arXiv:1709.00638v6 [math.DS] 31 Mar 2021 C ytm( system di Anosov of notion the ytm,nnatnmu yaia systems. dynamical non-autonomous systems, xs nsvfmle ( families Anosov exist di Anosov an to behavior ihthe with 1,Eape3.Frhroe the Furthermore, 3). Example [1], smti,tu,tehproiiycudb nue yteR the by induced be detail). could more hyperbolicity the thus, isometric, uhta,i ( if that, such ustof subset usae r one wyfo seDfiiin28.Young 2.8). Definition of (see consisting families 0 anthat from the away that bounded such are and subspaces derivative second bounded with families let hsetoyi nain yuiomcnuay .Kwnand of Kawan entropy C. the conjugacy. for uniform las by invariant is non-stationar for entropy entropy This topological of notion a troduced systems dynamical hscnuayi locniee ocharacterize to considered also is conjugacy This h oino nfr ojgc sas nw as known also is conjugacy uniform of notion The ilpoethat prove will hsfc en that means fact This n ed o s o n diinlcniin osdrn t Considering condition. additional any for ask sin not isolated), do not we are examples and these way, certain [1] in a found in be that, can varie families great Anosov the of examples is (non-trivial result this of implication important most TUTRLSAIIYADACAATRZTO FAOO FAMI ANOSOV OF CHARACTERIZATION A AND STABILITY STRUCTURAL 2 e od n phrases. and words Key 2010 D r nsv n[4 epoefray( any for prove we [14] In Anosov. are nsvfmle eeitoue yP roxadA ihrin Fisher A. and Arnoux P. by introduced were families Anosov Let o-ttoaydnmclssesaecasfidby classified are systems dynamical Non-stationary m ( ahmtc ujc Classification. Subject Mathematics M rva xmlso nsvfmle ilb ie nti pap characterizati a this and in stability given structrural be the will manifolds, families Anosov of examples trivial A M f togtopology strong bstract i D ) etestcnitn ftefmle of families the of consisting set the be ) i etedson no ftefml fReana manifolds Riemannian of family the of union disjoint the be ∈ m Z g ( endo euneo opc imninmnfls( manifolds Riemannian compact of sequence a on defined M i ) A nsvfmle r o-ttoaydnmclsseswit systems dynamical non-stationary are families Anosov . i ∈ ossigo nsvfmle n by and families Anosov of consisting ) Z b 2 ( D ∈ M or A sa pnsbe of subset open an is ) o-ttoaysbhfso nt type finite of subshifts non-stationary iedpnin yaia systems dynamical time-dependient 1 1 nsvfmle,Aoo di Anosov families, Anosov ff ( f ( M i M ff oopim.Ruhy nAoo aiyi non-stationary a is family Anosov an Roughly, eomorphisms. seScin2 rwt the with or 2) Section (see ) i C ∈ oopim seDfiiin26.I sipratt on ou point to important is It 2.6). Definition (see eomorphisms ,sc that such ), Z sa pnsbe of subset open an is ) 1 EVNYD EU UNE ACEVEDO MUENTES JESUS DE JEOVANNY + uhta the that such 1 admsalprubtoso nAoo di Anosov an of perturbations small random M i s lhuhte r di are they although ’s, 72;3C5 37B55. 37C75; 37D20; g i .I 1. is f i ) ε D f i i ntroduction ∈ i saentncsaiyAoo di Anosov necessarily not are ’s coeto -close Z 2 ( hr xssasqec fpstv ubr ( numbers positive of sequence a exists there , ff M 1 oopim tutrlsaiiy o-ttoaydynami non-stationary stability, structural eomorphism, admdnmclsystems dynamical random D ,wihgnrlzsteYugsresult. Young’s the generalizes which ), C equi-conjugacy 1 nfracrancaso nsvfamilies. Anosov of class certain a for on m nfr topology uniform ( -di M f i nthe in noe ihtesrn oooy The topology. strong the with endowed ) ff A yo o-rva xmlsta tprovides it that examples non-trivial of ty ew nyakta h aiyb Anosov be family the that ask only we ce nfr conjugacy uniform yaia ytm se[] 7,[13]). [7], [6], (see systems dynamical y oopim on eomorphisms r eso h xsec finvariant of existence the show We er. se[3,[].KlaaadSoain- Snoha and Kolyada [8]). [13], (see b m nrdcdb ihradAnu in Arnoux and Fisher by introduced , ( n 1] hstersl n[4 proves [14] in result the thus [12], and ff M lsbtenteusal n stable and unstable the between gles eana erc se[] 1]for [12] [1], (see metrics iemannian euiomtplg on topology uniform he oopi,te r o necessarily not are they eomorphic, C h e ossigof consisting set the ) .Ltskn n[] aeformu- gave [4], in Latushkin, Y. 1 n[8,Pooiin22 proved 2.2, Proposition [18], in , tplg,te ( then -topology, yeblcbhvo.Non- behavior. hyperbolic h nteltrtr se .. [6]). e.g., (see, literature the in 1,mtvtdb generalizing by motivated [1], ednt by denote We . M M i ) i i for , ∈ M ff Z and oopimo class of eomorphism seDfiiin2.4). Definition (see hc a similar a has which , ff hc sendowed is which , oopim (see eomorphisms i non-autonomous ∈ g i ) Z i ∈ For . D Z A C htthere that t dynamical A ∈ m m m ( ( Anosov M M m LIES 1 ε ,we ), ( the ) ≥ M i ) i cal ∈ 1, ). Z 2 JEOVANNYDEJESUSMUENTESACEVEDO

[1], which code non-stationary dynamical systems, and in particular Anosov families. In [13] we 1 proved that for any sequence of C -diffeomorphisms ( fi)i Z, there exists a sequence of positive 1 ∈ 1 numbers (δi)i Z such that if gi is a C -diffeomorphism which is δi-close to fi in the C -topology, ∈ then ( fi)i Z and (gi)i Z have the same topological entropy. ∈ ∈ If E is a vector bundle over a compact Riemannian manifold M, set Γ(E) the Banach R-vector space of continuous sections of E over M, endowed with the C0-topology. An automorphism L : E E is called hyperbolic if σ(L) S1 = , where σ(L) is the spectrum of L. For p M, let → ∩ ∅ ∈ Ep denote the fiber of E over p. For a diffeomorphism f on M, let f : Γ(TM) Γ(TM) be the 1 ∗ → bounded linear operator given by f (ζ) = Df ζ f − , where TM is the tangent bundle over M. J. Mather in [10], [11] proved that f∗ is an Anosov◦ ◦ diffeomorphism if and only if f is a hyperbolic automorphism. We can define the operator f for non-stationary dynamical systems.∗ In this case, that operator could be unbounded or non-hyperbolic∗ (see Section 4). We will give some conditions on an Anosov family to obtain the hyperbolicity of f defined for the family (see Theorem 4.5). ∗ Structural stability for non-stationary dynamical systems with respect to the uniform topology on m( ) will be defined in Section 2 (see Definition 2.5). A. Castro, F. Rodrigues and P. Varanda, in [2],D TheoremM 2.3, proved the stability of sequences of Anosov diffeomorphisms which are small perturbations of a fixed Anosov diffeomorphism. Liu in [8], Theorem 1.1, proved this same fact 2 1 for the random case. In this work we will show that b( ) is structurally stable in ( ), which A2 M D M generalizes the above results, since the elements in b( ) are not necessarily small perturbations of a fixed diffeomorphism. A M In the next section we will define the class of objects to be studied in this work: Anosov fami- lies. Furthermore, we will introduce the strong and uniform topologies on m( ) and the uniform conjugacy to classify non-stationary dynamical systems. In Section 3 weD willM show some results that provide a great variety of examples of Anosov families. Another examples and properties of Anosov families can be found in [1], [12] and [13]. A characterization of Anosov families will be given in Section 4, which generalizes the characterization of J. Mather in [10] and [11] for Anosov ff 2 2 di eomorphisms. In Section 5 we will prove the openness of b( ) in ( ) with respect to the A M2 D M uniform topology. We will see in Section 6 that each family in b( ) admits stable and unstable at every point of . Finally, the structural stability of 2( ) willA beM proved in Section 7. M Ab M 2. Anosov families

In this work we will consider a sequence of Riemannian manifolds Mi with fixed Riemannian metric , for i Z. Consider the disjoint union h· ·ii ∈ = M = M i. M i i × ai Z [i Z ∈ ∈ will be endowed with the Riemannian metric , induced by , , setting M h· ·i h· ·ii (2.1) , = , for i Z. h· ·i|Mi h· ·ii ∈ We denote by i the induced norm by , i on TMi and we will take defined on as = k·kZ h· ·i k·k M Mi i for i . If di( , ) is the metric on Mi induced by , i, then is endowed with the metrick·k| k·k ∈ · · h· ·i M min 1, d (x, y) if x, y M d(x, y) = { i } ∈ i 1 if x Mi, y M j and i , j.  ∈ ∈  STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 3

Definition 2.1. A non-stationary dynamical system ( , , , ) (or n.s.d.s, for short) is a map M h· ·i F : , such that, for each i Z, = f : M M + is a diffeomorphism. Sometimes F M→M ∈ F|Mi i i → i 1 we use the notation = ( fi)i Z. The composition law is defined to be F ∈ fi+n 1 fi : Mi Mi+n if n > 0 − ◦···◦ → n = 1 1 i :  fi−+n fi−1 : Mi Mi+n if n < 0 F  ◦···◦ − → Ii : Mi Mi if n = 0, →  where Ii is the identity on Mi (see Figure 2.1).

fi 1 fi ... − ... −−−→ −−−→

Mi 1 Mi Mi+1 − Figure 2.1. A non-stationary dynamical system on a sequence of 2-torus with dif- ferent Riemannian metrics.

