We introduce generalized hyperbolicity for nonlinear dynamics in Banach spaces. The definition allows the stable/unstable splitting to be discontinuous and requires only inclusions, rather than equalities, in the invariance conditions for both subspaces. On smooth compact manifolds, generalized hyperbolicity is equivalent to Axiom A and the strong transversality condition, providing a finite-dimensional calibration of the proposed Banach-space theory. For C^1-diffeomorphisms of the whole Banach space such that Df and D(f^{-1}) are globally bounded and Df is uniformly continuous, we establish the principal dynamical consequences of generalized hyperbolicity: Lipschitz shadowing, density of periodic points in the chain-recurrent set, and robustness under perturbations small in the uniform C^1 distance. The shadowing result requires no continuity of the splitting, and shadowing trajectories need not be unique. Under the additional assumption that the splitting is uniformly continuous, we prove semi-structural stability. If, in addition, the subspace E^s_x\oplus (Df(x))^{-1}E^u_{f(x)} is independent of x, we obtain structural stability.
We say that a linear isomorphism T:\mathbb{B}\to\mathbb{B} is generalized hyperbolic if there exists a splitting \mathbb{B} = E^s \oplus E^u into two closed subspaces such that for some k>0 conditions (?) hold and
We say that a diffeomorphism f:\mathbb{B} \to \mathbb{B} satisfies generalized hyperbolicity if there exist C > 0, \lambda \in (0, 1) and a family of projections P_x, Q_x: \mathbb{B} \to \mathbb{B} satisfying the following. Denote E^s_x = P_x(\mathbb{B}), E^u_x = Q_x(\mathbb{B}).
(GH1) Splitting. For all x \in \mathbb{B}, P_x + Q_x = Id, \|P_x\|, \|Q_x\| \leq C. In particular E^s_x \oplus E^u_x = \mathbb{B}.
(GH2) Inclusion. The subspaces E^s_x and E^u_x are invariant under the forward and backward iterations of Df, respectively:
(GH3) Exponential estimates. For any x \in \mathbb{B}, n > 0
For a C^1-diffeomorphism of a smooth compact manifold, the tangent-bundle analogue of generalized hyperbolicity is equivalent to Axiom A and the strong transversality condition.
Assume first that f satisfies Axiom A and the strong transversality condition. By Proposition 59, every derivative sequence admits a (C_0,\lambda_0)-structure, with constants independent of the trajectory.
On \Omega(f), use the ordinary hyperbolic splitting TM|_{\Omega(f)}=E_\Omega^s\oplus E_\Omega^u, whose exponential estimates and projection bounds are uniform; denote corresponding constants by C_\Omega>0 and \lambda_\Omega\in(0,1).
The invariant set M\setminus\Omega(f) contains no periodic points. Choose one representative p from each orbit in this set and write a chosen (C_0,\lambda_0)-structure along its derivative sequence as
Define E_{f^k(p)}^s:=E_k^s, E_{f^k(p)}^u:=E_k^u. These definitions are unambiguous because the orbit is nonperiodic. On \Omega(f), set E_x^s=E_{\Omega,x}^s, E_x^u=E_{\Omega,x}^u. With C=\max\{C_0,C_\Omega\}, \lambda=\max\{\lambda_0,\lambda_\Omega\}, the associated projections satisfy (GH1)–(GH3) on M. No continuity of the projections is required.
Conversely, assume that f satisfies generalized hyperbolicity. Condition (GH3) gives E_p^s\subseteq B^+(p), E_p^u\subseteq B^-(p). Therefore,
and hence (58) holds. By Theorem 58, f is structurally stable. The C^1 stability theorem [Mane1987] then implies Axiom A and the strong transversality condition.
Let f, h \in \mathfrak{D}(\mathbb{B}) and assume that f satisfies generalized hyperbolicity with constants C>0, \lambda \in (0,1) and splitting \mathbb{B} = E^s_x \oplus E^u_x. Then:
(P1) \mathfrak{D}(\mathbb{B}) is a group under composition, and g = h \circ f \circ h^{-1} satisfies generalized hyperbolicity with constants R_h^2 C and \lambda, where R_h bounds \|Dh\| and \|Dh^{-1}\|;
(P2) f^{-1} satisfies generalized hyperbolicity with the same constants and with the splitting obtained by interchanging E^s_x and E^u_x;
(P3) for every n \geq 1, f^n satisfies generalized hyperbolicity with constants C, \lambda^n and the same splitting;
(P4) let p \in [1,+\infty] and let f_i \in \mathfrak{D}(\mathbb{B}_i), i \in \mathcal I, satisfy f_i(0) = 0, conditions (?)–(?) with R and r independent of i, and generalized hyperbolicity with C, \lambda independent of i. Then f(\{x_i\}) = \{f_i(x_i)\} belongs to \mathfrak{D}(\mathbb{B}), where \mathbb{B} = \bigl(\bigoplus_{i \in \mathcal I} \mathbb{B}_i\bigr)_{\ell^p}, and satisfies generalized hyperbolicity with constants C, \lambda.
(P1) For f_1, f_2 \in \mathfrak{D}(\mathbb{B}) the chain rule gives \|D(f_1 \circ f_2)(x)\| \leq R_1R_2 and \|D(f_1 \circ f_2)(x+v) - D(f_1 \circ f_2)(x)\| \leq R_1 r_2(|v|) + R_2 r_1(R_2|v|), so f_1 \circ f_2 \in \mathfrak{D}(\mathbb{B}); in particular g \in \mathfrak{D}(\mathbb{B}). The spaces \tilde E^{s,u}_x = Dh(h^{-1}(x))E^{s,u}_{h^{-1}(x)} satisfy (GH1)–(GH3) for g with constants R_h^2C and \lambda.
(P2) The two inclusions in (GH2) and the two estimates in (GH3) interchange under f \to f^{-1}, E^s_x \leftrightarrow E^u_x.
(P3) Conditions (GH1), (GH2) are preserved by iteration, and |Df^{nm}(x)v^s| \leq C\lambda^{nm}|v^s| = C(\lambda^n)^m|v^s|.
(P4) Since f_i(0) = 0 and \|Df_i\| \leq R, we have |f_i(x_i)| \leq R|x_i|, so f maps \mathbb{B} into \mathbb{B}; the same applies to f^{-1}. Since r is nondecreasing and |v_i| \leq \|v\|,
hence f \in \mathfrak{D}(\mathbb{B}) and Df(x) = \bigoplus_i Df_i(x_i). The projections P_x = \bigoplus_i P^{(i)}_{x_i} and Q_x = \bigoplus_i Q^{(i)}_{x_i} are diagonal, so \|P_x\| = \sup_i \|P^{(i)}_{x_i}\| \leq C, and (GH2), (GH3) hold coordinatewise.
The semiflows of parabolic and delay equations that partly motivate this work belong to a different analytical category: they are typically non-invertible and generated by unbounded, densely defined operators, so (?)–(?) do not apply as stated. Extending generalized hyperbolicity to that setting requires additional ideas and is left for future work; we return to this question in Section ?.
For d>0, we say that a sequence \left\{y_{k} \in \mathbb{B}\right\}_{k \in I} is a d-pseudotrajectory if
If the interval I is not specified, we assume that I = \mathbb{Z}.
We say that a point x \in \mathbb{B} is chain-recurrent for a diffeomorphism f if, for any d>0, there exists a finite d-pseudotrajectory \{y_k\}_{k \in [0, N]} with y_0 = y_N = x. Denote the set of all chain-recurrent points by CR(f). Note that CR(f) is closed.
We say that f has the finite Lipschitz shadowing property (FinLipSh) if there exist M, d_{0} > 0 such that for any d < d_{0}, finite interval I = (a, b), and d-pseudotrajectory \left\{y_{k}\right\}_{k \in I} there exists an exact trajectory \{x_{k}\}_{k \in I} satisfying |x_{k} - y_{k}| \leq M d, k \in I.
We say that f has the Lipschitz shadowing property (LipSh) if there exist M, d_{0} > 0 such that for any d < d_{0} and d-pseudotrajectory \left\{y_{k}\right\}_{k \in \mathbb{Z}} there exists an exact trajectory \left\{x_{k}\right\}_{k \in \mathbb{Z}} satisfying |x_{k} - y_{k}| \leq M d, k \in \mathbb{Z}.
We say that f has the Lipschitz periodic shadowing property (LipPerSh) if there exist M, d_{0} > 0 such that for any d < d_{0} and periodic d-pseudotrajectory \left\{y_{k}\right\}_{k \in \mathbb{Z}} with y_{k+N} = y_k (for some N > 0) there exists a periodic exact trajectory \left\{x_{k}\right\}_{k \in \mathbb{Z}} satisfying
Let f\in\mathfrak{D}(\mathbb{B}) satisfy generalized hyperbolicity. Then:
f has the Lipschitz shadowing property;
f has the Lipschitz periodic shadowing property;
periodic points are dense in the chain-recurrent set: \overline{Per(f)}=CR(f);
generalized hyperbolicity is robust in the uniform C^1 distance: if f satisfies generalized hyperbolicity with constants C>0 and \lambda\in(0,1), then, for every \lambda_1\in(\lambda,1), there exist \delta>0 and C_1>0 such that every g\in\mathfrak{D}(\mathbb{B}) satisfying \|g-f\|_{C^1}<\delta is generalized hyperbolic with constants C_1 and \lambda_1.
