Hyperbolic Sets That are Not Locally Maximal
This paper addresses the following topics relating to the structure of hyperbolic sets: First, hyperbolic sets that are not contained in locally maximal hyperbolic sets. Second, the existence of a Markov partition for a hyperbolic set. We construct new examples of hyperbolic sets which are not contained in locally maximal hyperbolic sets. The first example is robust under perturbations and can be constructed on any compact manifold of dimension greater than one. The second example is robust, topologically transitive, and constructed on a 4-dimensional manifold. The third example is volume preserving and constructed on R4. We show that every hyperbolic set is included in a hyperbolic set with a Markov partition. Additionally, we describe a condition that ensures a hyperbolic set is included in a locally maximal hyperbolic set.
Hyperbolic Sets That are Not Locally Maximal Todd Fisher December 6, 2004 Abstract This papers addresses the following topics relating to the structure of hyperbolic sets: First, hyperbolic sets that are not contained in locally maximal hyperbolic sets. Second, the existence of a Markov partition for a hyperbolic set. We construct new examples of hyperbolic sets which are not contained in locally maximal hyperbolic sets. The first example is robust under perturbations and can be constructed on any compact manifold of dimension greater than one. The second example is robust, topologically transitive, and constructed on a 4-dimensional manifold. The third example is volume preserving and constructed on R4. We show that every hyperbolic set is included in a hyperbolic set with a Markov partition. Additionally, we describe a condition that ensures a hyperbolic set is included in a locally maximal hyperbolic set. 1 Introduction A general field of study in dynamical systems is examining complicated, or chaotic dynamics. A recipe for complicated dynamics is expansion plus recurrence. In the 1960 s, Anosov [2] and Smale [12] began studying compact invariant sets called hyperbolic sets, whose tangent space splits into invariant uniformly contracting and uniformly expanding directions. On a compact manifold these sets often possess a great deal of recurrence and complicated dynamics. The pioneering article by Smale [12] states many of the standard results for hyperbolic sets. There are many classically studied examples of hyperbolic sets, including hyperbolic toral automorphisms, solenoids, hyperbolic attractors, and horseshoes. These examples show that the structure of hyperbolic sets can be very rich. All of these classical examples possess a useful property called local maximality. We recall the definition of locally maximal hyperbolic sets. Let f : M ! M be a diffeomorphism of a compact smooth manifold M. A hyperbolic set is called locally maximal (or isolated) if there exists a neighborhood V of in M such that = T n2Z fn(V ). Locally maximal hyperbolic sets possess many useful properties; a few are the following: 1 1. Locally maximal hyperbolic sets have a local product structure. 2. Locally maximal hyperbolic sets have a Markov partition. 3. Locally maximal hyperbolic sets with positive entropy have only finitely many invariant probability measures maximizing entropy. The structure of locally maximal hyperbolic sets is well understood at this point. In [5, p. 272] Katok and Hasselblatt pose the following: Question 1.1 Let be a hyperbolic set, and let V be an open neighborhood of . Does there exist a locally maximal hyperbolic set such that V ? A related question is: Question 1.2 Given a hyperbolic set does there exist a neighborhood V of such that = T n2Z fn(V ) is a locally maximal hyperbolic set containing ? It appears that Question 1.1 was originally asked by Alexseyev in the 1960 s. These questions remained open until 2001, when Crovisier [4] showed there is a diffeomorphism f of the 4-torus with a hyperbolic set , such that is not contained in any locally maximal hyperbolic set. Crovisier s construction, based on the elegant example of Shub in [6], is delicate, and leaves open the question of whether such examples are somehow unusual. Immediate questions raised by this construction are: 1. Does there exist an open set U (in the C1 topology) of diffeomorphisms such that every f 2 U possesses an invariant hyperbolic set that is not contained in a locally maximal one? 2. Does there exist an example answering Question 1.2 in the negative on other manifolds, in a lower dimension, on all manifolds? 3. Does there exist a topologically transitive hyperbolic set that is not contained in a locally maximal one? 4. Does there exist a volume-preserving diffeomorphism with a hyperbolic set that is not contained in a locally maximal one? In Section 3 we answer the first two of these questions affirmatively, using techniques different from Crovisier s. Theorem 1.3 Let M be a compact smooth manifold and U be the set of all diffeomorphisms f of M such that there exists a hyperbolic set of f which is not contained in a locally maximal hyperbolic set. If dim(M) 2, then int(U) 6= ;. 