We use the notation ( , , , ), to indicate that we are fixing the Riemannian metric given in (2.1). M h· ·i F The study of time-depending dynamical systems is known in the literature with several differ- ent names: non-stationary dynamical systems, non-autonomous dynamical systems, sequences of mappings, among other names (see [1], [2], [3], [4], [13], [14] and the references there in).

Let X1 and X2 be n-dimensional compact Riemannian manifolds. We will introduce the metric on the spaces Diffm(X , X ) = φ : X X : φ is a Cm-diffeomorphism for m = 0, 1, 2. 1 2 { 1 → 2 } p Let k be the Riemannian norm on Xk and hk : Xk R be a C∞ isometric embedding, for k = k·k1, 2 and p N large enough. We choose a system→ of exponential charts (U , exp ) l1 and ∈ { xi xi }i=1 (V , exp ) l2 which covers X and X respectively, where U is an open subset of T X and V { yi yi }i=1 1 2 xi xi 1 yi is an open subset of Tyi X2 with diameter less than ̺/2, where ̺ is an injectivity radius of X2. We Rn can identify isometrically Uxi and Vyi with open subsets of and, by an abuse on the notation, we will continue calling them by Uxi and Vyi . Let dk( , ) be the metric induced by k on Xk, for k = 1, 2. Consider two homeomorphisms φ : X X· ·and ψ : X X . The d0 metrick·k on Diff0(X , X ) is given by 1 → 2 1 → 2 1 2 0 1 1 (2.2) d (φ, ψ) = max max d2(φ(x), ψ(x)), max d1(φ− (y), ψ− (y)) . x X1 y X2 ( ∈ ∈ ) If ϕ Diff0(X , X ), take ∈ 1 2 1 ϕ˜ = h ϕ exp for i = 1,..., l andϕ ˆ = h ϕ− exp for i = 1,..., l . xi 2 ◦ ◦ xi 1 y 1 ◦ ◦ yi 2 1 1 The d metric on Diff (X1, X2) is given by 1 0 (2.3) d (φ, ψ) = d (φ, ψ) + max sup Du(φ˜ x ) Du(ψ˜ x ) + max sup Dv(φˆy ) Dv(ψˆ y ) , 1 i l i i 1 i l i i 1 u Ux k − k 2 v Vy k − k ≤ ≤ ∈ i ≤ ≤ ∈ i 4 JEOVANNYDEJESUSMUENTESACEVEDO

n 2 where Dwϕ is the derivative of any diffeomorphism ϕ at w R . If φ and ψ are C -diffeomorphisms, the d2 metric is given by ∈ 2 = 1 + 2 ˜ 2 ˜ + 2 ˆ 2 ˆ (2.4) d (φ, ψ) d (φ, ψ) max sup Du(φxi ) Du(ψxi ) max sup Dv (φyi ) Dv(ψyi ) , 1 i l 1 i l 1 u Ux k − k 2 v Vy k − k ≤ ≤ ∈ i ≤ ≤ ∈ i where D2 ϕ is the second derivative of any C2-map ϕ at w Rn. w ∈ Fix m 1. The set ≥ m m ( ) = = ( fi)i Z : fi : Mi Mi+1 is a C -diffeomorphism D M {F ∈ → } can be endowed with the strong topology (or Whitney topology) or the uniform topology: m Definition 2.2. Given = ( fi)i Z and = (gi)i Z in ( ), take F ∈ G ∈ D M m = m duni f ( , ) sup min d ( fi, gi), 1) , F G i Z { { }} ∈ m m m where d ( , ) isthe C -metric on Diff (Mi, Mi+1) (see (2.2), (2.3) and (2.4)). The uniform topology on m( ·) is· induced by dm ( , ). We denote by τ the uniform topology on m( ). D M uni f · · uni f D M m Definition 2.3. For each = ( fi)i Z ( ) and a sequence of positive numbers (ǫi)i Z, a strong basic neighborhood of Fis the set∈ ∈D M ∈ F m m m B ( , (ǫi)i Z) = ( ): d ( fi, gi) <ǫi for all i . F ∈ {G∈D M } The Cm-strong topology is generated by the strong basic neighborhoods of each m( ). F ∈D M A sequence (hi)i Z, where hi : Mi Mi is a continuous maps for any i Z, is equicontin- uous if for any ε >∈0 there exists δ >→0 such that for any i Z, if x, y M∈ with d (x, y) < δ ∈ ∈ i i then di(hi(x), hi(y)) < ǫ. Non-stationary dynamical systems are classified via uniform topological conjugacy: Definition 2.4. A uniform topological conjugacy between two non-stationary systems ( , , , ) Z = M h· ·i F and ( , , , ) is a map : , such that, for each i , Mi hi : Mi Mi is a M h· ·i G H 1M→M ∈ H| → homeomorphism, (hi)i Z and (hi− )i Z are equicontinuous and ∈ ∈ h + f = g h : M M + , i 1 ◦ i i ◦ i i → i 1 that is, the following diagram commutes:

f 1 f0 f1 − M 1 M0 M1 M2 − −−−−−→ −−−−−→ −−−−−→

h 1 h0 h1 h2 ··· − ···  g 1  g0  g1   −    M 1 M0 M1 M2 y− −−−−−→ y −−−−−→ y −−−−−→ y In that case, we will say the families are uniformly conjugate. The reason for considering uniform conjugacy instead of topological conjugacy is that every non-stationary dynamical system is topologically conjugate to the constant family whose maps are all the identity (see [1], Proposition 2.1). The structural stability for elements in m( ) will be given considering the uniform topology on m( ) (see Definition 2.2). D M D M Definition 2.5. We say that m( ) is uniformly structurally stable if there exists ε > 0 such m Fm ∈D M that any ( ), with duni f ( , ) < ε, is uniformly conjugate to . A subset is called uniformlyG∈D structurallyM stable if allF theG elements in are uniformly structurallyF stable. A A STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 5

Next, the definition of Anosov families will be given. It is important to keep fixed the Riemann- ian metric on each Mi, since the notion of Anosov family depends on the Riemannian metric. The hyperbolicity for the family could be induced by the Riemannian metrics , i (see [1], Example 4). h· ·i Definition 2.6. An Anosov family on is a non-stationary dynamical system ( , , , ) such that: M M h· ·i F i. the tangent bundle T has a continuous splitting Es Eu which is D -invariant, i. e., for each p , T =MEs Eu with D (Es ) = Es ⊕and D (Eu) =FEu , where T is ∈ M pM p ⊕ p pF p (p) pF p (p) pM the tangent space at p; F F ii. there exist constants λ (0, 1) and c > 0 such that for each i Z, n 1, and p M , we have: ∈ ∈ ≥ ∈ i n n s n n u D ( )(v) cλ v if v E and D ( − )(v) cλ v if v E . k p Fi k ≤ k k ∈ p k p Fi k ≤ k k ∈ p s u The subspaces Ep and Ep are called stable and unstable subspaces, respectively. If we can take c = 1 we say the family is strictly Anosov. We will denote by 1( ) the set consisting of Anosov family on such that each f is a C1-diffeomorphism. A M M i In a natural way, Fisher and Arnoux in [1] generalized the notion of Markov partition to non- stationary dynamical systems, whose associated symbolic representation is a combinatorially de- fined two-sided sequence of maps, which they called a nonstationary subshift of finite type. This is a key tool for the further study of Anosov families. Remark 2.7. Proposition 3.5 in [14] proves the notion of Anosov family does not depend on Riemannian metrics chosen uniformly equivalent on the total space . On the other hand, taking a fixed Riemannian manifold M and considering M = M for each iM Z, we can suitably change i ∈ the Riemannian metric on each Mi for each i Z such that the sequence (Ii)i Z, where Ii : M M is the identity map, has a hyperbolic behavior∈ (see [1], Example 4). We want∈ to exclude that→ kind of cases. Therefore, we will suppose that