Let M, d_0 > 0 be the constants from the Lipschitz periodic shadowing property in Theorem 11 (?). Fix x \in CR(f). For any d \in (0, d_0) consider a finite pseudotrajectory y_{k \in [0, N]} such that y_0 = y_N = x. This segment can be continued to an N-periodic pseudotrajectory
By Theorem 11 (?) there exists an N-periodic exact trajectory x_k^{(d)} that Md-shadows \{y_k\}; in particular |x - x_0^{(d)}| = |y_0 - x_0^{(d)}| \leq Md. Since x_0^{(d)} \in Per(f) it follows that \overline{Per(f)} = CR(f).
The proof of Theorem 11 (?) combines Theorem 11 (?)–(?) with the robustness of generalized hyperbolicity for sequences of operators (Lemma 35).
Indeed, consider a diffeomorphism f:\mathbb{B} \to \mathbb{B} satisfying generalized hyperbolicity with constants C, \lambda. Fix \lambda_1 \in (\lambda, 1). For any trajectory \{x_k\}, the sequence of linear operators \{Df(x_k)\} satisfies generalized hyperbolicity. Take M, d_0 > 0 as in Theorem 11 (?)–(?) and \varepsilon > 0, C_1 > C provided by Lemma 35. Choose \delta \in (0, \min(d_0, \varepsilon/2)) such that
Let g\in\mathfrak{D}(\mathbb{B}) satisfy
Consider the set \mathcal{O} of all trajectories of the diffeomorphism g. For any trajectory \omega \in \mathcal{O} choose a point y_{\omega} \in \omega.
We now construct spaces E_y^{s, u} for all points y \in \omega. Consider an exact trajectory y_{k\in \mathbb{Z}} of diffeomorphism g with y_0 = y_{\omega}. Note that |y_{k+1} - f(y_k)| = |g(y_k) - f(y_k)| \leq \delta, k \in \mathbb{Z}, hence y_k is a \delta-pseudotrajectory of the diffeomorphism f. Denote B_k = Dg(y_k). Consider two cases.
Case 1. If the sequence \{y_{k}\} is not periodic, then the points \{y_k\} are pairwise distinct. By Theorem 11 (?) there exists an exact trajectory \{x_k\} of diffeomorphism f that M\delta-shadows \{y_k\}:
Denote A_k = Df(x_k). According to (?), (?) the following holds
By item (Robust1) in Lemma 35, the sequence \{B_k\} satisfies generalized hyperbolicity with constants C_1, \lambda_1.
Case 2. Suppose that the sequence \{y_{k}\} is m-periodic. Then the sequence \{B_k\} is m-periodic as well. By Theorem 11 (?), there exists an m-periodic exact trajectory \{x_k\} satisfying (?). The sequence \{A_k = Df(x_k)\} is m-periodic and satisfies (?). By item (Robust2) in Lemma 35, the sequence \{B_k\} satisfies m-periodic generalized hyperbolicity with constants C_1, \lambda_1.
In both cases denote the spaces corresponding to generalized hyperbolicity for \{B_k\} by \tilde{E}_k^s, \tilde{E}_k^u. Define (slightly abusing notation) for point y_k the spaces
If \omega is an m-periodic trajectory, then \tilde{E}_k^{s, u} = \tilde{E}_{k+m}^{s, u}, and hence \tilde{E}_{y_k}^{s, u} is well-defined. Moreover, Definition 2 does not impose any condition on the dependence of E_y^{s, u} and the corresponding projections on the point y. Hence the splitting defined by (?) satisfies the conditions of generalized hyperbolicity in Definition 2.
We say that f \in \mathfrak{D}(\mathbb{B}) is structurally stable if for every \varepsilon > 0 there exists d_0 = d_0(\varepsilon) > 0 such that for any diffeomorphism g \in \mathfrak{D}(\mathbb{B}) with \|f-g\|_{C^1} < d_0 there exists a continuous map h_1: \mathbb{B} \to \mathbb{B} satisfying \sup_{x \in \mathbb{B}}|h_1(x)| < \varepsilon such that (Id+h_1) is a homeomorphism and g \circ (Id+h_1) = (Id+ h_1) \circ f. If, moreover, the map h_1 can be chosen so that Id+h_1 and (Id+h_1)^{-1} are uniformly continuous, we say that f is uniformly structurally stable.
Every diffeomorphism f \in \mathfrak{D}(\mathbb{B}) satisfying generalized hyperbolicity is structurally stable.
We say that f \in \mathfrak{D}(\mathbb{B}) satisfies uniformly continuous generalized hyperbolicity if it satisfies generalized hyperbolicity with a family of projections \{P_x, Q_x\} for which the maps x \to P_x, x \to Q_x are in addition uniformly continuous, i.e.
We say that f \in \mathfrak{D}(\mathbb{B}) is uniformly semi-structurally stable if for every \varepsilon > 0 there exists d_0 = d_0(\varepsilon) > 0 such that for any diffeomorphism g \in \mathfrak{D}(\mathbb{B}) with \|f-g\|_{C^1} < d_0 there exist uniformly continuous maps h_1, h_2: \mathbb{B} \to \mathbb{B} satisfying \sup_{x \in \mathbb{B}}|h_1(x)|, \; \sup_{x \in \mathbb{B}}|h_2(x)| < \varepsilon, such that
If f \in \mathfrak{D}(\mathbb{B}) satisfies uniformly continuous generalized hyperbolicity, then there exist M>0, d_0 > 0 such that for any d \in (0, d_0) and diffeomorphism g \in \mathfrak{D}(\mathbb{B}) with \|f-g\|_{C^1} < d the following holds.
There exist uniformly continuous maps h_1, h_2:\mathbb{B} \to \mathbb{B} satisfying
hence f is uniformly semi-structurally stable.
If additionally the spaces E^s_x \oplus (Df(x))^{-1}E^u_{f(x)} do not depend on x, then h_1, h_2 can be chosen such that
In that case, the maps (Id+h_1), (Id+h_2) are homeomorphisms, and hence f is uniformly structurally stable.
Assume that f satisfies uniformly continuous generalized hyperbolicity with constants C > 0, \lambda \in (0, 1). Then the pair (f, Df) satisfies uniformly continuous cocycle generalized hyperbolicity. Let \varepsilon_0 be as in Lemma 52. Choose \varepsilon_1 \in (0,\varepsilon_0) and \lambda_1\in(\lambda,1), and let C_1>1 and \delta>0 be the corresponding constants. Let R > 1 be the constant in (?), L = L(C_1, \lambda_1)>0 be the constant provided by Lemma 49. Set \varepsilon=\frac{1}{3L}. Choose d_0 < \min(\frac{\delta}{R}, \frac{1}{3L}), such that (?) holds with this \varepsilon for |v_1|, |v_2| < 2Ld_0. Fix d\in(0,d_0) and g\in\mathfrak{D}(\mathbb{B}) satisfying \|g-f\|_{C^1}<d. Denote \Delta = g - f. Then
Moreover, for every x\in\mathbb{B}, setting y=g^{-1}(x), we have
Together with \sup_x|f(x)-g(x)|<d<\delta, this verifies (52).
By Lemma 52, the pair (g, Df) satisfies uniformly continuous cocycle generalized hyperbolicity with constants C_1, \lambda_1.
Denote the splitting corresponding to the pair (f, Df) by
and splitting provided by Lemma 52 corresponding to pair (g, Df) by
We do not claim that the pair (g, Dg) satisfies uniformly continuous cocycle generalized hyperbolicity, although we believe it to be true.
According to (52) the following equality holds \tilde{E}^s_x \oplus \tilde{F}_x = {E}^s_x \oplus {F}_x. Hence
By Lemma 49 there exist operators O_f, O_g: X \to Y satisfying
Denote M = 2L.
Step 1. First, we find a map h_1: \mathbb{B} \to \mathbb{B}, h_1 \in Y, satisfying (?). Substituting g = f + \Delta in equation (?) and using the definition of s_f we get
Canceling f(x) and regrouping terms we get
In order for (?) to hold, it suffices that
Consider the (nonlinear) operator G_f:BUC(\mathbb{B}, \mathbb{B}) \to BUC(\mathbb{B}, \mathbb{B}) defined by
Using (?) and (Proof) we conclude that for |h_1|, |h'_1| \leq 2L d the following inequalities hold
Using (49) we conclude that \|O_f \circ G_f(h_1) - O_f \circ G_f(h'_1)\| \leq \frac{2}{3}\|h_1-h'_1\|, hence O_f \circ G_f is a contraction. For \|h_1\| \leq 2Ld, equations (?), (Proof) imply \|G_f(h_1)\| \leq \varepsilon 2Ld + d, and
The latter implies existence and uniqueness of a map h_1 with \|h_1\| \leq 2Ld satisfying O_f \circ G_f (h_1) = h_1, and hence (?). Therefore h_1 satisfies (?). Note that \mbox{Im }O_f = Y and hence h_1 \in Y.
Step 2. To construct a map h_2 \in Y, satisfying (?) we follow an approach similar to that of Step 1. For completeness, we provide the full proof below. The conjugacy (?) is equivalent to
Using definition of s_f and canceling f(x) we get
In order for (?) to hold, it suffices that
Consider the (nonlinear) operator G_g:BUC(\mathbb{B}, \mathbb{B}) \to BUC(\mathbb{B}, \mathbb{B}) defined by
Using (?) and (Proof) we conclude that for |h_2|, |h'_2| \leq 2L d the following inequalities hold
Using (49) we conclude that \|O_g \circ G_g(h_2) - O_g \circ G_g(h'_2)\| \leq \frac{1}{3}\|h_2-h'_2\| and hence O_g \circ G_g is a contraction. If \|h_2\| \leq 2Ld then (?) and (Proof) imply \|G_g(h_2)\| \leq \varepsilon 2Ld + d, and
The latter implies existence and uniqueness of a map h_2 with \|h_2\| \leq 2Ld satisfying O_g \circ G_g (h_2) = h_2, and hence (?). Therefore, h_2 satisfies (?) and (?). Note that \mbox{Im }O_g = Y, hence h_2 \in Y. This proves Theorem 16 (?).
