### November 27, 2006

Both authors are supported in part by National Science Foundation.
<ph f="cmbx">A </ph> <math xmlns="http://www.w3.org/1998/Math/MathML"> <mi>C</mi> <mi>r</mi> </math> <ph f="cmmi"> </ph><ph f="cmbx">Closing Lemma for a Class of Symplectic Diffeomorphisms</ph>

### Zhihong Xia & Hua Zhang

Department of Mathematics, Northwestern University, Evanston, IL 60208 E-mail address : xia@math.northwestern.edu & zhang@math.northwestern.edu
• Abstract. We prove a ${C}^{r}$  closing lemma for a class of partially hyperbolic symplectic diffeomorphisms. We show that for a generic ${C}^{r}$  symplectic diffeomorphism, $r=1,2,\dots ,$  , with two dimensional center and close to a product map, the set of all periodic points is dense.

1 Introduction and Main Result

One of the fundamental problems in dynamical systems is the so-called ${C}^{r}$  closing lemma. The problem goes back to Poincaré in his study of the restricted three body problem. It asks whether periodic points are dense for a typical symplectic or volume preserving diffeomorphism on a compact manifold. Let $M$  be a compact manifold, with either a symplectic or a volume form $\omega$  . Let ${\text{Diff}}_{\omega }^{r}\left(M\right)$  be the set of ${C}^{r}$  symplectic or volume-preserving diffeomorphisms on $M$  . A set in a topological space is said to be residual if it is the intersection of countably many open and dense subsets of of the topological space. A dynamical property is said to be ${C}^{r}$  generic on ${\text{Diff}}_{\omega }^{r}\left(M\right)$  if there is a residual set $R$  such that the property holds for every $f\in R$  . In the class symplectic and volume preserving diffeomorphisms, the closing lemma is the following conjecture:
Conjecture 1 (Closing Lemma for symplectic and volume-preserving diffeomorphisms).
Assume $M$  is compact. There exists a residual subset $R\subset {\text{Diff}}_{\omega }^{r}\left(M\right)$  such that if $f\in R$  , the set of periodic points $P=\left\{x\in M|{f}^{p}\left(x\right)=x,\text{for some integer}p\right\}$  is dense in $M$  .
Smale [12listed the problem as one of the mathematical problems for this century.
For $r=1$  , the above conjecture is proved to be true by Pugh [9and later improved to various cases by Pugh & Robinson [10. A different proof was given by Liao [5and Mai [6. For higher smoothness $r>1$  , besides the hyperbolic cases (the Anosov closing lemma, for uniformly hyperbolic and non-uniformly hyperbolic diffeomorphisms), there is no non-trivial results. On the other hand, example shows that the local perturbation method used in the proof of the ${C}^{1}$  closing lemma no longer works for the smoother case. New and global perturbation methods are required (Gutierrez [2). M. Herman [3has a counter example for the ${C}^{r}$  closing lemma with $r$  large for symplectic diffeomorphisms where the symplectic form is not exact.
In this paper, we prove a ${C}^{r}$  closing lemma for arbitrary positive integer $r$  for a class of partially hyperbolic symplectic diffeomorphisms.
A diffeomorphism $f:M\to M$  is partially hyperbolic if the tangent bundle $TM$  admits a $Tf$  invariant splitting $TM={E}^{u}\oplus {E}^{c}\oplus {E}^{s}$  and there is a Riemannian metric on $M$  such that there exist real numbers ${\lambda }_{1}>{\lambda }_{2}>1>{\mu }_{2}>{\mu }_{1}>0$  satisfying $m\left(Tf{|}_{{E}^{u}}\right)\ge {\lambda }_{1}>{\lambda }_{2}\ge \parallel Tf{|}_{{E}^{c}}\parallel \ge m\left(Tf{|}_{{E}_{c}}\right)\ge {\mu }_{2}>{\mu }_{1}\ge \parallel Tf{|}_{{E}^{s}}\parallel >0.$  Here the co-norm $m\left(A\right)$  of a linear operator $A$  between two Banach spaces is defined by $m\left(A\right):=in{f}_{\parallel v\parallel =1}\parallel A\left(v\right)\parallel =\parallel {A}^{-1}{\parallel }^{-1}$  .