2 The hyperbolic sets in Crovisier s example and those constructed in Section 3 are all contained in hyperbolic sets with a Markov partition. Sinai originated the use of Markov partitions in dynamical systems. He constructed Markov partitions for Anosov diffeomorphisms in [11] and [10]. Bowen later extended these results to the existence of a Markov partition for a locally maximal hyperbolic set [3]. A Markov partition allows a symbolic representation of dynamics. This raises the following question, which is a weakened version of Question 1.2. Question 1.4 Let be a hyperbolic set for a diffeomorphism f : M ! M and V an open neighborhood of . Does there exist a hyperbolic set with a Markov partition such that V ? Using a Shadowing Theorem due to Katok [1] and adapting Bowen s construction [3] we are able to show the following: Theorem 1.5 If is a hyperbolic set and V is a neighborhood of , then there exists a hyperbolic set with a Markov partition such that V . In Section 5 we use Theorems 1.3 and 1.5 to extend the results for hyperbolic sets not contained in any locally maximal hyperbolic sets. The example is four dimensional, robust, and topologically transitive. For each y 2 M let O+(y) = {fn(y) | n 2 N} and O (y) = {fn | n 2 N} be respectively, the forward orbit and backward orbit of y. A point y 2 is transitive in if cl(O+(y)) = and cl(O (y)) = , and the set is transitive if contains a transitive point. Theorem 1.6 Let M be a compact smooth surface and T2 the flat 2-torus. On the compact manifold M T2 there exists a C1 open set of diffeomorphisms, U, such that for any F 2 U there exists a transitive hyperbolic set for F that is not contained in a locally maximal hyperbolic set. The proofs of Theorems 1.3 and 1.6 use diffeomorphisms that do not preserve any smooth volume form. The arguments can be modified to show the following: Theorem 1.7 There exists a symplectic diffeomorphism of R4 that contains a hyperbolic set that is not contained in a locally maximal hyperbolic set. Notice, in the above theorem we obtain a symplectic diffeomorphism, but with the loss of compactness for M and robustness of the construction. The proof of Theorem 1.7 uses classical techniques from Hamiltonian dynamics. The idea is to embed a diffeomorphism as a subsystem of a twist map. 2 Background We now give some useful definitions and results concerning hyperbolic sets. Let M be a smooth manifold, U M an open set, and f : U ! M a C1 diffeomorphism onto its image. 3 Definition: A compact f-invariant set M is called a hyperbolic set for f if there is a Df-invariant splitting T M = Eu Es and positive constants C and < 1 such that, for any point x 2 and any n 2 N, satisfies: kDfn x vk C nkvk, for v 2 Esx , and kDf n x vk C nkvk, for v 2 Eu x . Note, it is always possible to make a smooth change of the metric near the hyperbolic set so that C = 1. Such a metric is called an adapted metric. For > 0 sufficiently small and x 2 the local stable and unstable manifolds are respectively: Ws (x, f) = {y 2 M | for all n 2 N, d(fn(x), fn(y)) }, and Wu (x, f) = {y 2 M | for all n 2 N, d(f n(x), f n(y)) }. The stable and unstable manifolds are respectively: Ws(x, f) = S n 0 f n (Ws (fn(x), f)) , and Wu(x, f) = S n 0 fn (Wu (f n(x), f)) . For a hyperbolic set for a Cr diffeomorphism f the stable and unstable manifolds are Cr injectively immersed submanifolds characterized by uniform contraction and uniform expansion, respectively. For two points p and q in a hyperbolic set denote the set of points in the transverse intersection of Ws(p) and Wu(q) as Ws(p) t Wu(q). Two hyperbolic periodic points p and q contain a transverse heteroclinic point if Ws(p) t Wu(q) 6= ;. We will often extend a hyperbolic set, , by adding the orbit of a transverse heteroclinic point between two periodic points contained in . Denote the orbit of a point z under a map f by Of (z). Lemma 2.1 Let be a hyperbolic set for a diffeomorphism f containing periodic points p and q of the same index. If there exists a point z 2 Ws(p) t Wu(q), then 0 = [ Of (z) is a hyperbolic set. Proof. For x 2 keep the splitting of TxM to be the same for 0. Let Es fn(z) = Tfn(z)(Ws(p)) and Euf n(z) = Tfn(z)(Wu(q)). The Lambda Lemma [9, p. 203] shows the above will be a continuous splitting on 0. To see that the vectors in Es x0 are uniformly contracted for any x0 2 0 we first start with an adapted metric for . By continuity of Df there is a neighborhood of U of p and V of q such that kDffn(z)|Es fn(z)k < and kDf 1 fn(z)|Euf n(z)k < for fn(z) 2 U [ V . Since there are only a finite number of fn(z) outside of U [ V , it follows that there exists a C 1 such that kDffn(z)|Es fn(z)k < C and kDf 1 fn(z)|Euf n(z)k < C for all n 2 Z. 2 A consequence of hyperbolicity is that the orbit of a point depends sensitively on the initial position. A remarkable result is that approximate periodic orbits are close to real orbits. 4 Let (X, d) be a metric space, U M open and f : U ! M. For a 2 Z [ { 1} and b 2 Z [ {1} a sequence {xn}a<n<b U is an -pseudo orbit if d(f(xn 1), xn) < for all a < n < b. An -pseudo orbit for f is -shadowed by the orbit O(x) of x 2 U if d(xn, fn(x)) < for all a < n < b. The following is a more general version of the Shadowing Theorem for hyperbolic sets. Theorem 2.2 (Shadowing Theorem) [1] Let M be a Riemannian manifold, d the natural distance function, U M open, f : U ! M a diffeomorphism, and U a compact hyperbolic set for f. Then there exist a neighborhood U( ) and 0, 0 > 0 such that for all > 0 there is an > 0 with the following property: If f0 : U( ) ! M is a diffeomorphism 0-close to f in the C1 topology, Y a topological space, g : Y ! Y a homeomorphism, 2 C0(Y,U( )), and dC0 ( g, f0 ) := supy2Y d( g(y), f0 (y)) < then there is a 2 C0(Y,U( )) such that g = f0 and dC0 ( , ) < . Furthermore, is locally unique: If g = f0 and dC0 ( , ) < 0, then = . To get the more standard version of the Shadowing