Step 3. Under the additional assumption in Theorem 16 (?), we prove (?). From (?) and (?) we obtain the following.
Since h_1, h_2 \in Y and the spaces E^s_x \oplus (Df(x))^{-1}E^u_{f(x)} = E^s_x \oplus F_x do not depend on x, for some functions h_3, h_4 \in Y the following hold
By (?) the identity f\circ (Id+h_3) = (Id+h_3) \circ f holds. Arguing as in Step 1, we conclude that
Since h_3 \in Y, Lemma 49 implies that
Similarly to the operator O_f \circ G_f, the map h_3 \to O_f(s_f(x, h_3(x))) is a contraction (decreasing d if necessary) on the set of h_3 with \|h_3\| \leq 2Md. Therefore, equation (?) has a unique solution for h_3 satisfying (?). On the other hand, h_3 \equiv 0 satisfies both (?) and (?). By uniqueness h_3 \equiv 0.
The argument for h_4 is analogous. Equation (?) implies that g \circ (Id+h_4) = (Id+h_4) \circ g. Arguing as in Step 2, we obtain
As in the proof for h_3, this equation has a unique solution h_4, satisfying (?). Hence, h_4 \equiv 0. This completes the proof of (?) and Theorem 16 (?).
The additional condition in Theorem 16 (?) is automatic for an invariant hyperbolic splitting. Indeed, if Df(x)E_x^u=E_{f(x)}^u, then (Df(x))^{-1}E_{f(x)}^u=E_x^u, and consequently E_x^s\oplus(Df(x))^{-1}E_{f(x)}^u = E_x^s\oplus E_x^u = \mathbb{B}.
In [LaniWayda1995] strong results on shadowing and stability (persistence in their terminology) were obtained. At the same time, their proof strongly relies on an expansivity argument, which is not available in our setting. For other cases in which uniqueness or expansivity simplifies the analysis of the interplay between shadowing, structural stability, and hyperbolicity or exponential dichotomy, see, for example, [Henry94].
Let \lambda \in (0, 1), R > 1/\lambda, M > 1 and diffeomorphisms a_k: \mathbb{R} \to \mathbb{R} satisfy
Let \mathbb{B} = \ell^p(\mathbb{Z}) with p \in [1, +\infty] and consider the map f: \mathbb{B} \to \mathbb{B} defined as
Then f is a diffeomorphism satisfying generalized hyperbolicity.
By the definition of f, the inequalities |y_{k+1}| \leq R |x_k| hold, hence \|f(\{x_k\})\| \leq R \|\{x_k\}\|, and f is well-defined. Moreover, f^{-1} is defined by f^{-1}(\{y_k\}) = \{x_k\}, where x_k = a_k^{-1}(y_{k+1}), with \|f^{-1}(\{y_k\})\| \leq R\|\{y_k\}\|; hence \{x_k\} is well-defined.
For \{x_k\} \in \mathbb{B} define a linear operator A(\{x_k\}):\mathbb{B} \to \mathbb{B} by
It is invertible, with A^{-1}(\{x_k\})\{w_k\} = \{v_k\}, where v_k = w_{k+1}/a'_k(x_k), and by (19) we have \|A(\{x_k\})\|, \|A^{-1}(\{x_k\})\| \leq R. Let us show that Df(\{x_k\}) = A(\{x_k\}), where Df is a Fréchet derivative. Indeed,
According to (19), we have |z_k| \leq M|v_k|^2 and hence inequality (?) holds with r(|v|) = M|v|. Similarly, equation (?) holds and r satisfies (?).
Let us show that f satisfies generalized hyperbolicity. Denote
Note that E^s \oplus E^u = \mathbb{B}. Let P_x, Q_x be the projections onto E^s and E^u respectively (independent of x). Then E^s_x = P_x(\mathbb{B}) = E^s, E^u_x = Q_x(\mathbb{B}) = E^u, and \|P_x\| = \|Q_x\| = 1, so condition (GH1) holds with C \geq 1. By construction, A(x)E^s \subset E^s and A^{-1}(x)E^u \subset E^u, hence condition (GH2) holds. By (19) we have \|A(x)|_{E^s}\|, \|A^{-1}(x)|_{E^u}\| \leq \lambda, and hence condition (GH3) holds with C = 1 and the same \lambda as above.
Consider f:\mathbb{B} \to \mathbb{B} defined by f(\{x_k\}) = \{a(x_k)\}. Then f \in \mathfrak{D}(\mathbb{B}) and satisfies generalized hyperbolicity.
By (?), the map a satisfies (?) with R = 1/\lambda_1 and (?) with r(t) = Mt, and this r satisfies (?); hence a \in \mathfrak{D}(\mathbb{R}). As explained above, a satisfies generalized hyperbolicity with constants C = (\lambda_1\lambda_2)^{-n_0}, \lambda = \lambda_2 and the splitting \mathbb{R} = G^s_x \oplus G^u_x. Since a(0) = 0 and
Proposition 4 (P4), applied with \mathbb{B}_k = \mathbb{R} and f_k = a for every k \in \mathbb{Z}, shows that f \in \mathfrak{D}(\mathbb{B}), that Df(\{x_k\})\{v_k\} = \{a'(x_k)v_k\}, and that f satisfies generalized hyperbolicity with the same constants C and \lambda, with respect to the splitting
and the corresponding projections P_x, Q_x.
Examples 19 and 20 isolate, respectively, strict invariant inclusions and a discontinuous, dimension-varying splitting. The examples are not covered by the Banach-space frameworks of [SteinleinWalther1990, LaniWayda1995, Henry1981, SellYou]. That the examples are rigid is therefore the point rather than a limitation: they are minimal witnesses. Moreover, by Theorem 11 (?) and Proposition 4, each generates a conjugacy-closed family that is open in the uniform C^1 distance within \mathfrak{D}(\mathbb{B}), so the phenomena they exhibit are stable, not exceptional.
Let \mathbb{B} = L^p(\mathbb{S}^1, \mathbb{R}) with p \in [1, +\infty], or \mathbb{B} = C^0(\mathbb{S}^1, \mathbb{R}). Define a linear isomorphism T:\mathbb{B} \to \mathbb{B}, as
Then T is generalized hyperbolic.
By (?), there exist \lambda \in (0, 1) and interval neighborhoods U_s of s and U_u of u such that
Case 1. \mathbb{B} = L^p(\mathbb{S}^1, \mathbb{R}) with p \in [1, +\infty]. Set b_p(x):=a(x)|g'(x)|^{1/p}, and set b_\infty:=a. Since s and u are hyperbolic attracting and repelling fixed points, respectively, b_p(s)<1, b_p(u)>1. For p<+\infty, a change of variables gives
so b_p is the effective weight for the L^p estimates. Decreasing the neighborhoods U_s,U_u, while preserving the inclusions in (Proof), (Proof), we can assume that there exists \lambda_p\in(0,1) such that
Let \xi_1,\xi_2 be the boundary points of U_u, set V=\mathbb S^1\setminus U_u, and choose N>1 such that g^N(V)\subset U_s. Define the projections P, Q: \mathbb{B} \to \mathbb{B} by
and set E^s := P(\mathbb{B}), E^u := Q(\mathbb{B}). Note that
Hence TE^s \subset E^s, T^{-1}E^u \subset E^u, P+Q = Id. For a suitable global bound R_p\geq1 on b_p and b_p^{-1}, set C_p=(R_p/\lambda_p)^N. The choice of U_s,U_u,V,N and (?) imply that
and T is generalized hyperbolic with the splitting \mathbb{B} = E^s \oplus E^u.
Case 2. \mathbb{B} = C^0(\mathbb{S}^1, \mathbb{R}). For this case, take p=+\infty in the preceding choice, so that b_\infty=a. Consider the continuous functions \Xi_1:[\xi_1, g(\xi_1)] \to \mathbb{R}, \Xi_2:[\xi_2, g(\xi_2)] \to \mathbb{R}, such that
Define the projection P:\mathbb{B} \to \mathbb{B} as follows
By (?) if v \in \mathbb{B}, then Pv \in \mathbb{B}. Define the projection Q:\mathbb{B} \to \mathbb{B} by Qv = v - Pv. Set E^s = P(\mathbb{B}), E^u = Q(\mathbb{B}). Note that
if v^s \in E^s then \operatorname{supp} v^s \subset V, if v^u \in E^u then \operatorname{supp} v^u \subset \operatorname{Cl}(\mathbb{S}^1 \setminus g(V));
if \operatorname{supp} v^s \subset g(V) then v^s \in E^s, if \operatorname{supp} v^u \subset U_u = \mathbb{S}^1 \setminus V then v^u \in E^u.
Hence spaces E^s, E^u satisfy inclusions (1). The same argument as in Case 1, with p=+\infty, proves (?), (?). Therefore, T is generalized hyperbolic.
Let f \in \mathfrak{D}(\mathbb{B}) satisfy generalized hyperbolicity. Let X = B(\mathbb{B}, \mathbb{B}) be the Banach space of bounded maps \mathbb{B} \to \mathbb{B} with the supremum norm. Consider the linear isomorphism T_f: X \to X defined as follows
Then T_f is generalized hyperbolic.