To avoid triviality, we assume at least two of the subbundles are non-zero. Partial hyperbolicity is a ${C}^{1}$  open condition, as can be easily verified by it's associated invariant cone fields.
We remark that our definition of partial hyperbolicity here is not the most general one. One can allow the parameters ${\lambda }_{1},{\lambda }_{2},{\mu }_{1},{\mu }_{2}$  to depend on each trajectory in general cases. The systems that we are considering satisfy the definition given here.
For symplectic cases, the stable distribution ${E}^{s}$  and the unstable distribution ${E}^{u}$  have the same dimension. Moreover, one can choose the parameters such that ${\lambda }_{1}={\mu }_{1}^{-1}$  and ${\lambda }_{2}={\mu }_{2}^{-1}$  .
We are now ready to state our main theorem.
Theorem 1.1. Let ${M}_{1}$  be a compact symplectic manifold and ${f}_{1}\in {\text{Diff}}_{{\omega }_{1}}^{r}\left({M}_{1}\right)$  an Anosov diffeomorphism. Let ${M}_{2}$  be a compact symplectic surface (orientable surface) with an area form ${\omega }_{2}$  and let ${f}_{2}\in {\text{Diff}}_{{\omega }_{2}}^{r}\left({M}_{2}\right)$  be an area preserving diffeomorphism on ${M}_{2}$  . Let $\omega ={\omega }_{1}+{\omega }_{2}$  be the symplectic form defined on ${M}_{1}×{M}_{2}$  . Assume that ${f}_{1}$  dominates ${f}_{2}$  , i.e., ${f}_{1}×{f}_{2}:{M}_{1}×{M}_{2}\to {M}_{1}×{M}_{2}$  is partially hyperbolic with $T{M}_{2}$  as its center splitting. Then there exists a neighborhood $U$  of ${f}_{1}×{f}_{2}$  in ${\text{Diff}}_{\omega }^{r}\left({M}_{1}×{M}_{2}\right)$  and a residual subset $R\in U$  such that for any $g\in U$  , the set of periodic points of $g$  is dense in ${M}_{1}×{M}_{2}$  .
The proof took advantage of the partial hyperbolicity and a recent result of Xia [15on surface diffeomorphisms.

2 Partial Hyperbolicity and Symplectic Structure

For a ${C}^{r}$  partially hyperbolic diffeomorphism $f$  , the stable and unstable bundles are uniquely integrable and are tangent to foliations ${W}_{f}^{s}$  and ${W}_{f}^{u}$  with ${C}^{r}$  leaves.
$f$  is dynamically coherent if the distributions ${E}^{c}$  , ${E}^{c}\oplus {E}^{s}$  and ${E}^{c}\oplus {E}^{u}$  are integrable, they integrate to foliations ${W}_{f}^{c}$  , ${W}_{f}^{cs}$  and ${W}_{f}^{cu}$  respectively and ${W}_{f}^{c}$  and ${W}_{f}^{s}$  sub-foliate ${W}_{f}^{cs}$  , ${W}_{f}^{c}$  and ${W}_{f}^{u}$  sub-foliate ${W}_{f}^{cu}$  .
We have the following proposition from Pugh & Shub [11.
Proposition 2.1. Let $f$  be a partially hyperbolic diffeomorphism. If the center foliation ${W}_{f}^{c}$  exists and is of class ${C}^{1}$  , then $f$  is stably dynamically coherent, i.e., any $g$  which is ${C}^{1}$  sufficiently close to $f$  is dynamically coherent.
