Non-Markovian invariant measures are hyperbolic (Q1208953)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Non-Markovian invariant measures are hyperbolic
scientific article

    Statements

    Non-Markovian invariant measures are hyperbolic (English)
    0 references
    16 May 1993
    0 references
    Let \(M\) be a \(d\)-dimensional Riemannian \(C^ r\) manifold, \(r\geq 1\), and denote by \({\mathcal B}\) its Borel \(\sigma\)-algebra. Let \(T\) be either \(\mathbb{R},\mathbb{R}^ +,\mathbb{Z}\), or \(\mathbb{N}\). Let \((\Omega,{\mathcal F},P)\) be a probability space and let \(\{\nu_ i \mid i \in T\}\) be an ergodic family of measure preserving transformations of \((\Omega,{\mathcal F},P)\) such that \(\nu_{t+s}=\nu_ t \circ \nu_ s\) for all \(t,s \in T\). By definition a random dynamical system is a measurable map \(\varphi:T \times M \times \Omega \to M\), such that \(\varphi (0,\cdot)=\text{id}\) and \(\varphi(t+s,\omega)=\varphi(t,\nu,\omega)\circ \varphi (s,\omega)\) for all \(t,s \in T\) outside some \(P\)-nullset in \(\Omega\). Here \(\varphi(t,\omega):M \to M\) is the map which arises when \(t\) and \(\omega\) are kept fixed. A random dynamical system is said to be smooth if \(\varphi(t,\omega)\) is differentiable for each \(t \in T\) and \(\omega \in \Omega\). It is said to be one-sided or two-sided, respectively, according to whether \(T\) is \(\mathbb{R}^ +\) or \(\mathbb{N}\) or whether \(T\) is \(\mathbb{R}\) or \(\mathbb{Z}\). A probability measure \(\mu\) on \(M \times \Omega\) (on the product \(\sigma\)-algebra \({\mathcal B} \times{\mathcal F})\) is said to be an invariant measure for \(\varphi\) if it is invariant under \(\theta_ t\), \(t \in T\), and if it has marginal \(P\) on \(\Omega\). Here \[ \theta_ t:M \times \Omega \to M \times \Omega, \quad (x,\omega) \mapsto(\varphi(t,\omega)x,\nu,\omega), \] where \(\varphi (t,\omega) x=\varphi (t,x,\omega)\). The author establishes the estimate \(\alpha_ \mu \leq dE^ \mu(\max\{0,-\lambda_ d^ \mu\})\), where \(\alpha_ \mu\) is the relative entropy of \(\mu\), \(\lambda^ \mu_ d\) is the smallest Lyapunov exponent associated with \(\mu\), and \(E^ \mu\) denotes integration with respect to \(\mu\). For \(\mu\) ergodic this implies \(\alpha_ \mu \leq d(\max\{0,-\lambda^ \mu_ d\})\), since then the exponents \(\lambda^ \mu_ i\) are constant.
    0 references
    0 references
    ergodic family of measure preserving transformations
    0 references
    random dynamical system
    0 references
    Lyapunov exponent
    0 references
    0 references