̺ := inf ̺i > 0, i Z ∈ 1 where ̺i is an injectivity radius of Mi. Consequently, for each i Z and p Mi, the exponential map exp : B(0 ,̺) B(p,̺)isadiffeomorphism and ∈ ∈ p p → v = d (exp (v), p), for all v B(0 ,̺), k k i p ∈ p where B(0 ,̺) is the ball in T M with radius ̺ and center 0 T M , the zero vector in T M , and p p i p ∈ p i p i B(p,̺) is the ball in Mi with radius ̺ and center p. Definition 2.8. An Anosov family satisfies the property of the angles (or s.p.a.) if the angle be- tween the stable and unstable subspaces are bounded away from zero. Remark 2.9. In [14], Corollary 3.8, we proved that if s.p.a. then there exists a Riemannian norm , uniformly equivalent to on T , with whichF is strictly Anosov, with constant λ˜ (λ,k·k1).∗ is uniformly equivalentk·k to the normM given by F ∈ k·k∗ = s u (2.5) (vs, vu) ⋆ max vs , vu , for (vs, vu) Ep Ep, p . k k {k k∗ k k∗} ∈ ⊕ ∈M Consequently, there exists C 1 such that ≥ (2.6) (1/C) v v C v , for every v Es Eu, p . k k⋆ ≤k k ≤ k k⋆ ∈ p ⊕ p ∈M 1In [1], Example 4, we have ̺ = 0. 6 JEOVANNYDEJESUSMUENTESACEVEDO

From now on, will denote the norm given in (2.5). k·k⋆ In [14] we proved that 1( ) is an open subset of 1( ) with respect to the strong topology. A M F M This fact means that if ( fi)i Z is an Anosov family, then there exists a two-sided sequence of positive ∈ 1 1 1 numbers (ǫi)i Z such that if (gi)i Z ( ) and d ( fi, gi) <ǫi for any i Z, then (gi)i Z ( ). ∈ ∈ ∈D M ∈ ∈ ∈A M If we do not ask for any additional condition on the family ( fi)i Z, the sequence (ǫi)i Z could not be bounded away from zero. ∈ ∈ 3. Some examples of Anosov families It is clear that if for each i Z f is an fixed Anosov diffeomorphism φ : M M, where M ∈ i → is a compact Riemannian manifold with Riemannian metric , , then ( fi)i Z is an Anosov family, h· ·i ∈ 2 considering Mi = M i endowed with the metric induced by , . Furthermore, if φ is C and 1+1 ×{ } h· ·i each fi is a C small perturbation of φ, then ( fi)i Z is an Anosov family (see [18], Proposition 2.2). In this section we will show some results provide∈ many examples of Anosov families which are not necessarily sequences of Anosov diffeomorphisms (or small perturbations of a single Anosov diffeomorphism). Definition 3.1. Let and be non-stationary dynamical systems on and , respectively. We F F M M say that is a gathering of if there exists a strictly increasing sequence of integers (ni)i Z such =F = eF f ∈ that Mi Mni and i fni+1 1 fni+1 fni : e F − ◦···◦ ◦ ˜ = ˜= fi 1 fni 1 fni 1 fi fni+1 1 fni e e − − ◦···◦ − − ◦···◦ Mni 1 Mni Mni+1 ··· − −−−−−−−−−−−−−→ −−−−−−−−−−−→ ··· It is not difficult to prove that: Proposition 3.2. Any gathering of an Anosov family is also an Anosov family. Example 3.3. It follows from Proposition 3.2 that if φ : M M is an Anosov diffeomorphism, ni → then for each sequence of positive integers (ni)i Z, if fi = φ , then ( fi)i Z is an Anosov family. ∈ ∈ In dimension 2, a necessary and sufficient condition for a family of matrices = (Ai)i Z acting on the 2-torus T2 = R2/Z2 by multiplication on the column vectors, is that thereA exist constants∈ c > 0 and σ > 1 such that n(x) cσn for all x T2 and n 1, i Z kAi k ≥ ∈ ≥ ∈ where is the norm on T2 inherited from R2 (see [17], Proposition 2.1). k·k The following example, which is due to Arnoux and Fisher [1], proves that Anosov families are not necessarily sequences of Anosov diffeomorphisms.

Example 3.4. For any sequence of positive integers (ni)i Z set ∈ 1 0 1 ni Ai = for i even and Ai = for i odd, ni 1 ! 0 1 ! 2 acting on the 2-torus Mi = T . The family (Ai)i Z is called the multiplicative family determined by ∈ 2 the sequence (ni)i Z. Let be the Riemannian metric on Mi inherited from R . For each i Z, let s = (a , b ), u = (∈c , d ) andk·kλ (0, 1) be such that a d + c b = 1, ∈ i i i i i i i ∈ i i i i di for i even, ai = [nini+1...], bi = 1, = [ni 1ni 2...], and λi = ai, ci − − and ci for i odd, bi = [nini+1...], ai = 1, = [ni 1ni 2...] and λi = bi. di − − STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 7

Here, 1 + = [nini 1...] 1 . ni + ni+1+ ··· For all i Z and n 1, we have ∈ ≥ n n 1 1 1 (3.1) Ai si cλi+n 1 λi si and Ai ui c− λi−+n 1 λi− ui , k k ≤ − ··· k k k k ≥ − ··· k k = si ui where c max sup , k k , sup , k k (c < because v (1/2, √2) for all v si : i i j s j i j u j ∞ k k ∈ ∈ { ∈ Z u : i Z ).n n k k o n k k oo }∪{ i ∈ } Note that, if there exists λ (0, 1) such that λ λ for all i, we have ∈ i ≤ n n n 1 n A s cλ s and A u c− λ− u for all n 1. k i ik ≤ k ik k i ik ≥ k ik ≥ This shows that, if there is a λ (0, 1) such that λi λ for all i, then (Ai)i Z is an Anosov family, ∈ ≤ ∈ with constants λ and c as defined above, the stable subspaces are spanned by si and the unstable subspaces are spanned by ui. However, we have: Proposition 3.5. Any multiplicative family is an Anosov family with constant λ = √2/3 and 2c.

Proof. Notice that, if λ j (2/3, 1) for some j Z, then λ j 1 (0, 2/3) and λ j+1 (0, 1/2). Indeed, = 1 ∈ = ∈ − ∈ ∈ if λ j + 1 (2/3, 1) we must have n j 1 and n j+1 2. Hence, n j n + + j 1 ··· ∈ ≥ 1 1 1 λ = < and λ + = < 1/2. j 1 1 + j 1 1 − n j 1 + 1 (1/2) n j+1 + 1+ n j+2+ − ··· ··· n Next, by induction on n, we prove that cλi+n 1 λi < 2cλ , for each i Z and n 1. Fix i Z. − n··· ∈ ≥ ∈ m It is clear that if n = 1, 2, then cλi+n 1 λi < 2cλ . Let n 2 and assume that cλi+m 1 λi < 2cλ − ··· ≥ n+1 − ··· for each m 1,..., n . Clearly, if λ + 2/3, then cλ + λ < 2cλ . On the other hand, if ∈ { } i n ≤ i n ··· i λi+n > 2/3, then λi+n 1 < 1/2 and by induction assumption we have − n 1 1 n 1 2 n 1 n+1 cλi+nλi+n 1 λi < (λi+nλi+n 1)2cλ − < 2cλ − <λ 2cλ − = 2cλ . − ··· − 2 · It follows from (3.1) and the above facts that n n n 1 n A s 2cλ s and A u (2c)− λ− u , k i ik ≤ k ik k i ik ≥ k ik for each i Z and n 1, which proves the proposition.  ∈ ≥ By Proposition 3.2 we have any gathering of a multiplicative family is an Anosov family. On the other hand, if F S L(N, 2), then i ∈ 1 0 1 ni,ki 1 1 0 1 ni,1 (3.2) Fi = − , ni,ki 1 ! 0 1 ! ··· ni,2 1 ! 0 1 ! N for some non-negative integers ni,1,..., ni,ki , that is, S L( , 2) is a semigroup generated by 1 0 1 1 M = and N = 1 1 ! 0 1 ! (see [1], Lemma 3.11).

Corollary 3.6. Consider a sequence (Fi)i Z inSL(N, 2) and the factorization of each Fi as in (3.2). Z∈ If ni,ki and ni,1 are non-zero for each i , then (Fi)i Z is an Anosov family. ∈ ∈ Proof. Notice that (Fi)i Z is a gathering of an multiplicative family. It follows from Proposition 3.5 ∈ that (Fi)i Z is an Anosov family.  ∈ 8 JEOVANNYDEJESUSMUENTESACEVEDO

Corollary 3.6 provides a great variety of examples of Anosov families. Next, suppose that X = B1,..., Bk S L(N, 2). If each Fi X, then (Fi)i Z is an Anosov family (see [17], Propo- { } ⊆ ∈ ∈ sition 2.7). Another examples are provided by Theorem 5.1 in [1]: if (Ai)i Z S L(N, 2) is a non- eventually constant sequence of matrices with non-negative entries, then it∈ is an⊆ Anosov family on T2.