Inequalities (?) imply that T_f is a bounded linear isomorphism. Let C, \lambda be the corresponding constants, and let P_x, Q_x: \mathbb{B} \to \mathbb{B} be the family of projections associated with the generalized hyperbolicity of f. Define the projections P, Q: X \to X by the following
Condition (GH1) implies that for any v \in X we have Pv, Qv \in X and |Pv|, |Qv| \leq C|v|. Set E^s = P(X), E^u = Q(X). Condition (GH2) implies inclusions (1). Condition (GH3) implies
Therefore T_f is generalized hyperbolic.
If h satisfies Axiom A and the strong transversality condition, then S_h is generalized hyperbolic as a linear operator on \mathbb{B} = L^p(M, TM) for p \in [1, + \infty], and also on \mathbb{B} = C^0_b(M, TM).
We say that a sequence of operators \mathcal{A} = \{A_{k\in I}:\mathbb{B} \to \mathbb{B}\} (I as in (?)) with uniformly bounded norms \|A_k\|, \|A_k^{-1}\| \leq R satisfies the bounded solution property on interval I with constant L>0 if for any sequence \{w_{k+1} \in \mathbb{B}\}_{k \in \tilde{I}} with |w_{k+1}| \leq 1 there exists a sequence \{v_k \in \mathbb{B}\}_{k \in I} with |v_k| \leq L such that
If I = \mathbb{Z} we say that \mathcal{A} satisfies the bounded solution property with constant L>0.
If there exists a bounded linear right inverse S_{\mathcal{A}} of the operator T_{\mathcal{A}} defined by (?) with I defined in (?) and \|S_{\mathcal{A}}\| \leq L, we say that \mathcal{A} has the strong bounded solution property on interval I with constant L > 0.
We say that the sequence \mathcal{A} admits an exponential dichotomy on I, where I \in \{\mathbb{Z}^+, \mathbb{Z}^-, \mathbb{Z}\}, if there exist constants C>0, \lambda\in(0,1) and families of bounded projections P_k,Q_k:\mathbb{B} \to \mathbb{B} (k\in\mathbb Z), satisfying the following properties. Denote E^s_k:=P_k(\mathbb{B}), E^u_k:=Q_k(\mathbb{B}).
(ED1) Splitting. P_k+Q_k=Id, \|P_k\|,\|Q_k\|\le C, hence E^s_k\oplus E^u_k = \mathbb{B}.
(ED2) Invariance. The splitting is carried exactly onto itself: A_kE^s_k=E^s_{k+1}, A_k^{-1}E^u_{k+1}=E^u_{k} for k\in I. Equivalently A_kP_k=P_{k+1}A_k and A_kQ_k=Q_{k+1}A_k.
(ED3) Exponential estimates. For n \geq 0 the following estimates hold
Assume that the restrictions of the sequence \mathcal{A} to the future and to the past, {\mathcal{A}}^{+}:=\{A_k\}_{k\ge 0}, {\mathcal{A}}^{-}:=\{A_k\}_{k\le 0}, each admit an exponential dichotomy with splittings E^{s,+}_k\oplus E^{u,+}_k=\mathbb{B} and E^{s,-}_k\oplus E^{u,-}_k=\mathbb{B}, respectively. We say that \mathcal{A} satisfies the transversality condition if E^{s,+}_0\ + E^{u,-}_0 = \mathbb{B}.
Let \mathcal{A} = \{A_k: \mathbb{B} \to \mathbb{B}\}_{k \in \mathbb{Z}} be a sequence of isomorphisms satisfying \|A_k\|, \|A^{-1}_k\| \leq R for some R>0. We say that \mathcal{A} satisfies generalized hyperbolicity if there exist C > 0, \lambda \in (0, 1) and families of projections P_k, Q_k: \mathbb{B} \to \mathbb{B} satisfying the following properties. Denote E^s_k = P_k(\mathbb{B}), E^u_k = Q_k(\mathbb{B}). For every k \in \mathbb{Z} the following holds
(GH1*) Splitting. P_k + Q_k = Id, \|P_k\|, \|Q_k\| \leq C. Note that E^s_k \oplus E^u_k = \mathbb{B}.
(GH2*) Inclusion. The spaces E^s_k and E^u_k are forward- and backward-invariant, respectively(Compare with condition (ED2), the only difference is inclusion instead of equality.)
(GH3*) Exponential estimates. For every n\ge 0
For a periodic sequence of operators with period m (i.e. A_{k+m} = A_k), we say that \mathcal{A} satisfies periodic generalized hyperbolicity if projections P_k, Q_k are also m-periodic, that is, P_{k+m} = P_k, Q_{k+m} = Q_k.
Let f \in \mathfrak{D}(\mathbb{B}), C > 0, \lambda \in (0,1). Then f satisfies generalized hyperbolicity with constants C, \lambda if and only if
for every nonperiodic trajectory \{x_k\}_{k \in \mathbb{Z}} of f, the sequence \{Df(x_k)\}_{k \in \mathbb{Z}} satisfies generalized hyperbolicity with constants C, \lambda;
for every m-periodic trajectory \{x_k\}_{k \in \mathbb{Z}} of f, the sequence \{Df(x_k)\}_{k \in \mathbb{Z}} satisfies m-periodic generalized hyperbolicity with constants C, \lambda.
If f satisfies generalized hyperbolicity, set E^{s,u}_k := E^{s,u}_{x_k}. Then (GH1*)–(GH3*) are (GH1)–(GH3) at the points x_k, and if x_{k+m} = x_k the splitting is automatically m-periodic.
Conversely, the trajectories of f partition \mathbb{B}. Choose a point y_\omega in each trajectory \omega and let \{E^s_k, E^u_k\} be a splitting as in the hypothesis for the sequence \{Df(f^k(y_\omega))\}. The definition E^{s,u}_{f^k(y_\omega)} := E^{s,u}_k is unambiguous: for a nonperiodic \omega the points f^k(y_\omega) are pairwise distinct, and for an m-periodic \omega this is exactly the m-periodicity of the splitting. Conditions (GH1)–(GH3) then hold with constants C, \lambda.
If f \in \mathfrak{D}(\mathbb{B}) satisfies generalized hyperbolicity, then there exists L>0 such that, for any exact trajectory \{x_k\} and interval I (finite or infinite), the associated sequence of linear operators \mathcal{A} = \{A_k = Df(x_k)\}_{k \in I} satisfies the strong bounded solution property on I with constant L.
For an interval I of the form (?) the operator S_{\mathcal{A}} can be defined explicitly: S_{\mathcal{A}}\{w_{k+1}\}_{\tilde{I}} = \{v_k\}_{k \in I}, where v_k is given by Perron sums
with the convention that w_i = 0 for i \notin I. By construction the operator S_{\mathcal{A}} is linear. Due to (GH1), (GH3)
and hence \|S_{\mathcal{A}}\| \leq C^2 \frac{1+\lambda}{1-\lambda}. Conditions (GH1), (GH2) imply that \{v_k\} satisfies (25) and hence T_{\mathcal{A}} \circ S_{\mathcal{A}} \{w_{k+1}\}_{k \in \tilde{I}} = \{w_{k+1}\}_{k \in \tilde{I}}.
For any L > 0 there exists \varepsilon > 0 such that the following holds. Let I be an interval (finite or infinite), and let \mathcal{A} = \{A_k\}_{k \in I} be a sequence satisfying the strong bounded solution property on I with constant L. If \mathcal{B} = \{B_k\} is another sequence of uniformly bounded isomorphisms satisfying \|B_k - A_k\| \leq \varepsilon, then it satisfies the bounded solution property on I with constant 2L.
Consider \varepsilon < \frac{1}{2L}, operator T_{\mathcal{A}} and its right inverse S_{\mathcal{A}}:
Let B_k = A_k + \Delta_k, then equation v_{k+1} - B_kv_k = w_{k+1}, for k \in \tilde{I} is equivalent to
Consider the operator \Delta: \ell^{\infty}(I, \mathbb{B}) \to \ell^{\infty}(\tilde{I}, \mathbb{B}) defined by \Delta \{v_k\}_{k \in I} = \{w'_{k+1} = \Delta_k v_k\}_{k \in \tilde{I}}. Note that \| \Delta \| \leq \varepsilon. Then any solution of the following equation is also a solution of (Proof)
Note that
Then the sequence
is a solution of (Proof). Due to (Proof) the series converges and
Let \mathcal A=\{A_k\}_{k\in\mathbb{Z}} be a sequence of linear isomorphisms such that \|A_k\|,\|A_k^{-1}\|\leq R. Assume that there exists L>0 such that for any finite interval I = (a, b) the sequence \{A_k\}_{k \in I} satisfies the bounded solution property on I with constant L. Then the complete sequence \{A_k\}_{k \in \mathbb{Z}} satisfies the bounded solution property on \mathbb{Z} with the constant 2L.
We use the following statement for linear isomorphisms of Banach spaces.
For any invertible continuous linear operator T on any Banach space X, if T has the finite Lipschitz shadowing property with constant L, then T has the Lipschitz shadowing property with constant 2L.
The statement of Theorem 1 in [NilsonPeris] gives only the shadowing property without specifying the constant. However, the proof can be repeated to establish the constant 2L. For completeness of exposition we provide the proof tracing the constant in Appendix A.
Consider the Banach space\mathfrak{X} =\ell^{\infty}(\mathbb{Z}, \mathbb{B})=\bigl\{\{v_n\}_{n\in\mathbb{Z}}: v_n\in \mathbb{B} \bigr\} and a linear operator \mathfrak{A} \colon \mathfrak{X} \to \mathfrak{X} defined by
Let us prove that under the assumptions of Lemma 33, the operator \mathfrak{A} has the finite shadowing property.
