Compact locally maximal hyperbolic sets for smooth maps: Fine statistical properties (Q936510)

The authors study a compact locally maximal hyperbolic set by means of geometrically defined functional spaces asserting that this paper provides a self-contained theory that not only reproduces all the known classical results, but also gives the insights on the statistical properties of the set. We consider an open set \(U\subset X\) of a smooth manifold \(X\) and a map \(T\in C^r(U,X)(r> 1)\), diffeomorphism on its image such that \(\Lambda=\bigcap T^nU\) is non-empty and compact. Suppose that \(\Lambda\) is a hyperbolic set for \(T\). We fix an open neighorhood \(U'\) of \(\Lambda\) such that \(TU'\subset U\) and \(T^{-1 U'\subset U}\) and small enough so that the restriction of \(T\) to \(U'\) is still hyperbolic. We also fix a small neighborhood \(V\) of \(\Lambda\), compactly contained in \(U'\). Let \(\lambda> 1\) (resp. \(\nu< 1\)) be two constants smaller (resp, larger) than the minimal expansion (resp. maximal contraction) of \(T\) in the unstable (resp. stable) direction. Denote by \(W^0\) the set of \(C^{r-1}\) function \(\phi\) associating, to each \(x\in U\) and each \(d_s\) dimensional subspace \(E\) of the tangent space \(T_x X\) at \(x\), a real number \(\phi(x,E)\), where \(d_s\) is the dimension of the stable manifold \(W^s(x)\). Denote by \(W^1\) the set of \(C^r\) function \(\phi: U\to R\). For \(x\in\Lambda\), set \(\overline\phi(x)= \phi(x, E^s(x))\) for \(\phi\in W^0\) and \(\overline\phi(x)= \phi(x)\) for \(\phi\in W^1\). The authors construct a functional space \({\mathcal S}\) and define a linear map \({\mathcal L}\) derived from \(T\) on it in the following way. Let \(G\) be the Grassmanian of \(d_s\)-dimensional oriented subspaces of the tangent bundle \(TX\) to \(X\). The set \({\mathcal E}\) of all triple \((x, E,\omega)\) with \((x, E)\in G\), \(\omega\in\Lambda^{d_s}E'\otimes \mathbb{C}\) consists a complex line bundle over \(G\), where \(E'\) is the dual space of \(E\). We consider the vector space \({\mathcal S}\) of all \(C^{r-1}\) sections of the line bundle \(E\). A \(C^r\)-function \(\pi\) on \(X\) is called a truncation function if \(\pi(x)\in[0,1]\) for \(x\in X\) and equal to 1 on a neighborhood of \(\Lambda\) and compactly supported in \(T(V)\). Now, for each such function \(\pi\), and each \(\phi\in W^\iota\), \(\iota\in \{0,1\}\), we define a transfer operator \(L_{\pi,\phi}:{\mathcal S}\to{\mathcal S}\) by \(L_{\pi,\phi}\alpha= T_*(\pi, e^\phi\alpha)\) for \(\alpha\in{\mathcal S}\), where \(T_*\) denotes the naturally defined pushforward under \(T\). It is clear that the iteration of \(L_{\pi,\phi}\) would shed light on the mixing property of \(T\). The operator \(L_{\pi,\phi}\) does not have good asymtotic properties on \({\mathcal S}\) with its \(C^{r-1}\) norm, but the authors show that it behaves well on the spaces \(B^{p,q,\iota}\), which are obtained by the comletion of \({\mathcal S}\) in some semi-norms \(\|\cdot\|_{p,q,\phi}\), defined for some pairs \(p,q\in\mathbb{N}\times\mathbb{R}_+\). If \(W\) is a submanifold of dimension \(d_s\) contained in \(U\), \(\varphi\) is a continuous function on \(W\) with compact support and \(\alpha\in{\mathcal S}\), then we have \[ \int_W L_{\pi,\phi}\alpha= \int_{T^{-1}W} \varphi\circ T\cdot\pi e^\phi\alpha. \] The authors consider the spectral radius \(\rho\) of \({\mathcal L}= L_{\pi,\phi}|_{B^{p,q},\phi}\) and proves that \(\rho\) is a simple eigenvalue \(L\), when \(T\) is orientation mixing. They give also the description for the correlation functions (Theorem 1.2 in the Introduction) as follows: Let \(\phi\in W^\iota\) for some \(\iota\in (0,1)\). Let \(p\in\mathbb{N}\), and \(q\in\mathbb{R}_+\) satisfy \(p+ q\leq r- 1+\iota\). Let \(\sigma>\max(\lambda^{-p}, \nu^q)\). Then there exists a unique measure \(\mu\) on \(\Lambda\), a positive number \(C\), a finite-dimensional vector space \(F\), a linear map \(M: F\to F\) with simple eigenvalue 1, linear maps \(\tau_1: C^p(U)\to F\) and \(\tau_2: C^q(U)\to F\) such that for every \(\psi_1\in C^p(U)\) and \(\psi_2\in C^q(U)\) and \(n\in\mathbb{N}\) the following holds: \[ |\int(\psi_1\cdot\psi_2) T^n d\mu- \tau_2(\psi_2)\,M^n,\;\tau_1(\psi_1)|\leq C\sigma^n|\psi_2|_{C^q(U)}|\psi_1|_{C^p(U)}. \] Concerning the variational principle, the authors prove the following: The spectral radius is equal to the topological pressure \(P_{\text{top}}(\overline\phi)\) of the function. In addition, the measure \(\mu\) is the unique probability measure satisfying the variational principle: \[ h_\mu(T)+ \int\overline\phi\,d\mu= P_{\text{top}}(\overline\phi), \] where \(h_\mu(T)\) is the topological entropy of \(T\). The measure is the so-called Gibbs measure of \(T: \Lambda\to\Lambda\), corresponding to the potential \(\overline\phi\). Finally, the authors try to give an idea of the breadth of the obtained results by discussing some natural examples to which they can be applied, e.g., perturbation theory.