4. Characterization of Anosov families Let M be a compact Riemannian manifold and denote by Γ(M) the set consisting of continuous sections of TM. Foradiffeomorphism f on M, set f : Γ(M) Γ(M) the bounded linear operator 1 ∗ → defined by f (ζ) = Df ζ f − for any ζ Γ(M). J. Mather in [10], [11] proved that f is an Anosov diffeomorphism∗ if and◦ only◦ if f is a hyperbolic∈ automorphism. Furthermore, he proved that this is equivalent to show that f I is∗ an automorphism on Γ(M), where I is the identity on Γ(M). We can define the operator f ∗for− non-stationary dynamical systems. In this section we will give some conditions on an Anosov∗ family to obtain the hyperbolicity of f defined for such family. ∗ Definition 4.1. For τ > 0 and i Z, set: ∈ (i) D(I , τ) = h : M M : h is C0 and d(h(p), I (p)) τ for any p M ; i { i → i i ≤ ∈ i} (ii) (τ) = (hi)i Z : hi D(Ii, τ) for any i Z ; D { ∈ ∈ ∈ } (iii) Γ(Mi) = ζ : Mi TMi : σ is a continuous section ; Γ ={ Γ → } (iv) τ(Mi) ζ (Mi) : supp Mi ζ(p) τ ; Γ = { ∈ Γ ∈ k k ≤ } = (v) ( ) (ζi)i Z : ζi (Mi) and supi Z ζi Γi < , where ζ Γi maxp Mi ζ(p) . Γ M = ∈ Γ∈ Γ ∈ k k ∞ Z k k ∈ k k (vi) τ( )  (ζi)i Z ( ): ζi τ(Mi) for each i ; M { ∈ ∈ M ∈ ∈ } Γ Γ Γ It is clear that ( ) is a proper subset of i∞= (Mi). Note that (Mi) is a Banach space with M −∞ the norm Γ . Therefore: k·k i Q Γ = Lemma 4.2. ( ) is a Banach space with the norm (σi)i Z supi Z σi Γi . M k ∈ k∞ ∈ k k 1 Definition 4.3. For any n.s.d.s. = ( fi)i Z ( ), define F ∈ ∈D M F : (F) Γ( ) D → M (ζi)i Z (Fi 1(ζi 1))i Z ∈ 7→ − − ∈ Γ Γ = 1 where Fi : (Mi) (Mi+1) is defined by the formula Fi(ζ)(p) D f 1(p)( fi)(ζ( f − (p))), for → i− i p M + ,ζ Γ(M ), and ∈ i 1 ∈ i (F) = ζ Γ( ): F(ζ) Γ( ) . D { ∈ M ∈ M } It is not difficult to prove that F is a linear operator and

(4.1) F = sup Fi(ζi) sup Dfi . ∞ k k (ζi) =1 k k ≤ i Z k k k k∞ ∈ Therefore, if supi Z Dfi < , then F is a bounded linear operator, and, in this case, ∈ k k ∞ (F) = Γ( ). D M However, in general, we do not have supi Z Dfi < (see Example 3.4). Consequently, F could be an unbounded operator. ∈ k k ∞ In this section we will suppose that that 1( ) with constants λ (0, 1) and c 1. Consider the splitting TM = Es Eu inducedF by ∈AforM each i Z. Set ∈ ≥ i i ⊕ i F ∈ Γs(M ) = σ Γ(M ) : theimageof σ is contained in Es i { ∈ i } STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 9 and Γu(M ) = σ Γ(M ) : theimageof σ is contained in Eu . i { ∈ i } It is clear that s u s s u u Γ(M ) = Γ (M ) Γ (M ), F (Γ (M )) = Γ (M + ) and F (Γ (M )) = Γ (M + ). i i ⊕ i i i i 1 i i i 1 For t = s, u, set t t (4.2) Γ ( ) = (σi)i Z Γ( ): σi Γ (Mi) for each i Z . M { ∈ ∈ M ∈ ∈ } Note that Γs( ) and Γu( ) are closed subspaces of Γ( ) and Γs( ) Γu( ) = 0˜ , the zero vector in Γ( ). M M M M ∩ M { } M Lemma 4.4. Γs( ) and Γu( ) are complementary subspaces of Γ( ) (that is, Γ( ) = Γs( ) Γu( )) if and onlyM if satisfiesM the property of angles. M M M ⊕ M F s u Proof. Suppose that s.p.a. Let us prove that Γ( ) = Γ ( ) Γ ( ). For ζ = (ζi)i Z Γ( ), take F M M ⊕ M ∈ ∈ M (4.3) ζ s(p) = πs(p)ζ (p) and ζu(p) = πu(p)ζ (p), for i Z, p M , i i i i ∈ ∈ i s u s u where π (p) and π (p) are the projections on Ep and Ep, respectively. Since satisfies the property of the angles, there exists a K > 0 such that F s u Z ζi (p) K ζ and ζi (p) K ζ , for i , p Mi. k k ≤ k k∞ k k ≤ k k∞ ∈ ∈ s = s Γs u = u Γu = s + u Γ = Therefore ζ (ζi )i Z ( ),ζ (ζi )i Z ( ) and ζ ζ ζ , which proves that ( ) Γs( ) Γu( ). ∈ ∈ M ∈ ∈ M M MNow,⊕ supposeM that does not satisfies the property of angles. Without loss of generality, we F N can assume there exist p M0 and a subsequence k1 < k2 < k3 < ... in such that the angle θki s∈ u + between the subspaces E ki and E ki converges to zero as i . We can suppose that the 0 (p) 0 (p) → ∞ Es FEu F π/ θ <θ i i angle between ki and ki is less than 4 and ki+1 ki for every 1. For each 1, take 0 (p) 0 (p) ≥ ≥ s s Fu u F v E k and v E k such that for all i 1, ki i (p) ki i (p) ∈ F0 ∈ F0 ≥ 1 1 vs vu = π θ , < vu < , and vs = 1 sin2(θ ) vu 2 cos(θ ) vu , ki ki ki ki ki ki ki ki ki − sin(θki 1 ) k k sin(θki ) k k − k k − k k − q whered vs vu is the angle between vs and vu . Since ki ki ki ki s + u 2 = s 2 + u 2 + s u d v v v v 2 cos(θki ) v v for each i 1, k ki ki k k ki k k ki k k ki kk ki k ≥ we can prove that vs + vu = 1 for each i 1. ki ki u k k s ≥ However, v as i and, therefore, v as i . Take ζ = (ζ ) Z such that ki ki i i k k k→∞s u → ∞ k k→∞ → ∞ s s ∈ u u ζ ( i (p)) = v +v and ζ Γ 1 for any i Z. Note that the sections ζ = (ζ ) Z and ζ = (ζ ) Z ki ki ki i i i i i i F k k ≤ s ∈ Γs u Γu = s +∈ u s ∈ defined in (4.3) are the only that satisfy ζi (Mi),ζi (Mi) and ζ ζ ζ . However ζi is not bounded and therefore does not belong∈ to Γs( ). Hence∈ ζ Γ( ) Γs( ) Γu( ), that is, Γs( ) and Γu( ) are not complementary in Γ( M) ∈ M \ M ⊕ M  M M M Every C1-Anosov diffeomorphism φ : M M defined on a compact Riemannian manifold M satisfies the property of angles, because the→ compactness of M. Hence, the subspaces Γs(M) and Γu(M) defined in (4.2) are complementary subspaces in Γ(M). Example 2.3 in [12] proves that there exist Anosov families which do not satisfy the property of angles. In that case, for any s u p M0, the angle between E n(p) and E n(p) converges to zero as n . ∈ F F → ±∞ 10 JEOVANNYDEJESUSMUENTESACEVEDO