Consider a finite interval [k_1, k_2] and sequences \xi^k\in \mathfrak{X}, k \in [k_1, k_2], with |\xi^k|\leq1; denote by \xi_n^k \in \mathbb{B} (k\in[k_1,k_2], n\in\mathbb{Z}) their elements. Below we show that there exists a sequence \nu^k \in \mathfrak{X}, k\in[k_1,k_2] with |\nu^k| \leq L such that
Denote the elements of \nu^k by \nu^k_n \in \mathbb{B}, i.e. \nu^k = \{\nu^k_n\}. Equations (?) are equivalent to
The idea is to read the doubly-indexed system (?) along the diagonals k-n=j: on each of them it becomes a finite inhomogeneous difference equation in the single index n for the operators \{A_n\}, to which the finite bounded solution property applies. Denote j := k-n and consider the sequences
Then (?) is equivalent to
Rewriting indexes we have
For any j\in\mathbb{Z}, according to (25) there exists a sequence \mu^j_n satisfying (?), and |\mu_n^j| \le L for n\in[k_1-j,k_2-j]. Hence, for all j\in\mathbb{Z} and n+j\in[k_1,k_2-1], (?) holds, which means that \nu_n^k defined by (?) satisfies (?). Hence, the operator \mathfrak{A} has the finite Lipschitz shadowing property with constant L. According to Theorem Proof, this implies that \mathfrak{A} has the Lipschitz shadowing property with constant 2L.
Let us now show that the sequence \mathcal{A} satisfies the bounded solution property for I = \mathbb{Z}. Indeed, fix \{w_k \in \mathbb{B}\} with |w_k| \leq 1. Consider the sequence \xi^k \in \mathfrak{X} defined as
Note that |\xi^k| = |w_{k+1}| \leq 1. Since \mathfrak{A} has the shadowing property, there exists a sequence \nu^k \in \mathfrak{X} with |\nu^k| \leq 2L such that
Denote \nu^k = \{\nu^k_n\}, then |\nu^k_n| \leq 2L. According to (?), the sequence v_k = \nu^k_k satisfies (25) for I = \mathbb{Z} and hence \mathcal{A} satisfies the bounded solution property with constant 2L for I = \mathbb{Z}.
Assume that \mathcal{A} = \{A_k\}_{k \in \mathbb{Z}} is m-periodic (A_{k+m}=A_k, k \in \mathbb{Z}) and satisfies generalized hyperbolicity with constants C, \lambda, with a not necessarily periodic splitting. Then there exists L = L(C, \lambda)>0 such that for every bounded m-periodic sequence w=\{w_k\}_{k\in\mathbb{Z}}\subset\mathbb{B} with |w_k| \leq 1 there exists an m-periodic sequence v=\{v_k\}_{k\in\mathbb{Z}}\subset\mathbb{B}, satisfying
For i\in\mathbb{Z}, let r(i)\in\{0,\ldots,m-1\} be its residue modulo m. Periodicity gives \Phi_A(k+m,\ell+m)=\Phi_A(k,\ell) and, writing i=r(i)+qm, we have
Consider the sequence
Although the splitting need not be periodic, (Proof) allows us to use the projections at the representative indices r(i). By (Proof) and (GH1*), (GH3*),
Hence the series in (?) converge. Summing the geometric tail yields |v_k| \leq L = C^{2}\frac{1+\lambda}{1-\lambda}. Changing the summation index by m and using
gives v_{k+m}=v_k. Moreover, using A_k\Phi_A(k,i)=\Phi_A(k+1,i) and P_{r(k+1)}+Q_{r(k+1)}=Id, direct cancellation in (?) gives v_{k+1}=A_kv_k+w_{k+1}.
Assume that \mathcal{A}=\{A_k\} satisfies generalized hyperbolicity with constants C>0, \lambda\in(0,1) and \|A_k\|,\|A_k^{-1}\|\leq R, for k\in\mathbb{Z}. Then for any \lambda_1 \in (\lambda, 1), there exist C_1 = C_1(C, \lambda, \lambda_1, R) and \varepsilon > 0 such that
(Robust1) any sequence \mathcal{B} = \{B_k\}_{k \in \mathbb{Z}}, where
also satisfies generalized hyperbolicity with constants C_1, \lambda_1;
(Robust2) if, in addition, the sequence \mathcal{A} is m-periodic and satisfies periodic generalized hyperbolicity, then any m-periodic sequence \mathcal{B} = \{B_k\}_{k \in \mathbb{Z}} satisfying (35) also satisfies periodic generalized hyperbolicity with constants C_1, \lambda_1.
Let B_k = A_k + \Delta_k. Denote
where
The operators \Delta_k^{ss}, \Delta_k^{su}, \Delta_k^{us}, and \Delta_k^{uu} are defined in the same way, replacing A_k by \Delta_k.
By (GH2*) we have A_k^{us}=0, and therefore
Denote F_k^u = A_k^{-1} E_{k+1}^u. By (GH2*) we have F_k^u \subseteq E_k^u, and the map A_k:F_k^u \to E_{k+1}^u is a bijection. Therefore, it has an inverse; denote it by Z_k: E_{k+1}^u \to F_k^u,
Let us first prove item (Robust1). Denote L' = 1+ C^2 \frac{\lambda^2}{1-\lambda^2}. Fix \varepsilon<1/(2R) sufficiently small so that, for
the following inequality holds
Since \|A_k^{-1}(B_k-A_k)\|<\frac{1}{2}, each B_k is an isomorphism and \|B_k^{-1}\|\leq 2R.
In the following we show that there exists a sequence of linear maps H_k: E_k^s \to F_k^u, \|H_k\| \leq \varepsilon_2, such that for linear spaces
holds the inclusion
In the standard robustness proof for exponential dichotomy, the latter inclusion is replaced by an equality, and the operators H_k are assumed to act on E_k^u. These differences lead to the corresponding changes in the arguments.
Consider the vectors v_k^s \in E_k^s and \tilde{v}_k^s = v_k^s + H_k v_k^s \in \tilde{E}_k^s. The following holds
Inclusion B_k \tilde{v}_k^s \in \tilde{E}_{k+1}^s is equivalent to equality
If a sequence of operators H_k: E_k^s \to F_k^u satisfies
then inclusions (?) hold.
Consider the following auxiliary problem. Given a sequence of bounded operators S_k: E_k^s \to F_k^u, we seek a bounded sequence of operators H_k: E_k^s \to F_k^u that satisfies
Let X be the Banach space of such operator sequences \{S_k\}_{k \in \mathbb{Z}} endowed with the norm \|\{S_k\}\| = \sup_k \|S_k\|. Consider the operator F:X\to X defined by
Due to condition (GH3*) for l \geq 0, the following inequalities hold
Hence, the right-hand side of (?) converges and
A straightforward substitution shows that H_k, S_k satisfy (?).
Consider the nonlinear operator Q:X \to X defined as Q(\{H_k\}) = \{S_k\}, where
Note that if \{H_k\} \in X is a fixed point of F \circ Q, i.e. \{H_k\} = F \circ Q (\{H_k\}), then \{H_k\} satisfies (?).
Let \varepsilon_2 satisfy the conditions (?). Let U be the closed ball in X of radius \varepsilon_2 centered at 0. In the following, we show that F \circ Q : U \to U and F \circ Q|_{U} is a contraction. Consider H = \{H_k\}, H' = \{H'_k\} \in U, and assume that \|H_k\|, \|H'_k\| \leq \varepsilon_2, \|H-H'\| = \delta. Let us estimate Q\{H_k\} - Q\{H'_k\} = \{S_k - S'_k\}. The following holds
Taking into account (?) and \delta \leq 2\varepsilon_2 we conclude that \|QH - QH'\| \leq \frac{1}{2L'}\|H-H'\|. Since F is linear, inequality (?) implies that
and hence F \circ Q is contracting on U. Note that \|Q\{H_k = 0\}\| = \|\{-Z_k\Delta_k^{us}\}\| \leq R\varepsilon_1, hence \|F \circ Q\{H_k = 0\}\| \leq L'R\varepsilon_1 and according to (?) and (?) for any H \in U we have
and hence F \circ Q(U) \subset U. By the contraction mapping principle, there exists a unique H = \{H_k\}, with |H_k| \leq \varepsilon_2 such that F \circ Q H = H. Consequently, the subspaces \tilde{E}_k^s defined by (?) satisfy (?).
Let us show that for small enough \varepsilon, \varepsilon_2 the spaces \tilde{E}_k^s satisfy
for some C_1 = C_1(C, \lambda, \lambda_1, R). Indeed, let us choose N such that 2C\lambda^N \leq \lambda_1^N and choose C_1 = \left(\frac{R+1}{\lambda_1}\right)^N. For small enough \varepsilon, \varepsilon_2 we have
Represent n > 0 in the form n = lN + r, with l \geq 0, r \in [0, N-1]. Note that
which implies (?).
Similarly, for small enough \varepsilon we construct a sequence of operators H_k^u:E_k^u \to A_{k-1}E_{k-1}^s and spaces \tilde{E}_k^u = \{ v_k^u + H_k^uv_k^u : v_k^u \in E_k^u\} such that B_k^{-1} \tilde{E}_{k+1}^u \subseteq \tilde{E}_{k}^u and
For small enough \varepsilon_2 the inequalities \|H_k\|, \|H_k^u\| \leq \varepsilon_2 imply that the operator
is invertible, and projections P_{\tilde{E}_{k}^{s, u}} = T_k\circ P_{{E}_k^{s, u}} \circ T^{-1}_{k} satisfy \|P_{\tilde{E}_{k}^{s, u}}\| \leq 2C and \tilde{E}_k^s \oplus \tilde{E}_k^u = \mathbb{B}.