Let $W$  be a foliation of a compact smooth manifold $M$  whose leaves are ${C}^{r}$  immersed submanifolds of dimension $k$  . For $x\in M$  , we call a set $P\left(x\right)\subset W\left(x\right)$  a ${C}^{r}$  plaque of $W$  at $x$  if $P\left(x\right)$  is the image of a ${C}^{r}$  embedding of the unit ball $B\subset {\mathbb{R}}^{k}$  into $W\left(x\right)$  . A plaquation $\text{P}$  for $W$  is a collection of plaques such that every point $x\in M$  is contained in a plaque $P\in \text{P}$  .
Let $f$  be a diffeomorphism such that $W$  is invariant under $f$  . A pseudo orbit $\left\{{x}_{n}{\right\}}_{n\in \mathbb{Z}}$  respects $\text{P}$  if for each $n$  , $f\left({x}_{n}\right)$  and ${x}_{n+1}$  lie in a common plaque $P\in \text{P}$  . $f$  is called plaque expansive with respect to $W$  if there exists $\epsilon >0$  such that if two $\epsilon$  -pseudo orbits $\left\{{x}_{n}\right\}$  and $\left\{{y}_{n}\right\}$  both respect $\text{P}$  and $d\left({x}_{n},{y}_{n}\right)<\epsilon$  for all $n$  , then ${x}_{n}$  and ${y}_{n}$  lie in a common plaque for all $n$  .
Hirsch, Pugh and Shub [4proved that plaque expansiveness with respect to the center foliation of a partially hyperbolic diffeomorphism is a ${C}^{1}$  open property and is satisfied when we have smooth center foliation ${W}^{c}$  (Theorem 7.1 and 7.2 in [4).
It is clear that under the condition of our main theorem, $f={f}_{1}×{f}_{2}$  is partially hyperbolic with smooth center foliation, so there exists neighborhood $U$  of ${f}_{1}×{f}_{2}$  in ${\text{Diff}}_{\omega }^{r}\left({M}_{1}×{M}_{2}\right)$  such that any $g\in U$  is partially hyperbolic, dynamically coherent and plaque expansive with respect to the center foliation ${W}_{g}^{c}$  .
Niţ icǎ and Török in [13proved the following
Proposition 2.2. Let $M$  be a compact manifold with a smooth volume form $\mu$  , if $f\in {\text{Diff}}_{\mu }^{1}\left(M\right)$  is partially hyperbolic, dynamically coherent and plaque expansive with respect to the center foliation ${W}_{f}^{c}$  , then the periodic center leaves of $f$  are dense in $M$  .
Now we have the following
Lemma 2.3. Under the condition of our main theorem, there exists neighborhood $U$  of ${f}_{1}×{f}_{2}$  in ${\text{Diff}}_{\omega }^{r}\left({M}_{1}×{M}_{2}\right)$  such that any $g\in U$  is partially hyperbolic, dynamically coherent and the periodic center leaves of $g$  are dense in ${M}_{1}×{M}_{2}$  .
The proof is a simple application of the above results and we only have to note that a symplectic diffeomorphism trivially support an invariant smooth volume form.
The following proposition is also from Niţ icǎ and Török [13.
Proposition 2.4. For a partially hyperbolic diffeomorphism on a compact manifold $M$  with center-stable and center-unstable foliations ${W}^{cs}$  and ${W}^{cu}$  , we have the following local product structure property:
There exist constants $\epsilon >0$  , $\delta >0$  and $K>1$  such that for any $x,y\in M$  with $d\left(x,y\right)<\epsilon$  , the following hold, 1) ${W}_{\delta }^{s}\left(x\right)$  and ${W}_{\delta }^{cu}\left(y\right)$  intersect at a unique point ${z}_{1}$  , ${W}_{\delta }^{u}\left(x\right)$  and ${W}_{\delta }^{cs}\left(y\right)$  intersect at a unique point ${z}_{2}$  , and moreover $max\left(d\left(x,{z}_{1}\right),d\left(y,{z}_{1}\right)\right)  , $max\left(d\left(x,{z}_{2}\right),d\left(y,{z}_{2}\right)\right)  .