Theorem take the space Y = Z, the diffeomorphism f0 = f, the constant 0 = 0, and the function g(n) = n + 1. Additionally, replace 2 C0(Y,U( )) by {xn}n2Z U( ) and 2 C0(Y,U( )) such that g = f0 by {fn(x)}n2Z U( ). Then d(xn, fn(x)) < for all n 2 Z. A consequence of the Shadowing Theorem is the strong structural stability of hyperbolic sets [5, p. 571]. This theorem states that given a hyperbolic set M of the diffeomorphism f : U ! M and any open neighborhood V U of there exists a C1 neighborhood U of f such that each f0 2 U contains a hyperbolic set 0 V topologically conjugate to . Moreover, the conjugacy is unique when U is small enough. The Shadowing Theorem and strong structural stability of hyperbolic sets do not require local maximality of the hyperbolic set. A corollary of the strong structural stability of hyperbolic sets is that a hyperbolic set being locally maximal is a robust property and motivates whether a hyperbolic set may robustly not be contained in a locally maximal one. For a hyperbolic set of a diffeomorphism f we will denote the continuation of for a diffeomorphism f0 near f by (f0). Similarly, for a point p 2 the continuation will be denoted by p(f0) and the continuation of the stable and unstable manifolds of p, respectively, as Ws(p, f0) and Wu(p, f0). Throughout we will use the fact that locally maximal hyperbolic sets possess a local product structure. A hyperbolic set possesses a local product structure provided there exist constants , > 0 such that if x, x0 2 and d(x, x0) < , then Ws (x, f) and Wu (x, f) intersect in exactly one point which is contained in . Proposition 2.3 [5, p. 581] For a hyperbolic set locally maximality and possessing local product structure are equivalent conditions. Many of the constructions will involve hyperbolic sets with a heteroclinic tangency. A hyperbolic set for a C1 diffeomorphism has a heteroclinic tangency if there exist points x, y 2 such that Ws(x)\Wu(y) contains a point of 5 tangency. A point of quadratic tangency for a C2 diffeomorphism is defined as a point of heteroclinic tangency where the curvature of the stable and unstable manifolds differs at the point of tangency. To obtain robust properties for a manifold M, where dim(M) > 2, we will use results from normal hyperbolicity, see [6] for more details. Let f be a smooth diffeomorphism of a smooth compact manifold M leaving a C1 closed submanifold V M invariant. Definition: A diffeomorphism f is normally hyperbolic at V , if there is a continuous Df-invariant splitting of the tangent bundle TVM = Nu TV Ns such that: infp2V m(Dfp|Nu p ) > 1, supp2V kDF|Ns p k < 1, infp2V m(Dfp|Nu p )/kDfp|TV k > 1, supp2V kDfp|Ns p k/m(Dfp|TV ) < 1, where m(A) = kA 1k 1 is the conorm of an operator A. For a diffeomorphism the conorm m(Dfp) = infkvk=1,v2TpM kDf(v)k. The above implies the normal behavior to V of Df is hyperbolic and dominates the behavior tangent to V . The diffeomorphism f is normally contracting if TVM = Ns TV . The main theorem for normal hyperbolicity states that if f0 is another diffeomorphism of M and f0 is C1 near f, then f0 is normally hyperbolic at some unique submanifold V 0 which is C1 near V . 3 Hyperbolic Sets That Are Not Locally Maximal This section concerns hyperbolic sets that are not contained in any locally maximal hyperbolic sets. Specifically, we prove Theorem 1.3. The idea is to construct a hyperbolic set containing two points p and q with a quadratic tangency. We construct so that if there were a locally maximal hyperbolic set, 0 , then 0 would contain the point of tangency, a contradiction. We will use the following construction to prove Theorem 1.3. Let f : M ! M be a diffeomorphism of a smooth compact manifold M containing a normally contracting two dimensional compact submanifold U (U = M if dim(M) = 2) so that f|U satisfies: 1. an attractor a with trapping region V , 2. a fixed point q 2 a, 3. a saddle fixed point p /2 a, 4. a point z such that z 2 Wu(p) t Ws(q), 5. and there exist closed intervals J of q in Wu(q) and I of z in Wu(p) with the following properties: Each x 2 I is connected by a local stable manifold to a point y 2 J. 6 V f(V) p w t z q Figure 1: Robust Example The closed interval I contains a point w of quadratic tangency between Wu(p) and Ws(t), for some t 2 J. q p z w t J I Figure 2: Intervals I and J The above properties are illustrated in Figures 1 and 2. Proposition 3.1 For a smooth compact manifold M, where dim(M) 2, there exists a diffeomorphism f with the above properties. The proof of Proposition 3.1 is reserved for the appendix. Proposition 3.2 There exists a neighborhood U of f in Diff1(M) such that every g in U has an extension of U satisfying properties (1) (5) as above. 