Now, following the Mather’s ideas in [10] and [11] we obtain the next characterization for Anosov families which s.p.a. and with bounded derivative. Theorem 4.5. F : Γ( ) Γ( ) is a bounded hyperbolic automorphism if and only if 1 M → M F ∈ ( ) s.p.a. and supi Z Dfi < . A M ∈ k k ∞ 1 Proof. Suppose that ( ) s.p.a. and supi Z Dfi < . Thus F is a bounded automorphism on Γ( ). If n 1, thenF ∈A M ∈ k k ∞ M ≥ n Γs (4.4) Fi+n 1 Fi(ζ) Γi+n cλ ζ Γi if ζ (Mi) k − ◦···◦ k ≤ k k ∈ and 1 n Γu (4.5) Fi+n 1 Fi(ζ) Γi+n c− λ− ζ Γi if ζ (Mi). k − ◦···◦ k ≥ k k ∈ It follows from Lemma 4.4 that Γs( ) and Γu( ), defined in (4.2), are complementary sub- spaces of Γ( ). Furthermore, we have theyM are invariantM by F. Therefore, M σ(F) = σ(F Γs( )) σ(F Γu( )) | M ∪ | M (see [5]). It follows from (4.4) and (4.5) that for any n 1 ≥ n n s n 1 n u F (ζ) cλ ζ for ζ Γ ( ) and F (ζ) c− λ− ζ for ζ Γ ( ). k k∞ ≤ k k∞ ∈ M k k∞ ≥ k k∞ ∈ M Hence n 1/n n 1/n 1 lim (F Γs( )) λ < 1 and lim (F Γu( ))− λ− > 1. n M n M →∞ k | k ≤ →∞ k | k ≥ This fact proves that 1 σ(F Γs( )) z C : z λ and σ(F Γu( )) z C : z λ− | M ⊆{ ∈ k k ≤ } | M ⊆{ ∈ k k ≥ } (see [5]). Hence F is a hyperbolic automorphism. Now, assume that F : Γ( ) Γ( ) is a bounded hyperbolic automorphism. Thus, there exist two closed subspaces Γ1 andMΓ2 →of Γ(M) such that M i. Γ( ) = Γ1 Γ2; ii. F(MΓ1) = Γ1 and⊕ F(Γ2) = Γ2; iii. there exist λ , λ (0, 1) such that 1 2 ∈ n 1/n n 1/n (4.6) lim (F Γ1 ) = λ1 and lim (F Γ2 )− = λ2; n n →∞ k | k →∞ k | k = Z Z 1 iv. σ(F) σ(F Γ1 ) σ(F Γ2 ), with σ(F Γ1 ) z :0 < z λ1 and σ(F Γ2 ) z : λ2− z (see [5]). | ∪ | | ⊆{ ∈ | | ≤ } | ⊆{ ∈ ≤| |} For each j Z, set ∈ Γ1 = = Γ = n 1/n j ζ (ζi)i Z ( ): ζi 0 for i , j and limsup F (ζ) λ1 { ∈ ∈ M n k k ≤ } →∞ and Γ2 = = Γ = n 1/n j ζ (ζi)i Z ( ): ζi 0 for i , j and limsup F− (ζ) λ2 . { ∈ ∈ M n k k ≤ } →∞ Thus Γ1 and Γ2 are subspaces of Γ( ) for each j Z. We can prove that j j M ∈ 1 1 2 2 (4.7) F(Γ ) = Γ + and F(Γ ) = Γ + for each j Z. j j 1 j j 1 ∈ Next, the projection Π : Γ( ) Γ(M ) j M → j (ζi)i Z ζ j ∈ 7→ STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 11 induces the isomorphism

r r Π j Γr : Γ Π j(Γ ) Γ(M j) | j j → j ⊆ (ζi)i Z ζ j ∈ 7→ for r = 1, 2 and j Z. Since Γ( ) = Γ1 Γ2, we have ∈ M ⊕ (4.8) Γ(M ) =Π (Γ1) =Π (Γ1) Π (Γ2). j j j j j ⊕ j j From (4.7) we can prove that

r r (4.9) Df (Π (Γ )) =Π + (Γ + ) for j Z and r = 1, 2. j j j j 1 j 1 ∈ Π Γr By the definition of j( j), we obtain

Π (Γ1) = ζ Γ(M ) : limsup Df n(ζ) 1/n λ j j i Γ j+n 1 { ∈ n k k ≤ } →∞ 2 n 1/n Π (Γ ) = ζ Γ(M ) : limsup Df − (ζ) λ . j j i Γ j n 2 { ∈ n k k − ≤ } →∞ ffi Π Γr 0 R Γ = It is not di cult to prove that j( j)isa C (M j, ) submodule of (M j) for r 1, 2, where C0(M , R) = φ : M R : φ is a continuous map . j { j → } Γ Γ1 Γ2 Therefore, it follows from [16], Theorem 2, that (M j) is isomorphic to TM j and j and j are, re- 1 2 Z = spectively, isomorphic to the subbundles E j and E j of TM j for each j . Taking λ max λ1,λ2 , it follows from (4.6) that there exists c > 0 such that ∈ { }

n n 1 n n 2 (4.10) D ( )(v) cλ v if v (E ) and D ( − )(v) cλ v if v (E ) . k p F j k ≤ k k ∈ j p k p F j k ≤ k k ∈ j p We have from (4.8), (4.9) and (4.10) that is an Anosov family with constants λ (0, 1) and r F ∈ c > 0 (the continuity of each (E j)p on p if follows from Proposition 3.4 in [14]). Next, we will prove that satisfies the property of angles. It is clear that Γs( ) Γ1 and u 2 F 1 s M1 ⊆ Γ ( ) Γ . Let us show that Γ Γ ( ). Suppose that there exists ζ = (ζi)i Z Γ , such that

n n n u n s 1 n u n s F (ζ )(p) = D (ζ (p)) D (v (p)) D (v (p)) c− λ− v (p) cλ v (p) . k i i k k Fi i k≥k Fi i k−k Fi i k ≥ k i k − k i k n n 1/n This implies that limsupn + F (ζ) and therefore lim supn + F (ζ) 1, which con- → ∞ k k→∞ → ∞ k k ≥ s tradicts (4.6). Consequently, ζi(p) = 0 for all p and hence ζ and hence ζ Γ ( ). This proves that Γ1 Γs( ). Analogously we can∈ M prove that Γ2 ∈Γ Mu( ). Therefore∈Γ( M) = Γs( ) Γu( ). It follows⊆ M from Lemma 4.4 that s.p.a. Finally, by⊆ the definitionM of F, it isM clear M ⊕ M = F  that F is bounded if and only if supi Z Dfi . ∈ k k ∞ = Proposition 3.5 provides of many examples of Anosov families ( fi)i Z with supi Z Dfi . A question that arises from Theorem 4.5 is: ∈ ∈ k k ∞

Question 4.6. What kind of operator is F if = ( fi)i Z does not satisfy the property of angles and = F ∈ (or) supi Z Dfi ? ∈ k k ∞ 12 JEOVANNYDEJESUSMUENTESACEVEDO

5. Openness of 2( ) Ab M As we said in Introduction, in [14] we proved ( ) is open in 1( ) endowed with the A M D M strong topology, that is, for each ( ) there exists a sequence of positive numbers (δi)i Z 1 F ∈A M ∈ such that B ( , (δi)i Z) ( ). In that case, if we do not ask for any additional condition on , F ∈ ⊆A M F it is not always possible to take the sequence δi bounded away from zero, that is, δi could decay as i . → ±∞ Set 2 = = 2 b( ) ( fi)i Z ( ): is Anosov, s.p.a. and sup Dfi C2 < , A M {F ∈ ∈D M F i Z k k ∞} ∈ 1 2 2 1 2 where φ C2 = max Dφ , Dφ− , D φ , D φ− for a C -diffeomorphism φ. The goal of this k k k k k k k k2 k k section is to show forn any = ( fi)i Z ( ), thereo exists a ε > 0 such that: F ∈ ∈Ab M B2( ,ε) = 2( ): d2 ( , ) <ε 2( ). F {G∈D M uni f F G }⊆Ab M The basic neighborhood B2( ,ε) is called uniform. F From now on, will be an Anosov family in 2( ) with constants λ (0, 1) and c 1. F Ab M ∈ ≥ We will work using exponential charts (see Section 2). Take

S = sup Dfi C2 and r (0,̺/20S ). F i Z k k ∈ F ∈ 2 Fix = (gi)i Z B ( , r). If p, q Mi and d(p, q) r, then we have d( fi(p), gi(q)) < ̺/2 and G1 1 ∈ ∈ F ∈ ≤ 1 d( fi− (p), gi− (q)) < ̺/2. Therefore, gi(B(p, r)) B( fi(p),̺/2) and gi− (B( fi(p), r)) B(p,̺/2). Consequently, ⊆ ⊆ = 1 g˜ p expf−(p) expp : B(0p, r) B(0 (p),̺/2) ◦G◦ → F 1 = 1 1 andg ˜−p exp−p − exp (p) : B(0 (p), r) B(0p,̺/2), ◦ G ◦ F F → are well-defined for each p . ∈M ˜ = Z Proposition 5.1. For τ > 0, there exist r˜ > 0, δ > 0 and Xi p1,i,..., pmi,i Mi for each i , such that M = mi B(p , r˜) for each i Z and, furthermore, for{ every B}2 ⊆( , δ˜), we have∈ that i ∪ j=1 j,i ∈ G ∈ F σ , = ˜ , ˜ 1 1 < τ. (˜r ) sup  sup Dz(D0( fp) g˜ p) ⋆ sup Dz(D0( fp− ) g˜−p ) ⋆ G p Xi,i Z z B(0p,r˜) k − k z B(0 f (p),r˜) k − k ∈ ∈  ∈ ∈ i   s u  Proof. Fix p Mi. Let z B(0p, r) and (v, w) E E . There exists K > 0 (which does not depends on p),∈ such that ∈ ∈ ⊕ 2 D0( f˜p)(v, w) Dz( f˜p)(v, w) ⋆ K[1 + Dfi ⋆] D fi ⋆ z (v, w) ⋆ k − k ≤ k k k k k k∗k k (see [9], p. 50). Thus, D (D ( f˜ ) g˜ )(v, w) = D ( f˜ )(v, w) D (˜g )(v, w) k z 0 p − p k⋆ k 0 p − z p k⋆ D ( f˜ )(v, w) D ( f˜ )(v, w) + D ( f˜ g˜ )(v, w) ≤k 0 p − z p k⋆ k z p − p k⋆ 2 K[1 + S ]S z ⋆ (v, w) ⋆ + d ( fi, gi) (v, w) ⋆. ≤ F F k k k k k k 2 Therefore, if δ˜1 < τ/2 andr ˜1 < τ/2(K(1 + S )S ), then for each B ( , δ˜1) and z B(0p, r˜1), F F G ∈ F ∈ we have Dz(D0( f˜p) g˜ p) < τ. k − k 1 Analogously we can prove there exist δ˜2 > 0 andr ˜2 > 0 such that, if B ( , δ˜2) and 1 1 G ∈ F z B(0 , r˜ ), then D (D ( f˜− ) g˜− ) τ. ∈ fi(p) 2 k z 0 p − p k⋆ ≤ STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 13