2) ${W}_{\delta }^{cs}\left(x\right)$  and ${W}_{\delta }^{cu}\left(y\right)$  intersect transversally, same is true for ${W}_{\delta }^{cs}\left(y\right)$  and ${W}_{\delta }^{cu}\left(x\right)$  .
3) ${W}_{\delta }^{cs}\left(x\right)\cap {W}_{\delta }^{cu}\left(x\right)={W}_{\delta }^{c}\left(x\right)$  and ${W}_{\delta }^{cs}\left(y\right)\cap {W}_{\delta }^{cu}\left(y\right)={W}_{\delta }^{c}\left(y\right)$  .
Theorem 6.1 of [4tells us that $\epsilon$  , $\delta$  and $K$  are lower semi-continuous with respect to the ${C}^{1}$  topology on ${\text{Diff}}^{1}\left(M\right)$  .
We will need a result for symplectic partially hyperbolic diffeomorphisms.
Lemma 2.5. Let $f$  be a symplectic partially hyperbolic diffeomorphism on a compact symplectic manifold $M$  , suppose we have the center foliation ${W}_{f}^{c}$  , then center manifolds of $f$  are symplectic submanifolds and the restrictions of $f$  on invariant center leaves are symplectic diffeomorphisms.
$\text{Proof}$  . For symplectic partially hyperbolic diffeomorphism $f$  , there exist $\lambda >\mu >1$  such that $m\left(Tf{|}_{{E}^{u}}\right)\ge \lambda >\tau \ge \parallel Tf{|}_{{E}^{c}}\parallel \ge m\left(Tf{|}_{{E}_{c}}\right)\ge {\tau }^{-1}>{\lambda }^{-1}\ge \parallel Tf{|}_{{E}^{s}}\parallel >0.$  Denote by $\omega$  the symplectic form on $M$  . Let $W\subset M$  be a center leaf, we should prove $\left(W,\omega {|}_{W}\right)$  is a symplectic manifold, i.e., $\omega {|}_{W}$  is a non-degenerate, closed two form on $W$  . Closeness is obvious since $\omega$  is closed on $M$  .
Suppose that $\omega {|}_{W}$  is degenerate, then there exists $x\in W$  and a unit vector $u\in {T}_{x}W$  such that for all ${v}_{c}\in {T}_{x}W$  , $\omega \left(u,{v}_{c}\right)=0$  .
We have the splitting ${T}_{x}M={E}_{x}^{s}\oplus {E}_{x}^{c}\oplus {E}_{x}^{u}$  , any $v\in {T}_{x}M$  can be written as $v={v}_{s}+{v}_{c}+{v}_{u}$  , where ${v}_{s}\in {E}_{x}^{s}$  , ${v}_{u}\in {E}_{x}^{u}$  and ${v}_{c}\in {E}_{x}^{c}={T}_{x}W$  .
We have $\omega \left(u,v\right)=\omega \left(u,{v}_{s}\right)+\omega \left(u,{v}_{c}\right)+\omega \left(u,{v}_{u}\right)$  and by the way $u$  was chosen, $\omega \left(u,{v}_{c}\right)=0$  .
There exists $K>0$  such that $|\omega \left({w}^{1},{w}^{2}\right)|  for arbitrary pair of unit vectors ${w}^{1},{w}^{2}\in {T}_{x}M$  .
Now we know $\omega \left(u,{v}_{s}\right)=0$  because $|\omega \left(u,{v}_{s}\right)|=|\omega \left(T{f}^{n}\left(u\right),T{f}^{n}\left({v}_{s}\right)|\le \left(\frac{\tau }{\lambda }{\right)}^{n}\parallel {v}_{s}\parallel |\omega \left({u}^{n},{v}_{s}^{n}\right)|\le K\left(\frac{\tau }{\lambda }{\right)}^{n}\parallel {v}_{s}\parallel ⟶0$  as $n⟶+\infty$  .