7 Proof. First, for U sufficiently small normal hyperbolicity implies the existence of a C1 submanifold, denoted U(g), that is C1 close to U. For U perhaps smaller the structural stability of hyperbolic sets implies there exists a continuation a(g) of the Plykin attractor, and p(g) of the hyperbolic saddle fixed point. Normal hyperbolicity implies there is a diffeomorphism h : U(g) ! U with derivative of h and h 1 uniformly bounded near the identity. Then f = h (g|U(g)) h 1 is a diffeomorphism of U which is C1 near f|U. Hence, f and g|U(g) are smoothly conjugate. Therefore, it is sufficient to show properties (1) (5) are robust under C1 perturbations for perturbations of f|U in Diff1(U). The strong structural stability of hyperbolic sets implies there exists a neighborhood V of fU in Diff1(U) such that any f 2 V contains a continuation a( f) of a and p( f) of p. We now show there exist continuations of z, w, I, and J finishing the proof of the proposition. The continuity of the stable and unstable manifolds implies for V perhaps smaller the points p( f) and q( f) will have a point z( f) 2 Wu(p, f) t Ws(q, f) a continuation of z. The existence of the continuation of w will follow since w is a point of quadratic tangency for p and t under the action of f|U. To see this we choose C2 local coordinates (x1, x2) near w such that the submanifolds Ws(t) and Wu(p) can be expressed near w by: Ws(t) = {(x1, x2) | x2 = 0}, and Wu(p) = {(x1, x2) | x2 = ax21 }, where a is a nonzero constant. Let D be a small neighborhood of t in Wu(q) such that for each y 2 D the submanifolds of Ws(y) in the local coordinate system can be expressed by: Ws(y) = {(x1, x2) | x2 = y(x1)} where y is C1 close to x2 = 0. Let f be a sufficiently small C1 perturbation of f|U. Then the continuation of the submanifolds Wu(p, f) and Ws(y, f) near w can be expressed by: Wu(p, f) = {(x1, x2) | x2 = (x1)} where is C1 close to ax21and Ws(y, f) = {(x1, x2) | x2 = y, f (x1)} where y, f is C1 near x2 = 0, and y is in the continuation of D. Then for x2 > 0 sufficiently large and some y in the continuation of D the stable manifold Ws(y, f) crosses Wu(p, f) transversally twice in the local coordinate system. Furthermore, for x2 < 0 and |x2| sufficiently large Ws(y, f) does not intersect Wu(p, f) in the local coordinate system for any y in the continuation of D. It then follows that there is some y 2 U such that Ws(y, f) and Wu(p, f) are tangent, and the tangency is near w. 8 Since the stable and unstable manifolds vary continuously in V it follows that there exist intervals I( f) and J( f) in Wu(p, f) and Wu(q, f) respectively. 2 The proof of Theorem 1.3 is a direct consequence of the following proposition. Proposition 3.3 Let g 2 U as defined in Proposition 3.2. Then the set (g) = a(g)[{p(g)}[O(z(g)) is a hyperbolic set not contained in any locally maximal hyperbolic set. Proof. From Lemma 2.1, it is clear that (g) is a hyperbolic set. Suppose for the sake of contradiction that there exists a locally maximal hyperbolic set 0 (g). We will show that w(g) 2 0, where w(g) is as in Proposition 3.1. The point w(g) is a point of tangency between Wu(p, g) and Ws(t, g), and both p(g) and t(g) are contained in the hyperbolic set (g). Hence, if w(g) 2 0 the hyperbolic splitting fails to extend continuously to w(g), a contradiction. We now prove that w(g) 2 0. Since 0 carries a local product structure, there exist constants , > 0 such that for all x, y 2 0, if d(x, y) < , then Wu (x) \Ws (y) contains a single point [x, y] 2 0. Fix such and . Let I0 I(g) be the closed subinterval whose endpoints are z(g) and w(g). We next show that I0 0. From property (3) of Proposition 3.2 and compactness of I(g) we have that there exists n > 0 such that, for all x 2 I(g), there is a point y 2 J(g) \Ws(x) such that d(fn(x), fn(y)) < /2. The set fn(I0)\ 0 is closed since I0 and 0 are closed. On the other hand, the local product structure on 0 implies that fn(I0)\ 0 is also open and nonempty in fn(I0). Specifically, if x 2 fn(I0) \ 0, then there is a y 2 fn(J(g)) 0 so that d(x, y) < /2. Note that the unstable manifold Wu(y) restricted to U(g) is contained in a(g) 0, since restricted to U(g) the set a(g) is a hyperbolic attractor under the action of g. Now, by local product structure, if y0 2 Wu(y) 0 is sufficiently close to y (within /2, say), then the point [x, y0] is contained in 0. By varying y0 in a neighborhood of y we thus obtain a neighborhood of x in fn(I0) contained in 0. Since fn(I0) \ 0 is nonempty, open and closed in fn(I0), and fn(I0) is connected, it follows from the invariance of 0 that I0 0. 2 Remark 3.4 The property that a hyperbolic set cannot be included in a closed locally maximal hyperbolic set is not necessarily stable under C1 perturbations. If the original tangency between Wu(p) and Ws(t) is topologically transverse but not of quadratic type, then under an arbitrarily small C1 perturbation of f there may be no point of heteroclinic tangency between a point of Wu(q) and p. Therefore, under the perturbed function it may be possible to include in a locally maximal hyperbolic set. For example, in the construction of f suppose M is the flat 2-torus and suppose instead there existed C3 local coordinates (x, y) near w such that locally Ws(t) and Wu(p) can be expressed by Ws(t) = {(x, y)| y = 0} Wu(p) = {(x, y)| y = ax3} 9 for some a 6= 0. Additionally, the only point of tangency between Wu(p) and stable manifolds of points inWu(q) occurs for the set O(w). Then = a[{p}[ O(z) cannot be included in a locally maximal hyperbolic set, but there exists an arbitrarily small C1 perturbation of F such that there is no heteroclinic tangency between a point of Wu(q) and p. Therefore, the new perturbed hyperbolic set can be included in a locally maximal hyperbolic set, namely the separatrix of Wu(p) containing z and a. 4 Markov Partitions for Hyperbolic Sets The proof of Theorem 1.5 relies heavily on the Shadowing Theorem and an adaptation of Bowen s construction for Markov partitions in [3]. Next, we recall a few definitions that are useful in this and the following section. We use the following results from symbolic dynamics. Specifically, we will be looking at subshifts of finite type, which were introduced by Parry [8]. Let A = [aij ] be an n n matrix with nonnegative