Taker ˜ = min r˜1, r˜2 and δ˜ = min δ˜1, δ˜2 . Notice that neitherr ˜ nor δ˜ depend on p. Since Mi is { } { =} = mi compact, we can choose a finite subset Xi p1,i,..., pmi,i Mi such that Mi j=1B(p j,i, r˜) for each i Z, which proves the proposition. { } ⊆ ∪  ∈ 1 λ˜ From now on, we will fix α (0, − ). Set ∈ 1+λ˜ 1 1 (λ˜ − λ˜)α λ˜ − (1 α) (1 + α)α σ := min − , − − . A ( 2(1 + α)2 2(1 + α) ) ˜ 1 Furthermore, we will suppose thatr ˜ and δ 2 min r˜,σA are small enough such that σ(˜r) < σA. Z ≤ { =} 1 ˜ s = 1 s Fix p Xi for some i . For q B(p, r˜), consider z exp−p (q), Ez Dq(exp−p expq)(Eq), u ∈ 1 ∈ u ∈ ◦ E˜ = D (exp− exp )(E ), z q p ◦ q q s = ˜ s ˜u Kα, ,z (v, w) Ez Ez : w ⋆ < α v ⋆ (0, 0) , F { ∈ ⊕ k k k k }∪{ } u = ˜ s ˜u and Kα, ,z (v, w) Ez Ez : v ⋆ < α w ⋆ (0, 0) . F { ∈ ⊕ k k k k }∪{ } The next two lemmas can be shown analogously to the Lemmas 4.2 and 4.3 in [14], respec- tively. Lemma 5.2. Let B1( , δ˜). For each i Z,if p X we have: G ∈ F ∈ ∈ i i. D (˜g )(Ku ) Ku for all z B(0 , r˜), and z p α, ,z α, ,g˜ p(z) p F ⊆ F ∈ 1 s s ii. Dz(˜gp)− (K ) K for all z B(0 (p), r˜). α, ,g˜ p(z) α, ,z F ⊆ F ∈ F See Figure 5.1.

u ˜u ˜u u Kα, ,z Ez Eq Kα, ,q F D(˜gp)z F Tz M Tq M Ks Ks α, ,z 1 α, ,q F D(˜gp)q− F ˜ s ˜ s Ez Eq

Figure 5.1. Stable and unstable invariant α-cones. q = g˜ p(z)

2 Lemma 5.3. There exists η (0, 1) such that, if B ( , δ˜) then, for p Xi and z B(0p, r˜) we have ∈ G ∈ F ∈ ∈ 1 u i. Dz(˜gp)(y) ⋆ η− y ⋆ for each y Kα, ,z; k 1 k ≥ 1k k ∈ sF ii. Dz(˜g− )(y) ⋆ η− y ⋆ for each y K . p α, ,g˜ p(z) k k ≥ k k ∈ F Now, take ξ 1 min r˜,σ (see (2.6)). Using the Lemmas 5.2 and 5.3 we can prove that: ≤ 2C { A} 2 Lemma 5.4. Let = (gi)i Z B ( ,ξ). For each p , take G ∈ ∈ F ∈M ∞ ∞ s = n s u = n u Fp D n(p)( − )(K n ) and Fp D n(p)( )(K n ). G G α, , (p) G− G α, , − (p) \n=0 F G \n=0 F G 14 JEOVANNYDEJESUSMUENTESACEVEDO

s u = s u The families Fp and Fp are D -invariant subspaces and T p Fp Fp for each p . G M ⊕ = s u ∈ M Furthermore, there exist η (0, 1) and c˜ 1 such that, with the splitting T p Fp Fp, is an Anosov family with constant∈ η and c,˜ which≥ s.p.a. (see Figure 5.2). M ⊕ G

u u Ep Kα, ,p F Tp M

Fs s ps,1 Kα ,p Fp,2 sF s Ep Fp,3

u u u Fp,1Fp,2Fp,3

igure r = n k s = = F 5.2. Fp,n k=1 D k(p) ± (K k ), for r s, u and n 1, 2, 3. G± G α, , ± (p) T F G Proof. See [14], Lemmas 4.4 and 4.5.  Consequently, we have: 2 2 2 Theorem 5.5. For all b( ), there exists ξ > 0 such that B ( ,ξ) b( ). That is, 2( ) is open in ( 2(F), ∈A τ ).M F ⊆ A M Ab M D M uni f 6. Local Stable and Unstable Manifolds for Anosov Families In [12], we give conditions for obtain stable and unstable manifolds at each point of each component Mi. In that case, the size of each submanifold could decay as i . In Theorems 2 → ±∞ 6.2 and 6.3 we will see that each b( ) admits stable and unstable manifold with the same size at each point. F ∈A M Given two points p, q , set ∈M 1 1 Θ = n n ∆ = n n p,q lim sup log d( i (q), i (p)) and p,q lim sup log d( i− (q), i− (p)). n n F F n n F F →∞ →∞ Definition 6.1. Let ε > 0. Fix p . ∈M s = Θ n n = (i) (p,ε) q B(p,ε) : p,q < 0 and i (q) B( i (p),ε) for n 1 : the local stable Wset at p; { ∈ F ∈ F ≥ } u = ∆ n n = (ii) (p,ε) q B(p,ε) : p,q < 0 and i− (q) B( i− (p),ε) for n 1 : the local unstableW set at{ p.∈ F ∈ F ≥ } 2 ˜ ˜ 1 ˜ 2 Take b( ) and λ as in Remark 2.9. Fix α (λ− 1)/2 and γ (λ , 1). It follows from PropositionF ∈A 5.1 thatM there exists δ > 0 such that ≤ − ∈ 1 1 (λ˜ − λ˜)α (γλ˜ − λ˜) 1 λ˜ σ(δ, ) < min − , − , − . F ( (1 + α)2 (1 + α)(1 + γ) 1 + λ˜ ) Thus, satisfies the assumption from Proposition 3.3 in [12]. Therefore, there exists ζ (0, 1) such that:F ∈ STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 15