Here ${u}^{n}=T{f}^{n}\left(u\right)/\parallel T{f}^{n}\left(u\right)\parallel$  , ${v}_{s}^{n}=T{f}^{n}\left({v}_{s}\right)/\parallel T{f}^{n}\left({v}_{s}\right)\parallel$  .
Similarly, we have $\omega \left(u,{v}_{u}\right)=0$  and hence $\omega \left(u,v\right)=0$  for any $v\in {T}_{x}M$  , this contradicts the fact that $\omega$  is non-degenerate on $M$  .
So $\left(W,\omega {|}_{W}\right)$  is a symplectic submanifold and if $W$  is invariant under $f$  , $f{|}_{W}$  preserves $\omega {|}_{W}$  and hence is a symplectic diffeomorphism on $W$  .

3 Some generic properties for area-preserving diffeomorphisms on surfaces

To prove our main theorem, we need some generic properties for surface diffeomorphisms.
Let $S$  be a compact surface, denote by ${\text{Diff}}_{\mu }^{r}\left(S\right)$  the set of area-preserving ${C}^{r}$  diffeomorphisms. For $f\in {\text{Diff}}_{\mu }^{r}\left(S\right)$  , denote by $HP\left(f\right)$  the set of hyperbolic periodic points of $f$  . The following generic property was first proved by Mather [7for maps on two sphere ${S}^{2}$  and later generalized to arbitrary compact surfaces by Oliveira [8.
Proposition 3.1. There is a residual subset $R\in {\text{Diff}}_{\mu }^{r}\left(S\right)$  such that if $f\in R$  and $p\in HP\left(f\right)$  is a hyperbolic periodic point of $f$  , then $\overline{{W}_{f}^{s}\left(p\right)}=\overline{{W}_{f}^{u}\left(p\right)}.$
We remark that if $r=1$  , the proposition is true for generic symplectic and volume preserving diffeomorphisms on any compact manifolds (cf. Xia [14).
The next Theorem is due to Xia [15, extending a recent result of Franks & Le Calvez [1on two sphere.
Theorem 3.2. Let $S$  be a compact orientable surface and $\mu$  be an area form on $S$  . For any positive integer $r$  , there is a residual subset $R\in {\text{Diff}}_{\mu }^{r}\left(S\right)$  such that if $f\in R$  , then both the sets ${\cup }_{p\in HP\left(f\right)}{W}^{s}\left(p\right)$  and ${\cup }_{p\in HP\left(f\right)}{W}^{u}\left(p\right)$  are dense in $S$  . Moreover, if an open set $U\subset S$  contains no periodic point, then there is a hyperbolic periodic point $p\in HP\left(f\right)$  such that both the stable and unstable manifold of $p$  is dense in $U$  .
The proof uses prime end compactification and a rich literature on area preserving surface diffeomorphisms.

4 Proof of the Main Theorem

Our main perturbation lemma is from Xia [14.
Lemma 4.1. Let $M$  be a compact symplectic manifold and $f\in {\text{Diff}}_{\omega }^{r}\left(M\right)$  , there exist ${\epsilon }_{0}>0$  and $c>0$  such that for any $g\in {\text{Diff}}_{\omega }^{r}\left(M\right)$  such that $\parallel f-g{\parallel }_{{C}^{r}}<{\epsilon }_{0}$  and any $0<\epsilon \le {\epsilon }_{0}$  , $0<\delta \le {\epsilon }_{0}$  , if $x,y\in M$  and $d\left(x,y\right)  , there exists ${g}_{1}\in {\text{Diff}}_{\omega }^{r}\left(M\right)$  , $\parallel {g}_{1}-g{\parallel }_{{C}^{r}}<\epsilon$  satisfies ${g}_{1}{g}^{-1}\left(x\right)=y$  , ${g}_{1}\left(z\right)=g\left(z\right)$  for all $z/\in {g}^{-1}\left({B}_{\delta }\left(x\right)\right)$  , and ${g}_{1}^{-1}\left(z\right)={g}^{-1}\left(z\right)$  for all $z/\in {B}_{\delta }\left(x\right)$  .