integer entries. The graph of A is the directed graph GA with vertices V(GA) = {1, ..., n} and aij distinct edges with initial state i and terminal state j. For the graph GA with edge set E and adjacency matrix A let A be the space over E specified by A = {! = (!j)j2Z | t(!j) = i(!j+1) for all j 2 Z}, where t(!j) is the terminal state of edge !j and i(!j+1) is the initial state of edge !j+1. The map on A defined by A(!) = !0 where !0j = !j+1 is called the edge shift map. The subshift of finite type is the space ( A, A). A matrix A is irreducible if for each pair 1 i, j n there is a k 2 N such that (Ak)ij > 0. A standard result states that if A is irreducible, then A under the action of A contains a transitive point (see [9, p. 277]). Define the distance between two points !, !0 2 ( A, A) by d(!, !0) = X k2Z (!k, !0k) 4|k| , where (i, j) is the delta function. Then the local stable and unstable manifolds of a point ! are respectively, Ws loc(!) = {!0 2 ( A, A) | !0i = !i for i 0}, and Wu loc(!) = {!0 2 ( A, A) | !0i = !i for i 0}. Roughly speaking, a Markov partition of a hyperbolic set is a nice dynamical decomposition of the set and a finite to one semiconjugacy to a subshift of finite type. The partitions of a hyperbolic set are called rectangles. We now make these notions precise. For a hyperbolic set , the continuity of the stable and unstable distributions implies there exist constants > 0 such that f 1(Ws (f(p))) \ f(Wu (f 1(q))) = Ws (p) \Wu (q) 10 is a single point whenever p, q 2 satisfy d(p, q) . This point is denoted [p, q]. A set R is a rectangle provided R has diameter less than and p, q 2 R implies that Ws (p) \ Wu (q) 2 R where and are as above. A rectangle is proper if R = cl(int(R)). If R is a rectangle, then there exists a natural homeomorphism R ' Ws (x,R) Wu (x,R), where: Ws (x,R) = R \Ws (x) and Wu (x,R) = R \Wu (x). Definition: A Markov partition of is a finite collection of closed rectangles {Ri}Mi =1 that satisfy the following properties: 1. = Sm j=1 Rj , 2. Ri = cl(int(Ri)) for each i, 3. if i 6= j, then int(Rj) \ int(Ri) = ;, 4. if z 2 int(Ri) \ f 1(int(Rj )), then f(Wu(z,Ri)) Wu(f(z),Rj) and f(Ws(z,Ri)) Ws(f(z),Rj), and 5. if z 2 int(Ri) \ f 1(int(Rj )), then int(Rj) \ f(Wu(z,Ri) \ int(Ri)) = Wu(f(z),Rj) \ int(Rj) and int(Ri) \ f 1(Ws(z,Rj) \ int(Rj)) = Ws(z,Ri) \ int(Ri). The definition of a Markov partition implies that if x, y 2 Ri, then Ws (x) \ Wu (y) and Wu (x) \ Ws (y) are contained in Ri. Note this is different than a local product structure for . Specifically, if x, y 2 , the points x and y are in different rectangles, and d(x, y) < , then Ws (x) \Wu (y) or Wu (x) \Ws (y) may not be contained in . Proof of Theorem 1.5. Let U be a neighborhood of satisfying the hypothesis of the Shadowing Theorem. For U perhaps smaller U V and U = T n2Z fn(U) is hyperbolic. Let d( , ) be an adapted metric on U and extend the metric continuously to a neighborhood U0 of U. Fix > 0 and such that for any two points x, y 2 U if d(x, y) < then f 1(Ws (f(x))) \ f(Wu (f 1(y))) = Ws (x) \Wu (y) consists of one point, and the set S x2 B2 (x) is contained in U \ U0. Fix 0 < /2 as in the conclusion of the Shadowing Theorem so that every -orbit is /2-shadowed. 11 Let < /2 such that d(f(x), f(y)) < /2 and d(f 1(x), f 1(y)) < /2 when d(x, y) < for any x, y 2 U. Take a -dense set {pi}Ni =1 in , and let A be the transition matrix with aij = 1 if d(f(pi), pj) < 0 if d(f(pi), pj) Let ( A, A) be the subshift of finite type determined by A. We see by the way we have chosen , that every ! corresponds to a unique -orbit. For ! 2 A let (!) = p!0 . For ! and !0 close, then !0 = !00 so (!) = (!0) and is continuous. The sequence ( nA (!)) = p!n is an -pseudo orbit. We now verify that since d(f(p!n), p!n+1) < we have a well defined map from A into U. The Shadowing Theorem shows the -orbits { ( nA (!))}n2Z = {p!n}n2Z are uniquely -shadowed by (!) 2 U. Additionally, the Shadowing Theorem implies that A = f. Therefore, fn((!)) 2 U for all n 2 Z and it follows that (!) 2 U. Corresponding to the product structure [ , ] on U as defined in the preliminary section on symbolic dynamics, there is a product structure on A denoted also by [ , ] and defined by: [!, !0] = !i for i 0 !0i for i 0 for any !, !0 2 A with !0 = !00. Our choice of constants , , , and along with the continuity of implies that, [ , ] commutes with , so ([!, !0]) = [(!), (!0)]. Let Ri = {(!) | !0 = i} be the image of the i cylinder set in A. If x = (!) and y = (!0) are both in Ri, then !0 = !00 = i, so that [x, y] = [(!), (!0)] = ([!, !0]) 2 Ri. Hence, Ri is a rectangle. The continuity of implies that each Ri is closed. Let = SN i=1 Ri = ( A). Since A = f A the set is invariant. The set is also hyperbolic since U. Finally, we refine the rectangles R1, ...,RN to construct a Markov partition for . Define Ws(x,Rj) = Ws (x) \Rj and Wu(x,Rj) = Wu (x) \Rj . The sets Ri are each rectangles and by definition the set {Ri} covers . As a reminder the local stable manifold and unstable manifold of a point ! 2 A is defined in the preliminary section on symbolic dynamics. Claim 4.1 For any ! 2 A the following hold: (Ws loc(!)) Ws ((!),R!0 ), and (Wu loc(!)) Wu ((!),R!0 ). Proof. Let !0 2 Ws loc(!). Then !0i = !i for i 0. Since !0 = !00 we have that (!0) 2 R!0 . Also, we have the d(fn((!0)), fn((!))) for all n 0, 12 so that (!0) 2 Ws ((!)). Hence, (Ws loc(!)) Ws ((!)) \ R!0 = Ws ((!),R!0 ). A similar argument shows (Wu loc(!)) Wu ((!),R!0 ). 