Hence, the sequence \mathcal{B} satisfies generalized hyperbolicity with spaces \tilde{E}_k^s, \tilde{E}_k^u and constants C_1, \lambda_1. Item (Robust1) is proved.
To prove item (Robust2), we consider an m-periodic sequence of operators \mathcal{A} = \{A_k\} that satisfies periodic generalized hyperbolicity, with the corresponding periodic splitting \mathbb{B} = E_k^s \oplus E_k^u. Let \mathcal{B} = \{B_k\} be an m-periodic perturbation of \mathcal{A} satisfying (35). We show that the subspaces \tilde{E}_k^s, \tilde{E}_k^u constructed in the proof of (Robust1) are m-periodic.
The sequence of operators H = \{H_k\} \in X satisfying (F \circ Q)H = H can be obtained as a limit of iterates H^l = (F \circ Q)^l\{H_k = 0\} \in X: H = \lim_{l \to \infty}H^l. If sequences \mathcal{A}, \mathcal{B} and projections P_k, Q_k are m-periodic, then the operator (F \circ Q) maps m-periodic sequences to m-periodic sequences. Hence, for any l>0 the iterate H^l is m-periodic, and so is its limit H. Therefore, the subspaces \tilde{E}_k^s constructed in the proof of item (Robust1) are m-periodic. Similar considerations show that \tilde{E}_k^u is m-periodic as well. Consequently, \mathcal{B} satisfies periodic generalized hyperbolicity. This proves item (Robust2).
We expect that for sufficiently small \varepsilon, the sequence \mathcal{B} satisfies generalized hyperbolicity for any prescribed C_1 > C, \lambda_1 \in (\lambda, 1). However, the proof of Lemma 35 presented in this paper does not yield an optimal control of the constant C_1.
Let \mathbb{B} = \ell^p(\mathbb{Z}), p \in [1, \infty]. Consider a sequence of operators \mathcal{A} = \{A_k: \mathbb{B} \to \mathbb{B}\}_{k \in \mathbb{Z}} defined as
Then \mathcal{A} satisfies generalized hyperbolicity, but does not admit an exponential dichotomy on \mathbb{Z}^+ or \mathbb{Z}^-.
Consider the projections
A straightforward check shows that \mathcal{A} satisfies generalized hyperbolicity with projections P_k, Q_k, and C=1, \lambda = 1/2.
Assume that \mathcal{A} admits an exponential dichotomy on \mathbb{Z}^+ and denote its stable subspace at zero by \tilde E^s_0. Then \tilde E^s_0 is closed and
Denote by e_m the standard basis vectors in \ell^p(\mathbb{Z}) and set v^{(N)}=\sum_{m=1}^N2^{-m}e_m. Every v^{(N)} has finite support and |\Phi_A(n,0)v^{(N)}|\to0 as n\to+\infty. Hence v^{(N)}\in\tilde E^s_0. Moreover, v^{(N)}\to v:=\sum_{m=1}^{+\infty}2^{-m}e_m in \ell^p(\mathbb{Z}) for every p\in[1,+\infty]. However, the n-th coordinate of \Phi_A(n,0)v is equal to 1, and therefore |\Phi_A(n,0)v|\geq1 for every n>0. Thus v\notin\tilde E^s_0, contradicting the closedness of \tilde E^s_0. Hence \mathcal{A} does not admit an exponential dichotomy on \mathbb{Z}^+. Similarly, \mathcal{A} does not admit an exponential dichotomy on \mathbb{Z}^-.
If f\in\mathfrak{D}(\mathbb{B}) satisfies generalized hyperbolicity, then f satisfies the finite Lipschitz shadowing property (FinLipSh).
Let M = 2L. Fix d_0 > 0, satisfying the following inequalities
We first prove the statement for intervals of the form [-N,N].
For any d < d_0 and any d-pseudotrajectory \{y_k\}_{k \in [-N, N]} there exists an exact trajectory \{x_k\}_{k \in [-N, N]} such that |y_k - x_k| \leq Md.
We prove the claim by induction on N. For N=0, the statement is immediate.
Consider a d-pseudotrajectory \{y_{k \in [-(N+1), N+1]}\} with d<d_0. By the induction hypothesis, the d-pseudotrajectory \{y_{k \in [-N, N]}\} is Md-shadowed by an exact trajectory \{p_{k \in [-N, N]}\}
Note that p_{N+1} = f(p_N), p_{-(N+1)} = f^{-1}(p_{-N}). Due to (?), (6), (?) the following estimates hold
Hence, we conclude the following.
For d< d_0 a d-pseudotrajectory \{y_{k \in [-(N+1), N+1]}\} can be (RM+R)d-shadowed.
In the following, we show how to construct a better shadowing trajectory. Denote A_k = Df(p_k). Consider the sequence \{w_k \in \mathbb{B}\} defined by
Note that |w_k| \leq 1. By Lemma 31 there exists a sequence \{v_k \in \mathbb{B}\}_{k \in \mathbb{Z}}, |v_k| \leq L such that equalities (25) hold. Consider the sequence z_k = y_k + dv_k. Note that
Inequalities (?), (?), (?), (?) imply that
Then for k \in [-(N+1), N] the following inequalities hold
Hence \{z_{k \in [-(N+1), N+1]}\} is a 2 (RM+R+L)d \cdot r\left((RM+R+L)d\right)-pseudotrajectory. Inequality (Proof) implies 2 (RM+R+L)d \cdot r\left((RM+R+L)d\right) < d_0. According to Proposition ? the pseudotrajectory \{z_{k \in [-(N+1), N+1]}\} can be 2(RM+R)(RM+R+L)d \cdot r\left((RM+R+L)d\right)-shadowed by an exact trajectory q_{k \in [-(N+1), N+1]}. Then by (Proof)
and the pseudotrajectory \{y_{k \in [-(N+1), N+1]}\} is Md-shadowed by \{q_{k \in [-(N+1), N+1]}\}, which completes the induction step and proves Lemma Proof.
Every finite interval I\subset\mathbb{Z} is contained in [-N,N] for some N>0. Restricting the trajectory supplied by Lemma Proof to I proves Lemma 42.
Assume that f:\mathbb{B}\to\mathbb{B} is a C^1-diffeomorphism satisfying (?)-(?), and that there exists d_1, M > 0 such that, for any d < d_1 and any d-pseudotrajectory \{y_k\}, the sequence {\mathcal{B}} = \{B_k = Df(y_k)\}_{k \in \mathbb{Z}} satisfies the bounded solution property with constant M. Then f satisfies the Lipschitz shadowing property.
By (?) there exists d_0 \in (0, d_1) such that
For d_0 satisfying (Proof) and any d-pseudotrajectory \{y_k\}_{k \in \mathbb{Z}} with d<d_0 there exists d/2-pseudotrajectory \{z_k\}_{k \in \mathbb{Z}} such that
Consider a d-pseudotrajectory \{y_k\} with d < d_0. Similarly to the proof of Lemma 42 consider the sequence \{w_{k \in \mathbb{Z}}\} defined by
Consider B_k = Df(y_k). By the assumptions of Lemma 45 there exists a sequence v_k \in \mathbb{B} such that
Consider the sequence \{z_k = y_k + dv_k\}_{k \in \mathbb{Z}}. By construction, \{z_k\} satisfies (Proof). We now show that \{z_k\} is a d/2-pseudotrajectory. By the construction of the sequence \{v_k\} we have
Due to (?), (Proof) and (Proof) the following inequalities hold
Proposition Proof is proved.
Continue the proof of Lemma 45. For a given d-pseudotrajectory \{y_k\}_{k \in \mathbb{Z}} with d< d_0 consider the sequence of \delta_l-pseudotrajectories \{y_k^l\}_{k \in \mathbb{Z}} defined inductively.
Define \delta_0 = d, \{y_k^0 = y_k\}_{k \in \mathbb{Z}}.
By a given \delta_l-pseudotrajectory \{y_k^l\}_{k \in \mathbb{Z}} with \delta_l < d_0 according to Proposition Proof choose \delta_{l+1} = \delta_l/2-pseudotrajectory \{y_k^{l+1}\} such that |y_k^{l+1} - y_k^l| \leq M \delta_l. By construction we have \delta_l = \frac{d}{2^l} and
Hence, for any fixed k \in \mathbb{Z}, the sequence \{y_k^l\} is a Cauchy sequence with respect to l and there exists a limit x_k = \lim_{l \to \infty} y_k^l. Note that
The continuity of f and condition (?) imply that x_{k+1} = f(x_k). Therefore, \{x_k\} is an exact trajectory of f that 2Md-shadows \{y_k\}. Lemma 45 is proved.
Note that this construction uses the Axiom of Choice, and we do not know whether this dependence can be removed. For each trajectory \omega of g, we choose a base point y_\omega and a trajectory of f shadowing the trajectory of g through y_\omega. Once these choices are fixed, the construction in Lemma 35 and its equivariance under shifts define the splitting consistently along the whole trajectory \omega. The Axiom of Choice is used to make these selections simultaneously over all trajectories of g; the nonuniqueness of shadowing prevents a canonical choice.
We say that the pair (\alpha, A) satisfies cocycle generalized hyperbolicity if there exist C > 0, \lambda \in (0, 1) and a family of projections P_x, Q_x: \mathbb{B} \to \mathbb{B} satisfying the following.