Now we are ready to prove the main theorem.
$\text{Proof}$  . By Lemma 2.3, there exists a neighborhood $U$  of ${f}_{1}×{f}_{2}$  in ${\text{Diff}}_{\omega }^{r}\left({M}_{1}×{M}_{2}\right)$  such that any $g\in U$  is partially hyperbolic, dynamically coherent and the periodic center leaves of $g$  are dense in ${M}_{1}×{M}_{2}$  .
Now for any fixed $g\in U$  , suppose there is a periodic point free open set $V\subset {M}_{1}×{M}_{2}$  , we show that by an arbitrarily small perturbation, we can create a periodic point in $V$  . It is clear that the main theorem will follow.
Since periodic center leaves of $g$  are dense, by Proposition 2.4, we can find two periodic center leaves ${W}_{1}$  and ${W}_{2}$  which are sufficiently close such that there exist ${x}_{1},{y}_{1}\in {W}_{1}\cap V$  , ${x}_{2},{y}_{2}\in {W}_{2}\cap V$  with $z={W}_{\delta }^{u}\left({x}_{1}\right)\cap {W}_{\delta }^{s}\left({x}_{2}\right)\in V$  , $w={W}_{\delta }^{s}\left({y}_{1}\right)\cap {W}_{\delta }^{u}\left({y}_{2}\right)\in V$  . ${W}_{1}$  and ${W}_{2}$  are compact surfaces.
By taking certain power of $g$  we may assume that ${W}_{1}$  and ${W}_{2}$  are invariant under $g$  . From Lemma 2.5, ${W}_{1}$  and ${W}_{2}$  are symplectic submanifolds, ${g}^{1}=g{|}_{{W}_{1}}$  and ${g}^{2}=g{|}_{{W}_{2}}$  are symplectic diffeomorphisms. By making an arbitrarily small perturbation, we may assume that ${g}^{1}$  and ${g}^{2}$  satisfy the generic condition in Theorem 3.2. Now ${W}_{i}\cap V$  is a periodic point free open set in ${W}_{i}$  , we know that there exists ${p}_{i}\in HP\left({g}^{i}\right)$  such that ${W}_{{g}^{i}}^{u}\left({p}_{i}\right)$  and ${W}_{{g}^{i}}^{s}\left({p}_{i}\right)$  are both dense in ${W}_{i}\cap V$  , where $i=1,2$  .
Note that ${p}_{1}$  and ${p}_{2}$  are hyperbolic periodic points of $g$  .
We will show that by an arbitrarily small perturbation, we can change $z$  and $w$  into heteroclinic points of hyperbolic periodic points ${p}_{1}$  and ${p}_{2}$  and get a heteroclinic loop. As a result, there will be periodic points in arbitrary neighborhoods of $z$  and $w$  , including $V$  .
For arbitrary $\eta >0$  prescribed as the size of the perturbation, take $\epsilon$  such that $0<\epsilon   , where ${\epsilon }_{0}$  is from Lemma 4.1.
Since $z\in {W}^{u}\left({W}_{1}\right)$  and $z/\in {W}_{1}$  , there exists $\alpha$  with $0<\alpha <{\epsilon }_{0}$  , such that ${B}_{\alpha }\left(z\right)\cap {W}_{1}=\varnothing$  and ${g}^{-n}\left(z\right)/\in {B}_{\alpha }\left(z\right)$  for all $n\in \mathbb{N}$  .
Fix this $\alpha$  , there exists $\beta$  with $0<\beta <\alpha$  such that for all $\stackrel{~}{z}\in {W}^{u}\left({W}_{1}\right)$  with $d\left(\stackrel{~}{z},z\right)<\beta$  , we have ${g}^{-n}\left(\stackrel{~}{z}\right)/\in {B}_{\frac{\alpha }{2}}\left(z\right)$  for all $n\in \mathbb{N}$  .