2 Claim 4.2 For any x 2 the following hold: f(Ws(x,R!0 )) Ws(f(x),R!1 ) and f(Wu(x,R!1 )) Wu(f(x),R!0 ). where x = (!) for some ! 2 ( A, A). Proof. Let (!) = x 2 Ri, and let y 2 Ws(x,Ri). This implies that y = (!0) for some !0 2 ( A, A) such that w0 = w00 = i. Let ! = [!, !0] = Ws loc(!) \ Wu loc(!0). Hence, (! ) = [(!), (!0)] = [x, y] = y, and nA (! )i = !i. Since d is an adapted metric and y 2 Ws (x) we have that d(f(x), f(y)) < . This implies that ( A(! )) = f(y) 2 Ws(f(x),R!1 ) so f(Ws(x,R!0 )) Ws(f(x),R!1 ). A similar argument shows that f(Wu(x,R!1 )) Wu(f(x),R!0 ). 2 We now modify the rectangles Ri to obtain proper rectangles satisfying the Markov conditions. Subdivide each rectangle Rj for each Rk that intersects Rj as follows: R1 j,k = Rj \ Rk, R2 j,k = {x 2 Rj |Wu(x,Rj) \ Rk 6= ; and Ws(x,Rj) \ Rk = ;}, R3 j,k = {x 2 Rj |Wu(x,Rj) \ Rk = ; and Ws(x,Rj) \ Rk 6= ;}, and R4 j,k = {x 2 Rj |Wu(x,Rj) \ Rk = ; and Ws(x,Rj) \ Rk = ;}. The boundary of each rectangle can be written as @(Ri) = @s(Ri) [ @u(Ri), where @s(Ri) = {x 2 Ri | x /2 int(Wu(x,Ri))} and @u(Ri) = {x 2 Ri | x /2 int(Ws(x,Ri))}. Define X = {x 2 |Ws (x)\@s(Ri) = ; andWu (x)\@u(Ri) = ;}. It follows that X is open and dense in . For x 2 X define R(x) = \ {int(Rq j,k) | x 2 Rq j,k and Rj \ Rk 6= ;}. It then follows that for x, y 2 X either R(x) = R(y) or R(x) \ R(y) = ;. The following claims now follow from Bowen s construction, see [9, p. 432] for proofs. Since there are only a finite number of Rq j,k each R(x) is nonempty and open in . Since each Rq j,k is a rectangle, it follows that cl(R(x)) is a rectangle. By construction, each cl(R(x)) is proper. The collection {cl(R(x)) | x 2 X} forms a new partition {R0i } of . Since there are only a finite number of Rq j,k there are only a finite number of sets cl(R(x)). We now show these from a Markov partition of . 13 The same argument as in Claims 4.1 and 4.2 shows that for the subshift of finite type ( A0 , A0 ) defined by {R0i } and ! 2 ( A0 , A0 ) the following hold: f(Ws(x,R!0 )) Ws(f(x),R!1 ) and f(Wu(x,R!1 )) Wu(f(x),R!0 ). Lastly, it follows from an argument as in Claims 4.1 and 4.2 that if x 2 int(R0i )\ f 1(int(R0j )), then Ws(x,R0i ) R0i \ f 1(Ws(f(x),R0j )) R0i \ f 1(Ws (f(x)) = Ws(x,R0i). The above is a stronger version of property (5) for a Markov partition. Hence, has a Markov partition. 2 4.1 Condition to guarantee local maximality In this subsection a condition is given guaranteeing a hyperbolic set is contained in a locally maximal hyperbolic set. Corollary 4.3 If is a hyperbolic set and in the Shadowing Theorem it is possible for sufficiently small to chose = such that the set T n2Z fn( S x2 B (x)) is hyperbolic, then is contained in a locally maximal hyperbolic set. Proof. Take {pi}Ni =1 a sufficiently dense set as in the construction used to prove Theorem 1.5. Let be the hyperbolic set with a Markov partition constructed as in the proof of Theorem 1.5. Each x 2 is a -shadowing of an -pseudo orbit given by elements of {pi}Ni =1. Hence, we have x 2 \ n2Z fn( N[ i=1 B (pi)) = \ n2Z fn( N[ i=1 B (pi)), since = . This implies that 2 \ n2Z fn( N[ i=1 B (pi)). Next, fix x 2 \ n2Z fn( N[ i=1 B (pi)), then x has an -pseudo orbit given by the pi s so x is in the set . Therefore, = \ n2Z fn( N[ i=1 B (pi)) so is locally maximal. 2 14 5 Further Results of Hyperbolic Sets That are Not Locally Maximal In this section we prove Theorems 1.6 and 1.7. In each proof the idea is to extend the diffeomorphism f0 from Section 3 to higher dimension. The proof of Theorem 1.6 will use consequences from normal hyperbolicity and Theorem 1.5. 5.1 Proof of Theorem 1.6 Let M be a smooth compact surface, and let T2 denote the flat 2-torus. Fix F0 2 Diff1(M T2) of the form F0(x, y) = (g(x), f(y)) and satisfying the following properties: (see Figure 3) g has a hyperbolic saddle fixed point, a, with a transverse homoclinic point b. The submanifold {a} T2 is normally hyperbolic under F0. f contains a hyperbolic set 0, a hyperbolic attractor a with fixed point q, a saddle fixed point p /2 a, and a point z 2 Ws(q) t Wu(p). Furthermore, f satisfies the conclusions of Theorem 1.3 as in Section 3. M T 2 a b q p Figure 3: Definition of F0 the stable and unstable manifolds are two dimensional and have components inside and outside of the fiber {a} T2. Figure 3 depicts their components transverse to the fiber. To construct F satisfying the conclusions of Theorem 1.6 we will deform F0, such that Ws(a, p) and Wu(a, q) intersect transversally in the fiber {b} T2. Specifically, the deformation will be supported in a small neighborhood of g(b). Near g(b) the diffeomorphisms F and F0 will differ by a translation in the fiber direction. 15 Denote the orbit of a point x 2 M under g by Og(x). Choose a small neighborhood U of g(b) such that Og(b) g(b) M U. Fix U , a small neighborhood of g(b) contained in U, and : T2 ! T2 a translation sending q to p . Let G be a diffeomorphism of M T2 given by G = (Id, ). By possibly perturbing f near p we obtain that Wu(a, q) and G(Ws(a, p)) are transverse at (b, q). Let F 2 Diff1(M T2) be any diffeomorphism satisfying: F(x, y) = F0(x, y) for all x 2 M U and F(x, y) = (g, f) for all x 2 U , see Figure 4. By construction of F the fixed points (a, p) and (a, q) are heteroclinically related with (a, z) 2 Wu((a, p), F) t Ws((a, q), F) and (b, q) 2 Ws((a, p), F) t Wu((a, q), F). By Lemma 2.1 the set 1 = (a, p) [ (a, a) [ O(a, z) [ O(b, q) is a hyperbolic set for F. M T 2 a U b p q q Figure 4: Definition of F Lemma 5.1 The hyperbolic set 1 is included in a transitive hyperbolic set . 