Theorem 6.2. For each p , u(p, δ) is a differentiable submanifold of and there exists Ku > 0 such that: ∈ M W M u = u (i) T p (p, δ) Ep, W1 u u 1 (ii) − ( (p, δ)) ( − (p), δ), F W u ⊆W F n n u n (iii) if q (p, δ) and n 1 we have d( − (q), − (p)) K ζ d(q, p). ∈W ≥ F F ≤ u (iv) Let (pm)m N be a sequence in Mi converging to p Mi as m . If qm (pm, δ) converges∈ to q B(p, δ) as m , then q u(p, δ∈). → ∞ ∈ W ∈ →∞ ∈W Proof. See [12], Theorems 3.7 and 4.5.  Theorem 6.3. For each p , s(p, δ) is a differentiable submanifold of and there exists Ks > 0 such that: ∈ M W M s = s (i) T p (p, δ) Ep, (ii) (W s(p, δ)) s( (p), δ), (iii) ifF qW s(p, δ⊆W) and nF 1 we have d( n(q), n(p)) Ksζnd(q, p). ∈W ≥ F F ≤ s (iv) Let (pm)m N be a sequence in Mi converging to p Mi as m . If qm (pm, δ) converges∈ to q B(p, δ) as m ,asq s(p, δ).∈ → ∞ ∈ W ∈ →∞ ∈W Proof. See [12], Theorems 3.8 and 4.6.  Ku and Ks from Theorems 6.2 and 6.3, respectively, depend on the constant c of , on the constant C in (2.6) and on the minimum angle between the stable and unstable subspacesF of the splitting T = Es Eu, which is positive because we are supposing that s.p.a. M ⊕ F 7. Structural stability of 2( ) Ab M 2 2 In this section we will show that b( ) is uniformly structurally stable in ( ): for each 2 A M 2 D M 2 b( ) there exists a uniform basic neighborhood B ( , δ) of such that, each B ( , δ) Fis uniformly ∈A M conjugate to . Since and (see RemarkF 2.9),F then any two-sidedG ∈ sequenceF F k·k k·k⋆ (hi : Mi Mi)i Z is equicontinuous with respect to the metric if and only if is equicontinuous with respect→ ∈ . Therefore, in order to show our result it isk·k sufficient to prove it considering k·k⋆ the metric ⋆ on . Hence, we will consider the metric ⋆ given in 2.5. To simplify the k·k M k·k 2 notation, we will omit the symbol “⋆” for this metric. Therefore, we will fix b( ) and we can suppose that is strictly Anosov with respect to , with constant λ (0, 1).F ∈A M k·k ∈ 2 In order to prove the structural stability of Anosov families in b( ) we have adapted the Shub’s ideas in [15] to prove the structural stability of Anosov diffeomorphismsA M on compact Rie- mannian manifolds. We will divide the proof of this fact into a series of lemmas and propositions. Throughout this section, we will considerr ˜ > 0, ξ > 0, ̺ > 0 and η [λ, 1) as in Section 5. ∈ Let 0 <ε ̺/2. We can identify (ε) with Γ ( ) (see Defifition 4.1) by the homeomorphism ≤ D ε M Φ : (ε) Γ ( ) D → ε M (hi)i Z (Φi(hi))i Z, ∈ 7→ ∈ 1 where Φ (h )(p) = exp− (h (p)), for p M . Note that Φ (I ) is the zero section in Γ(M ). i i p i ∈ i i i i 2 Lemma 7.1. Fix κ > 0. There exist ξ′ (0,ξ] and r′ (0, r˜/3] such that, if (gi)i Z B ( , ξ˜), the map ∈ ∈ ∈ ∈ F 1 : (r′) (κ), (hi)i Z (gi 1 hi fi−1)i Z G D →D ∈ 7→ − ◦ ◦ − ∈ is well-defined. 16 JEOVANNYDEJESUSMUENTESACEVEDO

Proof. It is sufficient to prove that there exist ξ′ > 0 and r′ > 0 such that, for every i Z, if ff 2 1 ∈ gi : Mi Mi+1 is a di eomorphism with d (gi, fi) < ξ′ and h D(Ii, r′), then gihfi− D(Ii+1, κ). For each→ continuous map h : M M , we have ∈ ∈ i → i 1 1 1 1 d(g hf − (p), I + (p)) d(g hf − (p), g f − (p)) + d(g f − (p), I + (p)). i i i 1 ≤ i i i i i i i 1 = + Z Let : supi Z Dfi < . If r′ < κ/2( 1) and ξ′ < min 1,κ/2 . For i , if S ∈ k k ∞ S { } ∈ 1 1 1 1 1 1 g B ( f ,ξ′) = g Diff (M , M + ) : max d (g, f ), d (g− , f − ) <ξ′ i ∈ i { ∈ i i 1 { i i } we have 1 = 1 1 1 d(gi fi− (p), Ii+1(p)) d(gi fi− (p), fi fi− (p)) d (gi, fi) <ξ′ < κ/2. 1 ≤ Furthermore, if h D(I , r′) and g B ( f ,ξ′), then ∈ i i ∈ i 1 1 1 1 d(g hf − (p), g f − (p)) Dg d(hf − (p), f − (p)) (ξ′ + )r′ < κ/2. i i i i ≤k ik i i ≤ S 1 1 Therefore, if h D(Ii, r′) and gi B ( fi,ξ′), we have that gi h fi− D(Ii+1, κ), which proves the lemma. ∈ ∈ ◦ ◦ ∈  Fix (7.1) ζ (0, min 1 λ, 1 η,̺/2 ). ∈ { − − } It follows from Lemma 7.1 that there exist

(7.2) r′ (0, r˜/3) and ξ′ (0, min ξ, r′(1 λ ζ) ), ∈ ∈ { − − } 2 such that if (gi)i Z B ( ,ξ′) then ∈ ∈ F G : Γ ( ) Γ ( ) r′ M → ζ M Φ Φ 1 (σi)i Z ( i i 1 i− 1(σi 1))i Z ∈ 7→ G − − − ∈ = 1 is well-defined, where i 1(h) gi 1 h fi−1, for h D(Ii 1, r′). Consequently, G − − ◦ ◦ − ∈ − 1 1 1 Φi i 1Φ− (σ)(p) = exp− gi 1 exp 1 σ( f − (p)), for p Mi,σ Γr (Mi 1). i 1 p fi−1(p) i 1 ′ G − − ◦ − ◦ − − ∈ ∈ − 2 Let F be the operator given in Definition 4.3 for . Since b( ), F is a bounded hyperbolic linear operator and (F) = Γ( ) (see Theorem 4.5).F F ∈A M D M Lemma 7.2. Take any ζ as in (7.1). There exist ξ′ (0,ξ] and r′ (0, r˜/3] such that, for each 1 ∈ ∈ i Z, if B ( ,ξ′), then ∈ G ∈ F Lip([F G] Γ (M)) < ζ, − | r′ where Lip denotes a Lipschitz constant.

Proof. For σ Γr (Mi 1) and p Mi we have ∈ ′ − ∈ Φ Φ 1 = 1 1 [Fi 1 i i 1 − ](σ)(p) [D f 1 p ( fi 1) exp− gi 1 exp 1 ](σ( f − (p))). i 1 i−1( ) p fi−1(p) i 1 − − G − − − − − ◦ − ◦ − − = 1 Let q fi−1(p). Note that − = 1 Dq( fi 1) D0q (exp−p fi 1 expq) where0q is the zero vector in Tq Mi 1. − ◦ − ◦ − For any v Tq Mi 1 with v <ξ, we have ∈ − k k 1 = 1 1 Dv[Dq( fi 1) exp−p gi 1 expq] Dv[D0q (exp−p fi 1 expq) exp−p gi 1 expq] − − ◦ − ◦ ◦ − ◦ − ◦ − ◦ = 1 1 D0q (exp−p fi 1 expq) Dv(exp−p gi 1 expq). ◦ − ◦ − ◦ − ◦ As we saw in the proof of Proposition 5.1, 1 1 + + 2 D0q (exp−p fi 1 expq) Dv(exp−p gi 1 expq) K[1 ] v d ( fi 1, gi 1). k ◦ − ◦ − ◦ − ◦ k ≤ SF SF k k − − STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 17

Hence if σ = (σi)i Z,σ ˜ = (σ ˜ i)i Z Γr (Mi 1) and p Mi we have ∈ ∈ ∈ ′ − ∈ Φ Φ 1 Φ Φ 1 [Fi 1 i i 1 i− 1](σi 1)(p) [Fi 1 i i 1 i− 1](σ ˜ i 1)(p) (σi 1)(p) (σ ˜ i 1)(p) k − − G − − − − − − G − − − k≤Jk − − − k σi 1 σ˜ i 1 Γi ≤ Jk − − − k σ σ˜ , ≤ Jk − k∞ 2 where = K[1 + f ] f v + d ( fi 1, gi 1). We can choose ξ′ and r′ small enough such that < ζ. Thus J S S k k − − J Φ Φ 1 Φ Φ 1 [Fi 1 i i 1 i− 1](σi 1) [Fi 1 i i 1 i− 1](σ ˜ i 1) Γi ζ σ σ˜ , k − − G − − − − − − G − − − k ≤ k − k∞ and therefore [F G](σ) [F G](σ ˜ ) ζ σ σ˜ , k − − − k∞ ≤ k − k∞ which proves the lemma. 