(C-GH1) Splitting. P_x + Q_x = Id, \|P_x\|, \|Q_x\| \leq C.
Denote E^s_x = P_x(\mathbb{B}), E^u_x = Q_x(\mathbb{B}); then E^s_x \oplus E^u_x = \mathbb{B}.
(C-GH2) Inclusion. A(x)E^s_x \subseteq E^s_{\alpha(x)}, A^{-1}(x)E^u_{\alpha(x)} \subseteq E^u_{x},
(C-GH3) Exponential estimates.
If, additionally, \alpha and \alpha^{-1} are uniformly continuous and the projections P_x, Q_x depend uniformly continuously on x, i.e. satisfy (14), then we say that (\alpha,A) satisfies uniformly continuous cocycle generalized hyperbolicity.
Assume that (\alpha, A) satisfies the uniformly continuous cocycle generalized hyperbolicity with constants C, \lambda. Then there exists L = L(\lambda, C), such that the restriction T_{\alpha, A}: Y \to X is a bijection and admits a bounded inverse
This statement generalizes [NilsonAli, Theorem 1, Step 1], where the case in which A(x) is a constant generalized hyperbolic isomorphism was considered.
Consider the operator S_{\alpha, A}:X \to Y as in Lemma 31. Given w \in X, define v = S_{\alpha, A}w by
As in Lemma 31, the series (Proof) converges for every x and defines a function v that satisfies (?). Moreover, |v(x)| \leq L\|w\|, where L = C^2\frac{1+\lambda}{1-\lambda}. Define v_N(x) analogously by truncating the sum in (Proof) to |i| \leq N. Then v_N \to v uniformly and since each v_N is uniformly continuous, v is also uniformly continuous. The first sum in (Proof) takes values in E_x^s, and the second sum takes values in F_x. Hence, v(x) \in E_x^s \oplus F_x and v \in Y. Therefore, S_{\alpha, A} is well-defined and satisfies T_{\alpha, A}S_{\alpha, A} = Id, \|S_{\alpha, A}\| \leq L, so T_{\alpha, A}: Y \to X is surjective.
We now show that its kernel is trivial. Assume that for some v \in Y, v \ne 0,
Consider x \in \mathbb{B} such that v(x) \ne 0. Relation (?) implies that
Denote v_n = v(\alpha^n(x)) \in E^s_{\alpha^n(x)} \oplus F_{\alpha^n(x)}. Define v_n = v_n^s + v_n^u, where v_n^s \in E^s_{\alpha^n(x)}, v_n^u \in F_{\alpha^n(x)}. Note that
Hence,
By induction, we conclude
Conditions (C-GH1), (C-GH3) imply that
Note that
if v_{0}^s \ne 0 then the values |v_{-n}|, n \geq 0 are unbounded,
if v_{0}^u \ne 0 then the values |v_{n}|, n \geq 0 are unbounded.
Since v(x) \ne 0, at least one of v_0^s, v_0^u is nonzero, so the sequence \{v(\alpha^n(x))\}_{n \in \mathbb{Z}} is unbounded; this is impossible, since v \in Y \subset X is bounded. This shows that the kernel of T_{\alpha, A}: Y \to X is trivial, so T_{\alpha, A} is injective. Hence, T_{\alpha, A} is invertible and S_{\alpha, A} is its inverse.
Consider a cocycle (\alpha, A) satisfying uniformly continuous cocycle generalized hyperbolicity. We say that the linear mappings H^s_x:E^s_x \to F_x depend uniformly continuously on x if the mappings
are uniformly continuous in x in the operator norm. Similarly, we define the uniform continuity for H^u_{x}: E_{\alpha(x)}^u \to A(x)E_x^s \subset E_{\alpha(x)}^s.
Assume that the pair (\alpha, A) satisfies uniformly continuous cocycle generalized hyperbolicity with constants C > 0, \lambda \in (0, 1). Then for any \lambda_1 \in (\lambda, 1) and \varepsilon >0 there exist C_1 > 1 and \delta >0 such that for any homeomorphism \beta: \mathbb{B} \to \mathbb{B}, with \beta and \beta^{-1} uniformly continuous, satisfying
there exist mappings H^s_x: E_x^s \to F_x \subset E_x^u, H^u_{x}: E_{\alpha(x)}^u \to A(x)E_x^s \subset E_{\alpha(x)}^s that depend uniformly continuously on x such that \|H^s_x\|, \|H^u_x\| < \varepsilon and the pair (\beta, A) satisfies uniformly continuous cocycle generalized hyperbolicity with constants (C_1, \lambda_1) and spaces
There exists \varepsilon_0 such that if \varepsilon < \varepsilon_0 then
The proof uses ideas similar to those in Lemma 35. We highlight the main differences.
First, we make a construction of \tilde{E}^s_x. Fix \varepsilon >0, \lambda_1 \in (\lambda, 1). Denote F_x = A(x)^{-1}E^u_{\alpha(x)}
According to (C-GH2)
Since P_x, Q_x depend uniformly continuously on x, for any \varepsilon_1 \in (0, 1) there exists \delta > 0 such that, if \beta satisfies (52), then
Define J_x: E^u_{\alpha(x)} \to E^u_{\beta(x)} as J_x := Q_{\beta(x)}|_{E^u_{\alpha(x)}}, then J_x is an isomorphism, and
In the following we show that for any \lambda_2 \in (\lambda, \lambda_1) there exists C_2 > 0 and small enough \varepsilon_1 such that B^{ss} and B^{uu} satisfy
(B1) \|B^{ss}_{\beta^{n-1}(x)} \dots B^{ss}_{\beta(x)}B^{ss}_x\| \leq C_2\lambda_2^n, \quad n \geq 0, x \in \mathbb{B},
(B2) B^{uu}_{x} has the right inverse Z_x: E_{\beta(x)}^u \to F_x such that
The proof of (B1) is similar to the argument in (?). Fix \lambda_2 and choose N so that 2C\lambda^N \leq \lambda_2^N. Set C_2=(R(C+2)/\lambda_2)^N. Note that for any \delta_1 > 0 there exists \delta > 0 such that, whenever \beta satisfies (52), it holds \operatorname{dist}(\beta^n(x), \alpha^n(x)) \leq \delta_1 , (n \in [1, N]). For small enough \delta_1 we have
Similarly to (?) we conclude (B1).
The proof of (B2) is similar to the argument in (B1) and (?). Choosing N as in (B1), reducing additionally \delta_1 we may assume that \operatorname{dist}(\beta^{-n}(x), \alpha^{-n}(x)) \leq \delta_1, (n \in [1, N]). Define Z_x = A(x)^{-1}J_x^{-1}, which satisfies (?). As in (B1), we conclude that
For arbitrary n\geq0, write n=lN+r, where l\geq0 and 0\leq r<N. Splitting the product into blocks of length N and using \|Z_x\|\leq(1+\varepsilon_1)R, we obtain (?). Item (B2) is proved.
In the following, we construct a map H_x^s:E_x^s \to F_x with the following property: for any v^s \in E^s_x there exists w^s \in E_{\beta(x)}^s such that
For an arbitrary map H_x^s we can represent A(x)(v^s+ H^s_x(v^s)) = w^s + w^u, where
Condition (?) is equivalent to w^u = H^s_{\beta(x)}w^s, i.e.,
To satisfy (?) it is sufficient to verify the following (compare with (?))
As in Section ?, we consider the following auxiliary problem. Given a family of bounded linear operators S_x:E^s_x \to F_x that depends uniformly continuously on x, we seek a family of bounded linear operators H_x: E^s_x \to F_x that depends uniformly continuously on x, which satisfies H_x = Z_xH_{\beta(x)}B_x^{ss} + S_x (compare to (?)).
Denote by U the Banach space of uniformly bounded linear operators S_x : E^s_x \to F_x that depend uniformly continuously on x.
Define an operator F: U \to U by F(S_x) = H_x, where
Similarly to (?) the following holds
Since \beta and \beta^{-1} are uniformly continuous, the right-hand side depends uniformly continuously on x, and hence H_x \in U.
Define the nonlinear operator G:U \to U by G(H_x) = S_x, where
Similarly to Section ?, for small enough \delta > 0, \varepsilon_2 > 0 and \|H_x\|, \|H'_x\| \leq \varepsilon_2 we have \|GH_x-GH'_x\| \leq \frac{1}{2L_2'}\|H_x-H'_x\|. Since F is linear, inequality (?) implies that
According to (?), (?) and (B2) we have inequalities \|G(H_x = 0)\| = \|Z_xB^{us}_x\| \leq CR\varepsilon_1 and \|F \circ G (H_x = 0)\|\leq L_2'CR\varepsilon_1. Hence, there exists H_x \in U with \|H_x\| \leq 2L_2'CR\varepsilon_1 such that F\circ G (H_x) = H_x, which implies (?). Consequently, (?) holds.
Proceeding as in (?), by choosing \varepsilon_1 sufficiently small and, if necessary, decreasing \delta, we obtain generalized hyperbolicity estimates with exponent \lambda_1 for spaces \tilde{E}_x^s.
Now we carry out the corresponding construction for \tilde{E}^u_x. Here, the stable and unstable cases are genuinely asymmetric, more so than in Lemma 35: there the operators are perturbed while the index shift k \to k+1 is common to \mathcal{A} and its perturbation, whereas here the cocycle A is fixed and the base map is perturbed, \alpha \to \beta. Time-reversal A \to A^{-1} therefore does not turn the perturbed system into a base-only perturbation of the reversed original (its fiber map is sampled at \beta^{-1}(x), not \alpha^{-1}(x)), so H^s_x and H^u_x act between genuinely different spaces and cannot be related by symmetry; this is what forces the auxiliary splitting at \alpha(\beta^{-1}(x)) introduced below.