16 Proof. It suffices to show that 1 is included in a hyperbolic set with a Markov partition as constructed in Theorem 1.5 such that the matrix A is irreducible, so ( A, A) is transitive. The beginning of the proof of Theorem 1.5 gives the following claim: Claim 5.2 If is a hyperbolic set, then there exists a neighborhood U of and positive constants , , and such that corresponding to any dense set of points {pi} , there is a shift space ( A, A) and maps and satisfying the following: 1. for any ! 2 A the set { ( nA (!))} is an -pseudo orbit. 2. the set ( A) is a hyperbolic set with a Markov partition, 3. the set ( A) contains , and 4. any x 2 ( A) is a -shadowing of an -pseudo orbit given by some !. We now fix a set {pi}ni =1 that is dense such that there exist constants n1 < n2 satisfying the following: p1, ..., pn1 2 {a} a, pn1+1, ..., pn2 2 O(a, z), pn2+1 = (a, p), and pn2+2, ..., pn 2 O(b, q). Since a is mixing it follows from the above construction that for any pi and pj in a it is possible to construct an -pseudo orbit from pi to pj . If pi 2 O(b, q) and pj 2 a it is possible to construct an -pseudo orbit from pi to pj as follows: First, follow pi to pn. Since the set {pi} is dense there is a point pk 2 a such that d(f(pn), pk) < . Then, there is an -pseudo orbit from pk to pj . Similarly, for pi 2 a and pj 2 O(a, z) we can construct an -pseudo orbit from pi to pj . Lastly, for pi 2 O(a, z) and pj 2 O(b, q) it is possible to construct an -pseudo orbit by simply following pi to pj . Therefore, the transition matrix A is irreducible. For The map as given by the above claim the following diagram commutes: A A / A F / We now find a transitive point in . It will follow from surjectivity and continuity of . We will show that there exists a y 2 such that cl(O+(y)) = cl(O (y)) = . The Shadowing Theorem shows that for any ! 2 A, the set 17 (O+ A(!)) = O+ F ((!)) and the set (O+ A(!)) O+ F ((!)). Let y = (!0) where !0 is transitive in A under A. Then, = ( A) = (O+ A(!0)) O+ F ((!0)) = O+ F (y) . Hence, O+ F (y) = and similarly O F (y) = . Therefore, y is transitive in . 2 Proof of Theorem 1.6. Using normal hyperbolicity we show the above construction is robust. Structural stability of hyperbolic sets implies there exists a neighborhood U of F in Diff1(M) such that for any G 2 U there exists a hyperbolic set (G) conjugate to . Furthermore, for U perhaps smaller the conjugacy is unique and near the identity. Normal hyperbolicity implies that for U perhaps smaller there is a unique invariant manifold V (G) which is C1 near V and a diffeomorphism h : V (G) ! V such that the derivative of h and of h 1 is uniformly bounded and near the identity. Then f = h (G|V (G)) h 1 is a diffeomorphism of V which is C1 near F|V . Let V be a neighborhood of F|V satisfying Theorem 1.3. For U perhaps smaller f will be in V. Hence, G|V (G) is smoothly conjugate to f and contains a hyperbolic set as constructed in the proof of Theorem 1.3. Suppose is contained in a locally maximal hyperbolic set . Then there exists a neighborhood U of such that T n2Z Gn(U) = , and V (G) \ U is a neighborhood of \V (G) in V (G). By the invariance of the normally hyperbolic fiber it follows that T n2Z Gn(U \V (G)) = \V (G). Hence, under the action of G|V (G) the hyperbolic set \V (G) is contained in a locally maximal hyperbolic set, a contradiction. 2 5.2 Proof of Theorem 1.7 In this subsection we show we can strengthen the previous result to a symplectic diffeomorphism, but with the loss of compactness for M and robustness of the construction. The proof of Theorem 1.7 uses classical techniques from Hamiltonian dynamics. The idea is to embed a diffeomorphism as a subsystem of a twist map. As the proof uses classical results for symplectic diffeomorphisms we review some basic concepts, see [7, p. 97-98] for more details. Equip R2n with the standard symplectic form, ! = dx^dy = Pn i=1 dxi ^dyi. Let F be a symplectic diffeomorphism of R2n and denote F(x, y) = (X, Y ). Since F is symplectic and R2n is simply connected the closed 1-form = Y dX ydx is exact. Hence, there is a function L(x,X) such that dL = Y dX ydx. Since dL = @L @X dX + @L @x dx it follows that Y = @L @X and y = @L @x . (5.1) The symplectic diffeomorphism F is called a twist map if equation (5.1) implicitly defines a unique solution for all (x,X). For F symplectic the Hessian of L must be non-singular. Hence, X and Y can be solved in terms of x and y. The following technical lemma is the heart of proving Theorem 1.7. The lemma extends a diffeomorphism f of Rn to a diffeomorphism F of R2n so that: 18 F is volume preserving, the set Rn {0} is invariant, and a hyperbolic set of f is a hyperbolic set {0} of F. Lemma 5.3 Let f be a diffeomorphism of Rn that equals the identity outside of a ball of finite radius containing a hyperbolic set . Then F as defined above is a symplectic diffeomorphism of R2n such that F(x, 0) = (f(x), 0) and the set {0} is a hyperbolic set for F. Proof. Define the function L : R2n ! R as L(x,X) = 1 2kX f(x)k2. Furthermore, define Y = @L @X = (X f(x))T and y = @L @x = (X f(x))TDfx. where x and X are thought of as column vectors and y and Y are thought of as row vectors. Then y Df 1 x = XT f(x)T . Solving the above for X and Y yields X = f(x) + (y Df 1 x )T and Y = y Df 1 x . (5.2) Define F(x, y) = (X(x, y), Y (x, y)), then equation (5.2) implies that F is a symplectic diffeomorphism preserving the standard volume form. Furthermore, F(x, 0) = (f(x), 0) and DF(x, 0) = Dfx 