Fix this $\beta$  , there exists $\gamma$  with $0<\gamma   , such that for all ${z}_{1}\in {W}^{u}\left({W}_{1}\right)$  with $d\left({z}_{1},z\right)<\gamma$  , we have ${B}_{\frac{\alpha }{4}}\left({z}_{1}\right)\subset {B}_{\frac{\alpha }{2}}\left(z\right)$  and hence ${g}^{-n}\left({z}_{1}\right)/\in {B}_{\frac{\alpha }{4}}\left({z}_{1}\right)$  for all $n\in \mathbb{N}$  .
By continuity of the unstable foliation, there exists $\nu >0$  such that for all ${\stackrel{~}{x}}_{1}\in {W}_{1}$  with $d\left({\stackrel{~}{x}}_{1},{x}_{1}\right)<\nu$  , there exists ${z}_{1}\in {W}_{g}^{u}\left({\stackrel{~}{x}}_{1}\right)$  such that $d\left({z}_{1},z\right)<\gamma$  .
Since ${W}_{{g}^{1}}^{u}\left({p}_{1}\right)$  is dense in ${W}_{1}\cap V$  , there exists ${\stackrel{~}{x}}_{1}\in {W}_{{g}^{1}}^{u}\left({p}_{1}\right)$  with $d\left({\stackrel{~}{x}}_{1},{x}_{1}\right)<\nu$  and hence there is a ${z}_{1}\in {W}_{g}^{u}\left({\stackrel{~}{x}}_{1}\right)$  such that $d\left({z}_{1},z\right)<\gamma$  .
Now we can use the perturbation lemma 4.1 for $g$  using the parameters $0<\epsilon <{\epsilon }_{0}$  and $0<\delta =\frac{\alpha }{4}<{\epsilon }_{0}$  . We have $d\left({z}_{1},z\right)<\gamma   , so there exists ${g}_{1}\in {\text{Diff}}_{\omega }^{r}\left({M}_{1}×{M}_{2}\right)$  , $\parallel {g}_{1}-g{\parallel }_{{C}^{r}}<\epsilon$  satisfies ${g}_{1}{g}^{-1}\left({z}_{1}\right)=z$  , ${g}_{1}\left(x\right)=g\left(x\right)$  for all $x/\in {g}^{-1}\left({B}_{\delta }\left({z}_{1}\right)\right)$  , and ${g}_{1}^{-1}\left(y\right)={g}^{-1}\left(y\right)$  for all $y/\in {B}_{\delta }\left({z}_{1}\right)$  . We check that after the perturbation, $z\in {\stackrel{~}{W}}_{{g}_{1}}^{u}\left({p}_{1}\right)$  , where ${\stackrel{~}{W}}_{{g}_{1}}^{u}\left({p}_{1}\right)$  stands for the unstable manifold of the hyperbolic periodic point ${p}_{1}$  for ${g}_{1}$  , not the leaf of the unstable foliation containing ${p}_{1}$  in the partially hyperbolic setting.
It is clear that ${g}_{1}^{-1}\left(z\right)={g}^{-1}\left({z}_{1}\right)$  and since ${g}^{-n}\left({z}_{1}\right)/\in {B}_{\delta }\left({z}_{1}\right)$  for all $n\in \mathbb{N}$  , ${g}_{1}^{-n}\left(z\right)={g}^{-n}\left({z}_{1}\right)$  for all $n\in \mathbb{N}$  . Moreover, ${g}_{1}^{-n}\left({p}_{1}\right)={g}^{-n}\left({p}_{1}\right)$  for all $n\in \mathbb{N}$  since ${B}_{\delta }\left({z}_{1}\right)\cap {W}_{1}=\varnothing$  .