From now on we will suppose that ξ′ and r′ satisfy (7.2) and Lemma 7.2. Furthermore, we will 2 fix = (gi)i Z B ( ,ξ′). G ∈ ∈ F Lemma 7.3. G Γ ( ) has a fixed point in Γr ( ). | r′ M ′ M s u Γ = Γ Γ = Z Γ = + Proof. Since ( ) ( ) ( ), each σ (σi)i r′ ( ) can be written as σ σs σu, M s M ⊕ M u ∈ ∈ M where σs = (σi,s)i Z Γ ( ) and σu = (σi,u)i Z Γ ( ). Let G˜ be defined on Γr ( ) as ∈ ∈ M ∈ ∈ M ′ M 1 G˜ (σ) = (G(σ)) + (F− [σ + F(σ ) (G(σ)) ]) . s u u − u u Γ ˜ = = If σ r′ ( ) is a fixed point of G, we have (G(σ))s σs and (G(σ))u σu, that is, σ is a fixed point∈ of G.M Therefore, in order to prove the lemma, it is sufficient to find a fixed point of G˜ . First ˜ = Z + Z = Z + Z Γ we prove that G is a contraction. Let σ (σi,s)i (σi,u)i andσ ˜ (σ ˜ i,s)i (σ ˜ i,u)i in r′ ( ), where σ , σ˜ Γs( ) and σ , σ˜ Γu( ). Thus,∈ we can∈ prove that ∈ ∈ M s s ∈ M u u ∈ M ˜ ˜ + G(σ)i+1 G(σ ˜ )i+1 Γi+1 (λ ζ) σ σ˜ . k − k ≤ k − k∞ Hence, G˜ (σ) G˜ (σ ˜ ) (λ + ζ) σ σ˜ . Since λ + ζ < 1 (see (7.1)), G˜ is a contraction. Now k − k∞ ≤ k − k∞ we prove that G˜ (Γr ( )) Γr ( ). If 0 = (0i)i Z is the sequence of the zero sections, we have that ′ M ⊆ ′ M ∈ ˜ + + G(σ)i+1 Γi+1 (λ ζ) σ max λ (G(0)u)i+1 Γi+1 , (G(0)s)i+1 Γi+1 , k k ≤ k k∞ { k k k k } thus G˜ (σ) (λ + ζ) σ + G(0) . Now, for each i Z, p Mi, we have k k∞ ≤ k k∞ k k∞ ∈ ∈ = 1 = 1 1 Gi(0i)(p) exp−p gi exp f 1(p)(0 f 1(p)) exp−p (gi fi− (p)) k k k ◦ ◦ i− i− k k k = 1 = 1 1 d(gi fi− (p), p) d(gi fi− (p), fi fi− (p)) <ξ′ Γ ˜ + + ˜ Γ Consequently, if σ r′ ( ), then G(σ) < (λ ζ)r′ ξ′ < r′, that is, G(σ) r′ ( ). Therefore, G˜ has a fixed point∈ in Γ M( ). k k∞ ∈ M  r′ M s u Lemma 7.4. Let p M and v = (vs, vu), w = (ws, wu) B (0, r˜) B (0, r˜). If vs ws vu wu , then ∈ ∈ × k − k≤k − k 1 (˜g (v)) (˜g (w)) (η− ζ) v w (˜g (v)) (˜g (w)) . k p s − p sk ≤ − k u − uk≤k p u − p uk On the other hand, if v w v w , then k u − uk≤k s − sk 1 1 1 1 1 (˜g− (v)) (˜g− (w)) (η− ζ) v w (˜g− (v)) (˜g− (w)) . k p u − p uk ≤ − k s − sk≤k p s − p sk Proof. See [15], Lemma II.1.  18 JEOVANNYDEJESUSMUENTESACEVEDO

= s u = = n n Let v (vs, vu) B (0, r˜) B (0, r˜) and q expp(v). Suppose that, for n 1, d(g (p), g (q)) < r˜. Thus, if v v∈ , by Lemma× 7.4 we have ± k sk≤k uk 1 1 d(q, p) = v v + v 2 v 2(η− ζ)− (˜g (v)) (˜g (0 )) k k≤k sk k uk ≤ k uk ≤ − k p u − p p uk 1 1 2(η− ζ)− [ (˜g (v)) (˜g (0 )) + (˜g (v)) (˜g (0 )) ] ≤ − k p u − p p uk k p s − p p sk 1 1 1 1 2 √2(η− ζ)− g˜ (v) g˜ (0 ) 2 √2(η− ζ)− r˜. ≤ − k p − p p k ≤ − 1 1 Analogously if v v , then d(q, p) 2 √2(η− ζ)− r˜. Inductively, we can prove that: k uk≤k sk ≤ − Proposition 7.5. For each i Z,if p, q M and d(gn(p), gn(q)) < r˜ for each n [ N, N], then ∈ ∈ i i i ∈ − 1 N d(q, p) 2 √2(η− ζ)− r˜. ≤ − Finally, we have: 2 2 2 Theorem 7.6. b( ) is uniformly structurally stable in ( ), that is, any b( ) is uniformly structurallyA M stable. F M F ∈A M

2 Proof. Take (gi)i Z B ( ,ξ′). It follows from Lemma 7.3 that there exists = (hi)i Z, with ∈ ∈ F H ∈ hi D(Ii, r˜/3) for each i, such that gi hi = hi+1 fi for each i Z. We will prove that is ∈ ◦ ◦ ∈ 1 N H equicontinuous and each h is injective. Let α > 0. Take N > 0 such that 2 √2(η− ζ)− r˜ < α. i − Since ( fi)i Z is equicontinuous, the family of sequences ∈ n ( fi )i Z : n [ N, N] { ∈ ∈ − } is equicontinuous. Consequently, there exists β > 0 such that, for each i Z, if x, y Mi and n n ∈ Z ∈ d(x, y) < β, then d( fi (x), fi (y)) < r˜/3 for any n [ N, N]. Hence, for each i and n [ N, N], if x, y M and d(x, y) < β, then ∈ − ∈ ∈ − ∈ i n n n n + n n + n n d(gi hi(x), gi hi(y)) d(gi hi(x), fi (x)) d( fi (x), fi (y)) d( fi (y), gi hi(y)) ◦ ◦ ≤ ◦ n n n n n ◦ n = d(h + f (x), f (x)) + d( f (x), f (y)) + d( f (y), h + f (y)) i n ◦ i i i i i i n ◦ i r˜/3 + r˜/3 + r˜/3 = r˜. ≤ It follows from Proposition 7.5 that d(hi(x), hi(y)) < α. This fact proves that (hi)i Z is an equicon- = n n∈ tinuous family. Note that if hi(x) hi(y) for some x, y Mi, then d(gi hi(x), gi hi(y)) < r˜ for Z = ∈ ◦ ◦ 1 any n . Thus x y and therefore hi is injective. Analogously we can prove that (hi− )i Z is equicontinuous.∈ Consequently, 2( ) is uniformly structurally stable in 2( ). ∈  Ab M D M It follows directly from Theorems 5.5 and 7.6 that if = ( fi)i Z is an Anosov family satisfying the property of angles consisting of a finite sequence ofF diffeomorphisms,∈ there exists ξ > 0 such that, if B2( ,ξ), then 2( ) and is uniformly conjugate to . G ∈ F G∈A M F References [1] P. Arnoux and A. M. Fisher. “Anosov families, renormalization and non-stationary subshifts”. Ergodic Theory and Dynamical Systems, 25(3): 661-709, 2005. [2] A. Castro, F. Rodrigues and P. Varandas. “Stability and limit theorems for sequences of uniformly hyperbolic dynamics”. arXiv preprint arXiv:1709.01652, (2017). [3] C. Kawan. “Entropy of nonautonomous dynamical systems”. arXiv preprint arXiv:1708.00815. (2017). [4] C. Kawan and Y. Latushkin. “Some results on the entropy of nonautonomous dynamical systems”. Dynamical systems. 31(3): (2016), 251-279. [5] T. Kato. Perturbation theory for linear operators. Vol. 132. Springer Science & Business Media, 2013. [6] Kolyada, Sergiy and Lubom´ır Snoha. “Topologial entropy of nonautonomousdynamical systems”. Random Com- put. Dynamics 4, No. 2-3: 205-233, 1996. STRUCTURAL STABILITY AND A CHARACTERIZATION OF ANOSOV FAMILIES 19

[7] S. Kolyada, M. Misiurewicz, L. Snoha. Topological entropy of nonautonomous piecewise monotone dynamical systems on the interval. Fund. Math. 160 (1999), 161-181. [8] Pei-Dong Liu. “Random perturbations of Axiom A basic sets”. Journal of statistical physics, 90(1-2): 467-490, 1998. [9] Pei-Dong Liu and Min Qian. Smooth Ergodic Theory of Random Dynamical Systems. (Lecture Notes in Mathe- matics 1606). Springer, 2006. [10] J. Mather. “Anosov Diffeomorphism”, appendix to part I of: S. Smale, Differentiable Dynamical Systems, Bul- letin of the Amer. Math. Soc., 73, (1967) 792-795. [11] J. Mather. “Characterization of Anosov diffeomorphisms”. Indag. Math 30.5 (1968): 479-483. [12] Acevedo, Jeovanny de Jesus Muentes. “Local stable and unstable manifolds for Anosov families.” Hokkaido Mathematical Journal 48.3 (2019): 513-535. [13] J. Muentes. “On the continuity of the topological entropy of non-autonomous dynamical systems”. Bulletin of the Brazilian Mathematical Society, New Series, 49(1): 89-106, 2017. [14] J. Muentes. “Openness of Anosov families”. Journal of the Korean Mathematical Society, 55(3): 575-591, 2018. [15] Michael Shub. Global stability of Dynamical Systems. Springer-Verlag, (1987). [16] R. Swan. “Vector bundles and projective modules”. Transactions of the American Mathematical Society, 105, 264- 277 (1962). [17] Marcelo Viana. Lectures on Lyapunov exponents. Vol. 145. Cambridge University Press, 2014. [18] Lai-Sang Young. “Stochastic stability of hyperbolic attractors”. Ergod. Th. and Dynam. Sys, 6(2): 311-319, 1986.

Grupo de Investigacion´ Deartica,Universidad del Sinu´,El´ıas Bechara Zainum´ , Cartagena, Colombia Email address: [email protected]