{0}^{s}

is a

C^{2}

vector field transverse to

ℱ^{u}

, there exists

C_{1} > 0

such that

dist (π_{σ, x}, J) \leq C_{1} | π_{σ, x} (J) |

and

dist ((π_{σ, x} |_{J})^{- 1}, π_{σ, x} (J)) \leq C_{1} | π_{σ, x} (J) |

for any interval

J

which is tangent to

Y_{0}^{s}

and is contained in

Im φ_{σ}

. It is easy to see that

\begin{matrix} dist (h_{t} \circ Φ^{- t}, Φ^{t} (J_{0})) \leq 2 C_{1} \sum_{n = 0}^{n (t)} | π_{σ_{n} (t), x} (J_{t} (y_{n} (t); x_{h})) | \end{matrix}

(7)

for any

t \geq 0

. Hence, it is sufficient to show that the latter sum is bounded by a constant.

Fix

T_{1} > 0

so that

γ_{t} ⊄ Im φ_{σ}

for any

σ \in Σ

and

t \geq T_{1}

. We claim that

| φ_{σ_{n} (t), x} (I_{n} (t)) | \geq Δ_{1}

for any

t \geq T_{1}

and

n = 0, \dots, n (t)

. Suppose

| π_{σ_{n} (t), x} (I_{n} (t)) | < Δ_{1}

. Since

I_{n} (t)

is tangent to

Y_{t}^{u}

and

I_{n} (t) \cap φ_{σ_{n} (t)} ([- 1, 1]^{3})

contains

Φ^{t} \circ H_{t} (0, y_{n} (t))

, Proposition 4.2 implies that

I_{n} (t) \subset φ_{σ_{n} (t)} ([- 5 / 4, 5 / 4]^{3})

It contradicts that

I_{n} (t)

is a connected component of

γ_{t} \cap φ_{σ_{n} (t)} ([- 3 / 2, 3 / 2]^{3})

and

γ_{t} ⊄ Im φ_{σ_{n} (t)}

The claim implies

\begin{matrix} | π_{σ_{n} (t), x} (J_{t} (y_{n} (t); x_{h})) | \leq Δ_{1} \cdot Area (π_{σ_{n} (t)} (B_{n} (t))), \end{matrix}

(8)

where

Area

is the Lebesgue measure on

R^{2}

. Hence, it is sufficient to show that the intersection multiplicity of

{B_{n} (t) | σ_{n} (t) = σ}

is less than

8 C_{Σ}^{2}

for any

t \geq T_{1}

and

σ \in Σ

. In fact, the inequality ( 8 ) implies that the sum

\sum_{n = 0}^{n (t)} | π_{σ_{n} (t), x} (J_{t} (y_{n} (t); x_{h})) |

is bounded by

(8 C_{Σ}^{2}) \cdot (4 # Σ)

. The proof is completed by the inequality ( 7 ).

Suppose that intersection multiplicity of

{B_{n} (t) | σ_{n} (t) = σ}

is at least

8 C_{Σ}^{2}

for

t_{0} \geq T_{1}

and

σ_{0} \in Σ

. Then, there exist

(x_{0}, y_{0}) \in [- 2, 2]^{2}

and

s_{1}, s_{2} \in [- 2, 2]

such that

0 < s_{2} - s_{1} < (4 C_{Σ}^{2})^{- 1}

and

φ_{σ_{0}} (x_{0}, y_{0}, s_{i}) \in Φ^{t} (Im H_{t_{0}})

for

i = 1, 2

Put

L = φ_{σ_{0}} (x_{0} \times y_{0} \times [s_{1}, s_{2}])

. Since

L

is tangent to

X

and

∥ D Φ^{t} (X (z)) ∥ = ∥ X (Φ^{t} (z)) ∥ = 1

for any

z

, we have

| Φ^{- t_{0}} (L) | = | L | \leq (4 C_{Σ})^{- 1}

. Since

\partial Φ^{- t_{0}} (L)

is contained in a subset

Im H_{t_{0}}

φ_{σ_{*}} ([- 1, 1]^{3})

, we have

Φ^{- t_{0}} (L) \subset Im φ_{σ_{*}}

. In particular,

φ_{σ_{*}}^{- 1} \circ Φ^{- t_{0}} (L)

is an interval parallel to the

s

-axis. Such an interval intersects with

Im H_{t_{0}}

at most once. It contradicts

\partial Φ^{- t_{0}} (L) \subset Im H_{t_{0}}

Proof of Proposition 4.5 . Suppose ${Per}_{*}^{s} (Φ)$ is non-empty. Take periodic points $z_{*} = φ_{σ_{*}} (0, 0, 0) \in {Per}_{*} (Φ)$ and $z_{h} = φ_{σ_{*}} (x_{h}, y_{h}, s_{h}) \notin {Per}_{*} (Φ)$ , and an interval $J_{0} = φ_{σ_{*}} ([0, x_{h}] \times 0 \times 0)$ tangent to $Y_{0}^{s}$ as above. Notice that $Φ^{T_{*}} (J_{0}) \subset J_{0}$ and $\cap_{n \geq 0} Φ^{n T_{*}} (J_{0}) = {z_{*}}$ . Since $∥ D (Φ^{T_{*}} |_{J_{0}}) (z_{*}) ∥ = ∥ D {\hat{Φ}}^{T_{*}} |_{(E^{s} / T Φ) (z_{*})} ∥ = 1$ , we obtain $dist (Φ^{t} |_{J_{0}}, J_{0})$ tends to infinity as $t \to \infty$ . However, it contradicts Lemmas 4.8 and 4.9 . Therefore, ${Per}_{*}^{s} (Φ)$ is empty. Applying it to the flow $Φ^{- 1}$ , we obtain that ${Per}_{*}^{u} (Φ)$ also is.