Hence we have $d\left({g}_{1}^{-n}\left(z\right),{g}_{1}^{-n}\left({p}_{1}\right)\right)=d\left({g}^{-n}\left({z}_{1}\right),{g}^{-n}\left({p}_{1}\right)\right)$  $\le d\left({g}^{-n}\left({z}_{1}\right),{g}^{-n}\left({\stackrel{~}{x}}_{1}\right)\right)+d\left({g}^{-n}\left({\stackrel{~}{x}}_{1}\right),{g}^{-n}\left({p}_{1}\right)\right)⟶0$  as $n⟶+\infty$  .
This shows $z\in {\stackrel{~}{W}}_{{g}_{1}}^{u}\left({p}_{1}\right)$  . The two terms above both go to $0$  as $n$  goes to $+\infty$  since ${z}_{1}\in {W}_{g}^{u}\left({\stackrel{~}{x}}_{1}\right)$  and ${\stackrel{~}{x}}_{1}\in {W}_{{g}^{1}}^{u}\left({p}_{1}\right)$  .
Similarly we can use a perturbation of size less than $\epsilon$  to make $z$  on the stable manifold of ${p}_{2}\in {W}_{2}$  . Two more of these perturbations will make $w$  in the intersection of stable manifold of ${p}_{1}$  and unstable manifold of ${p}_{2}$  . Finally by a perturbation of size less than $4\epsilon =\eta$  , we have the desired heteroclinic loop. This concludes our proof.
References

1. J. Franks and P. Le Calvez. Regions of instability for non-twist maps. Ergodic Theory Dynam. Systems, 23(1):111–141, 2003.
2. C. Gutierrez. A counter-example to a ${c}^{2}$  closing lemma. Ergodic Theory & Dynamical Systems, 7(4):509–530, 1987.
3. M. Herman. Exemples de flots hamiltoniens dont aucune perturbation en topologie ${c}^{\infty }$  n'a d'orbites périodiques sur un ouvert de surfaces d'énergie. C.R. Acad. Sci. Paris, t., 312:989–994, 1991.
4. M. Hirsch, C. Pugh, and M. Shub. Invariant manifolds, Lect. Notes in Math., volume 583. Springer-Verlag, Berlin-New York, 1977.
5. S.T. Liao. An extension of the ${c}^{1}$  closing lemma. Acta Sci. Natur. Univ. Pekinensis, 2:1–41, 1979.
6. J. Mai. A simpler proof of ${c}^{1}$  closing lemma. Scientia Sinica, 10:1021–1031, 1986.
7. J. Mather. Topological proofs of some purely topological consequences of carathéodory's theory of prime ends. in Selected Studies. Eds. Th. M. Rassias and G. M. Rassias, pages 225–255, 1982.
8. F. Oliveira. On the generic existence of homoclinic points. Ergod. Th. & Dynam. Sys., 7:567–595, 1987.
9. C. Pugh. The closing lemma. Amer. J. Math., 89:956–1021, 1967.
10. C. Pugh and C. Robinson. The ${c}^{1}$  closing lemma, including hamiltonians. Ergod. Th. & Dynam. Sys., 3:261–313, 1983.
11. C. Pugh and M. Shub. Stably ergodic dynamical systems and partial hyperbolicity. J. of Complexity, 13:125–179, 1997.
12. S. Smale. Mathematical problems for the next century. Math. Intelligencer, 20(2):7–15, 1998.
13. V. Niţ icǎ and A. Török. An open dense set of stably ergodic diffeomorphisms in a neighborhood of a non-ergodic one. Topology, 40:259–278, 2001.
14. Z. Xia. Homoclinic points in symplectic and volume-preserving diffeomorphism. Commun. Math. Phys., 177:435–449, 1996.
15. Z. Xia. Area-preserving surface diffeomorphisms. Preprint, Mathematics ArXiv: math.DS/0503223, 2004.

Department of Mathematics, Northwestern University, Evanston, IL 60208 E-mail address : xia@math.northwestern.edu & zhang@math.northwestern.edu