0 Df 1 x . Hence, {0} is a hyperbolic invariant set for F. 2 Proof of Theorem 1.7. First let f0 be a diffeomorphism of the two dimensional disk D such that f0 contains a hyperbolic set as in Section 3. Let f be a diffeomorphism of R2 such that f|D = f0 and for some R > 0 sufficiently large f|R2 B(R,0) = Id. Lemma 5.3 then implies that f can be extended to a symplectic diffeomorphism F, such that {0} is a hyperbolic set for F. Furthermore, we have F|R2 {0} = f. It follows from the same argument used in Proposition 3.3 that {0} cannot be contained in a locally maximal hyperbolic set. 2 Remark 5.4 This example may not be robust, since the submanifold is not normally hyperbolic. A Proof of Proposition 3.1 In this appendix we construct a diffeomorphism satisfying the conclusions of Proposition 3.1. Proof. We start with a Plykin attractor a for a diffeomorphism f of S2 as constructed in [5, p. 540]. The diffeomorphism contains a hyperbolic repelling 19 fixed point p0. The set Ws( a) consists of S2 minus four hyperbolic repelling periodic points. Fix a periodic point q contained in a. Since a is a locally maximal topologically transitive hyperbolic set with periodic points dense and the closure of a heteroclinic class we have that S Ws(O(q)) = Ws( a). Hence, there exist points q0 2 O(q) and z 2 Ws(q0) such that p0 2 (z, f). Now puncture the sphere at p0 and replace p0 with a closed circle obtaining a closed disk D0. The induced homeomorphism from f on D is not a diffeomorphism, but can be deformed near @D to obtain a diffeomorphism f of D0 such that the following hold: 1. f|@D = Id, 2. Ws( a) = D\@D, and 3. there exists a periodic point q0 2 a, a fixed point p 2 @D0, and a point z 2 Ws(q0) such that p 2 (z, f). By construction, the derivative of f at p in the direction tangent to @D0 is 1. Also, the derivative of p in the direction tangent to Ws(y) is greater then or equal to 1. Hence, we can perturb f in a neighborhood of p, such that p is a hyperbolic saddle fixed point with Es tangent to @D0 and Eu perpendicular to @D. Fix n 2 N such that q0 is fixed under fn. Next embed D0 as a disk in the interior of a closed disk D and extend fn to a diffeomorphism f0 of D such that, within a neighborhood of the boundary of D, the diffeomorphism f0 is the identity. By construction, z 2 Wu(p) \ Ws(q0). Notice that a perturbation near the heteroclinic point f 1 0 (z) does not change Ws(q0) near z. Thus, by possibly perturbing f0 in a sufficiently small neighborhood of f 1 0 (z), we may determine that z 2 Wu(p) t Ws(q0). The transversality of Wu(p) and Ws(q0) at z now imply the existence of a neighborhood J0 of q0 in Wu loc(q0) and a neighborhood I0 of z in Wu(p) such that each x 2 I0 is connected by a local stable manifold to a point y 2 J0. Let z0 2 I0 z. Then by deforming f0 in a sufficiently small neighborhood of f 1 0 (z0), we create a point w 2 I0 of quadratic tangency between Wu(p) and Ws(t), for some t 2 J0. Let I be the segment of Wu(p) from z to w, and let J be the segment of Wu(q0) from q0 to t. It is now simple to construct a diffeomorphism f satisfying the conclusions of Proposition 3.1, when M is a compact boundaryless surface. We just take a sufficiently small coordinate chart ' : D ! M such that the diffeomorphism ' f0 ' 1 satisfies the conclusions of Proposition 3.1, and extend ' f0 ' 1 trivially to a diffeomorphism f of M. Suppose that dim(M) = n 3. Let f1 : S2 ! S2 be a diffeomorphism satisfying the conclusions of Proposition 3.1. We extend f1 to a diffeomorphism f2 of an n-dimensional open disk, Dn, such that the following hold: 1. in a neighborhood of the boundary @Dn the diffeomorphism f2 is the identity, 2. S2 is a normally contracting invariant submanifold under f2, and 20 3. f2|S2 = f1. Now take a sufficiently small coordinate chart, ' : Dn ! M such that ' f1 ' 1 satisfies the conclusions of Proposition 3.1. Lastly, we extend ' f1 ' 1 trivially to all of M completing the proof. 2 References [1] V. M. Alekseyev, A. Katok, and A. G. Kusnirenko. Three papers in dynamical systems. In Trans. of the Amer. Math. Sco., volume 116 of 2. American Mathematical Society, Providence, RI, 1981. [2] D. V. Anosov. Geodesic flows on Riemannian manifolds with negative curvature. In Proc. Steklov Inst. Math., volume 90. American Mathematical Society, Providence, RI, 1967. [3] R. Bowen. Equilibrium States and the Ergodic Theory of Anosov Diffeomorphisms, volume 470 of Lecture Notes in Mathematics. Springer- Verlag, New York, Heidelberg, Berlin, 1975. [4] S. Crovisier. Une remarque sur les ensembles hyperboliques localement maximaux.Copyright R. Math. Acad. Sci. Paris, 334(5):401 404, 2001. [5] B. Hasselblatt and A. Katok. Introduction to the Modern Theory of Dynamical Systems. Cambridge University Press, 1995. [6] M. Hirsch,Copyright Pugh, and M. Shub. Invariant Manifolds, volume 583 of Lecture Notes in Mathematics. Springer-Verlag, Berlin-New York, 1977. [7] K.R. Meyer and G.R. Hall. Introduction to Hamiltonian Dynamical Systems and the N-body Problem. Springer-Verlag, New York, 1992. [8] W. Parry. Intrinsic Markov chains. Trans. Amer. Math. Soc., 112:55 56, 1964. [9] C. Robinson. Dynmical Systems Stability, Symbolic Dynamics, and Chaos. CRC Press, 1999. [10] Ya. G. Sinai. Construction of Markov partitions. Funct. Anal. Its Appl., 2:245 253, 1968. [11] Ya. G. Sinai. Markov partitions and c-diffeomorphisms. Funct. Anal. Its Appl., 2:61 82, 1968. [12] S. Smale. Differentiable dynamical systems. Bull. Amer. Math. Soc., 73:747 817, 1967. 21