References

A. Arroyo and F. Rodriguez Hertz, Homoclinic bifurcations and uniform hyperbolicity for three-dimensional flows. Ann. Inst. H. Poincaré Anal. Non Linéaire 20 (2003), 805–841.
M. Asaoka, Classification of regular and non-degenerate projectively Anosov diffeomorphisms on three dimensional manifolds, preprint, http://www.math.kyoto-u.ac.jp/ ~asaoka/papers/regular.pdf
M. Asaoka, A classification of three dimensional regular projectively Anosov flows. Proc. Japan Acad. Ser. A Math. Sci. 80 (2004), no. 10, 194–197 (2005).
C. Bonatti,L. J. Díaz,M. Viana, Dynamics beyond uniform hyperbolicity. A global geometric and probabilistic perspective. Encyclopaedia of Mathematical Sciences, 102. Mathematical Physics, III. Springer-Verlag, Berlin, 2005. xviii+384 pp.
J. Cantwell and L. Conlon, The theory of levels. Index theory of elliptic operators, foliations, and operator algebras (New Orleans, LA/Indianapolis, IN, 1986), 1–10, Contemp. Math., 70, AMS, Providence, RI, 1988.
J. Cantwell and L. Conlon, Reeb stability for noncompact leaves in foliated 3-manifolds. Proc. Amer.Math.Soc., 33, 1981.
J. Cantwell and L. Conlon, Endsets of exceptional leaves; a theorem of G.Duminy. Foliations: geometry and dynamics (Warsaw, 2000), 225–261, World Sci. Publishing, River Edge, NJ, 2002.
C. I. Doering, Persistently transitive vector fields on three-dimensional manifolds. Dynamical systems and bifurcation theory (Rio de Janeiro, 1985), 59–89, Pitman Res. Notes Math. Ser., 160, Longman Sci. Tech., Harlow, 1987.
Y. Eliashberg and W. Thurston, Confoliations. University Lecture Series, 13. Amer. Math. Soc., Providence, RI, 1998.
E. Ghys, Rigidité différentiable des groupes fuchiens, Inst. Hautes Études Sci. Publ. Math. 78 (1993), 163–185.
A. J. Homburg, Piecewise smooth interval maps with non-vanishing derivative. Ergodic Theory Dynam. Systems 20 (2000), 749–773.
R. Man͂é, Hyperbolicity, sinks and measure in one-dimensional dynamics. Comm. Math. Phys. 100 (1985), no. 4, 495–524. and Erratum. Comm. Math. Phys. 112 (1987), no. 4, 721–724.
Y. Mitsumatsu, Anosov flows and non-stein symplectic manifolds. Ann. Inst. Fourier 45 (1995), 1407–1421.
Y. Mitsumatsu, Foliations and contact structures on 3-manifolds. Foliations: geometry and dynamics (Warsaw, 2000), 75–125, World Sci. Publishing, River Edge, NJ, 2002.
W. de Melo, S. van Strien, Sebastian, One-dimensional dynamics. Ergebnisse der Mathematik und ihrer Grenzgebiete (3), 25. Springer-Verlag, Berlin, 1993.
T. Noda, Projectively Anosov flows with differentiable (un)stable foliations. Ann. Inst. Fourier 50 (2000), 1617–1647.
T. Noda, Regular projectively Anosov flows with compact leaves. Ann. Inst. Fourier (Grenoble) 54 (2004), no. 2, 353–363.
T. Noda, T. Tsuboi, Regular projectively Anosov flows without compact leaves. Foliations: geometry and dynamics (Warsaw, 2000), 403–419, World Sci. Publishing, River Edge, NJ, 2002.
E. Pujals and M. Sambarino, Homoclinic tangencies and hyperbolicity for surface diffeomorphisms. Ann. of Math. (2) 151 (2000), no. 3, 961–1023.
J.Palis and F.Takens, Hyperbolicity and sensitive chaotic dynamics at homoclinic bifurcations. Cambridge Studies in Advanced Mathematics, 35. Cambridge University Press, Cambridge, 1993.
T. Tsuboi, Regular projectively Anosov flows on the Seifert fibered 3-manifolds, J. Math. Soc. Japan.. 56 (2004), no. 4, 1233–1253.

Regular pro jectively Anosov flows on three manifolds

Masayuki ASAOKA * * Supported by Grant-in-Aid for Encouragement of Young Scientists (B) Department of Mathematics, Kyoto University Kyoto 606-8502, Japan e-mail:asaoka@math.kyoto-u.ac.jp

November 27, 2006