In order to define H^u_{x}: E_{\alpha(x)}^u \to A(x)E_x^s we need to establish an analog of (?), i.e. to show that for any \tilde{v}^u \in \tilde{E}^u_{\beta(x)} there exists \tilde{w}^u \in \tilde{E}^u_{x} such that A(x)\tilde{w}^u = \tilde{v}^u, equivalently, for any v^u \in E^u_{\alpha(x)} there exists w^u \in E^u_{\alpha(\beta^{-1}(x))} such that
Note that
For small enough \delta we have the splitting \mathbb{B} = E^s_{\alpha(\beta^{-1}(x))} \oplus E^u_{\alpha(\beta^{-1}(x))}. Denote by \tilde{P}_x, \tilde{Q}_x the projections corresponding to this splitting. Similarly to the construction of \tilde{E}^s_x define
Note that since we considered the decomposition of A^{-1}, the operator C_x^{ss} is “expanding” and the operator C_x^{uu} is “contracting”. Equation (?) is equivalent to
Hence, the condition on H^u is
Similarly to (?) define the operator \tilde{Z}_x:E^s_{\alpha(\beta^{-1}(x))} \to A(x)E^s_x such that \tilde{Z}_x \circ C^{ss}_x = Id_{A(x)E^s_x}. Condition (?) is equivalent to
Exactly as for (B1) and (?), but using (C-GH3) for A^{-1} in place of the estimates for A, for small enough \varepsilon_1 and some C_2 > 0 the contracting block C^{uu} and the right inverse \tilde{Z}_x satisfy
Hence (?) is the analog of (?) with \beta, B^{ss}_x, B^{su}_x, B^{us}_x, Z_x replaced by \beta^{-1}, C^{uu}_x, C^{us}_x, C^{su}_x, \tilde{Z}_x, and the fixed-point argument for H^s_x applies verbatim: for any \varepsilon > 0 and small enough \delta there is an operator H^u_x satisfying (?) with \|H^u_x\| < \varepsilon.
The spaces \tilde{E}^u_x satisfy the exponential estimates with exponent \lambda_1 as for \tilde{E}^s_x, via (?). As in the construction of H^s_x we prove that H^u_x depends uniformly continuously on x.
Let us prove equality (52). According to construction \tilde{E}^s_x \subset E^s_x \oplus F_x and
Hence,
For any x \in \mathbb{B} and v^s_x \in E^s_x define the vectors
Then v^s_x + v^u_x \in \tilde{E}^s_x and v^u_{\alpha(x)} + v^s_{\alpha(x)} \in \tilde{E}^u_{\beta(x)}. Hence,
Note that for \varepsilon < 1/R the following holds
hence Id - A^{-1}(x) H^u_x A(x) H^s_x is bijective on E^s_x. In particular E^s_x \subset \tilde{E}^s_x \oplus A^{-1}(x)\tilde{E}^u_{\beta(x)}. Similarly, we show F_x = A^{-1}(x)E^u_{\alpha(x)} \subset \tilde{E}^s_x \oplus A^{-1}(x)\tilde{E}^u_{\beta(x)}, and hence E^s_x \oplus F_x \subset \tilde{E}^s_x \oplus A^{-1}(x)\tilde{E}^u_{\beta(x)}, which together with inclusion (?) implies (52).
Note that in Step 3 we strongly used that E^s_x\oplus F_x = \tilde{E}^s_x\oplus \tilde{F}_x and hence h_1 and h_2 are unique in the same space Y. This required special attention in Lemma 52 to establish equality (52). The additional condition in Theorem 16 (?) was used only to establish the inclusions (?).
Let the pair (\alpha,A) satisfy cocycle generalized hyperbolicity with constants C,\lambda, with no continuity assumption on the projections P_x,Q_x. Let X=C_b(\mathbb{B},\mathbb{B}) be the Banach space of bounded continuous functions with the supremum norm, and let T_{\alpha,A}:X\to X be defined by (?). Then there exist a constant L=L(\lambda,C) and a not necessarily linear map S_{\alpha,A}:X\to X such that T_{\alpha,A}\circ S_{\alpha,A}=Id_X, and S_{\alpha,A} is Lipschitz with constant L.
When do exponential dichotomies on \mathbb{Z}^{\pm} together with the transversality condition imply generalized hyperbolicity?
Assume that T has the finite Lipschitz shadowing property with constant L. Then for every d>0 and every d-pseudotrajectory \{x_j\}_{j \in \mathbb{Z}} of T there exists a \tfrac{d}{2}-pseudotrajectory \{x'_j\}_{j \in \mathbb{Z}} of T such that
Fix an integer m \ge 4L and set n_k := \frac{k(k-1)}{2}m, so that n_0 = n_1 = 0 and n_{k+1} - n_k = km for k \ge 1; the intervals [n_k, n_{k+1}), k \ge 1, partition [0, +\infty) \cap \mathbb{Z}. For each k \ge 1 the restriction \{x_j\}_{|j| \le n_{k+1}} is a finite d-pseudotrajectory, so by the finite Lipschitz shadowing property there is u_k \in X with
We interpolate linearly between consecutive shadowing points. For j \ge 0 write j = n_k + i with k \ge 1 and 0 \le i < km, and put
for j < 0 write -j = n_k + i in the same way and put x'_j := \big(1 - \tfrac{i}{km}\big) T^j u_k + \tfrac{i}{km}\, T^j u_{k+1}.
Displacement. Let j \ge 0, j = n_k + i. Then j \le n_{k+1} \le n_{k+2}, so (Proof) holds at j both for u_k and for u_{k+1}; as x'_j is a convex combination of T^j u_k and T^j u_{k+1},
The case j<0 is identical, so \|x_j - x'_j\| \le Ld for all j \in \mathbb{Z}.
Halved error. Let j \ge 0, j = n_k + i with 0 \le i < km. If i+1 \le km-1, then j+1 lies in the same block and a direct computation gives
If i+1 = km, then j+1 = n_{k+1} is the first index of the next block, where x'_{j+1} = T^{j+1}u_{k+1}, and the same identity follows since x'_j = \tfrac{1}{km}T^{j}u_k + \big(1-\tfrac{1}{km}\big)T^{j}u_{k+1} in this case. In either case j+1 \le n_{k+1} \le n_{k+2}, so (Proof) applies at j+1 for both u_k and u_{k+1}, and therefore
where we used k \ge 1 and m \ge 4L. The estimate for j < 0, including the junction between j=-1 and j=0, is analogous. Hence, \{x'_j\} is a \tfrac{d}{2}-pseudotrajectory.
A C^1-diffeomorphism f:M\to M is structurally stable if and only if
Assume that f satisfies Axiom A and the strong transversality condition. Then there exist constants C_0>0 and \lambda_0\in(0,1) such that, for every p\in M, the derivative sequence \left\{ Df(f^k(p)):T_{f^k(p)}M\to T_{f^{k+1}(p)}M \right\}_{k\in\mathbb{Z}} has a (C_0,\lambda_0)-structure in the sense of [PilCL].
There exist neighborhoods Z_i of \Omega_i, continuous subbundles E_i^u,E_i^s of TM|_{\mathcal O(Z_i)}, i=1,\ldots,m, and \theta\in(0,1) such that:
(1) E_i^u and E_i^s are invariant under Df;
(2) if x\in\mathcal O^+(Z_j)\cap\mathcal O^-(Z_i), then E_{i,x}^u\subseteq E_{j,x}^u, E_{i,x}^s\supseteq E_{j,x}^s;
(3) E_{i,x}^u \oplus E_{i,x}^s=T_xM, for any x\in\mathcal O(Z_i);
(4) if x\in Z_i, v^s\in E_{i,x}^s, and v^u\in E_{i,x}^u, then
Here \mathcal O^+(Z)=\bigcup_{n\geq0}f^n(Z), \mathcal O^-(Z)=\bigcup_{n\leq0}f^n(Z), \mathcal O(Z)=\mathcal O^+(Z)\cup\mathcal O^-(Z). The one-step estimates in item (4) are asserted for x\in Z_i, not for all x\in\mathcal O(Z_i).
Let \{\gamma_i\}_{i=1}^m be a C^{\infty} partition of unity subordinate to the finite open cover \{\mathcal O(Z_i)\}_{i=1}^m. There exists an integer q\geq0 such that, for every i and every x\in\operatorname{supp}\gamma_i,
the forward orbit of x lies in \bigcup_{j\geq i}Z_j except at most q iterates;
the backward orbit of x lies in \bigcup_{j\leq i}Z_j except at most q iterates.
Set b_*:=\sup_{x\in M}\max\left\{\|Df(x)\|_*,\|D(f^{-1})(x)\|_*\right\}, C_*:=\frac{b_*^q}{\theta^q}. Then:
(5) for every x\in\operatorname{supp}\gamma_i, n\geq0, v^s\in E_{i,x}^s, and v^u\in E_{i,x}^u,
(6) the stable and unstable bundles are uniformly transverse on \operatorname{supp}\gamma_i. Consequently, there exists C_{\mathrm{proj},*}>0, independent of i and x, such that the projections
satisfy \|\pi_{i,x}^s\|_*, \|\pi_{i,x}^u\|_* \leq C_{\mathrm{proj},*}, for x\in\operatorname{supp}\gamma_i.
The text on this page is from arXiv:2510.05499v3 by Sergey Tikhomirov, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2510/2510.05499.jsonl.