Invariant measures for random dynamical systems on the circle (Q1348962)

A system \((\varphi,X,T,(\Omega,{\mathcal F},P),(\vartheta_t))\) is called random dynamical system (abbreviated RDS), if \(x\mapsto\varphi(t,\omega)x\) and \(t\mapsto\varphi(t,\omega)x\) are continuous \(P\)-almost surely, \(\varphi\) satisfies the cocycle property, \(X\) is a Polish space, \(T\colon T\times X\times\Omega\to X\) is a measurable map, \((\Omega,{\mathcal F},P)\) is a probability space, and \(\vartheta_t\) is an ergodic flow on \(\Omega\). An invariant measure \(d\mu(x,\omega)=d\mu_\omega(x)dP(\omega)\) for \(\varphi\) is a probability measure on \(X\times \Omega\) with marginal distribution \(P\) which is invariant with respect to \(\Theta_t(x,\omega)=(\varphi(t,\omega)x,\theta_t(\omega))\). The support of an invariant measure \(\mu\) is the set-valued map \(\omega\mapsto\text{supp} \mu_\omega\), and \(\mu\) is said to be a point mass, if \(\text{supp} \mu_\omega\) is finite almost surely. In this article, the state space \(X\) is either \(\mathbb{R}\) or \(S^1\). For an RDS, there are two \(\sigma\)-algebras defined by \[ \begin{aligned} {\mathcal F}_{\leq 0} & =\sigma\{\omega\mapsto\varphi(t,\vartheta_{-s}\omega)x: 0\leq t\leq s,x\in X\}\quad{\text{past}},\\ {\mathcal F}_{\geq 0} & =\sigma\{\omega\mapsto\varphi(t,\vartheta_s\omega)x:s,t\geq 0,x\in X\}\quad \text{future.}\end{aligned} \] Definition 3.2. An RDS is said to be white noise, if its past \({\mathcal F}_{\leq 0}\) and its future \({\mathcal F}_{\geq 0}\) are independent. When \(\text{card}(\text{supp }\mu)\) is finite \((m_1(\omega)<\cdots<m_m(\omega))\), then intervals \(I_j(\omega)=[m_j(\omega),m_{j+1}(\omega)]\) are called basic intervals. Theorem 3.3: Suppose \(\varphi\) is a two-sided time white noise, and suppose the past measurable invariant measure \(\mu\), as well as future measurable invariant measure \(\nu\) are unique, respectively, and that they both are point measures. Then \(\text{ card}(\text{supp }\mu) =\text{card}(\text{supp }\nu)\), and each of the basic intervals induced by \(\text{supp }\mu\) contains exactly one of \(\text{supp} \nu\), and vice versa. Denote by \(D_x\varphi(t,\omega)\) the linearization of \(\varphi\) which is \(C^1\). Let \[ \lambda(\mu)=\lim_{t\to\infty}{1\over t}\log|D_x\varphi(t,\omega)|. \] This exists and is constant with respect to an ergodic invariant measure \(\mu\), and is called the Lyapunov exponent associated with \(\mu\). Proposition 4.2: Suppose that \(\mu\) is an ergodic invariant measure for \(C^1\) RDS on the unit circle, and that \(\sup_{0\leq t\leq 1}\log^+|D_x\varphi(t,\omega)|\) is integrable with respect to \(\mu\). If \(\lambda(\mu)\neq 0\), then \(\mu\) is a point measure. Proposition 4.3: Suppose that \(\mu\) is an ergodic invariant measure for \(C^1\) RDS on the unit circle, and that \(\sup\{\log|D_x\varphi(t,\omega)|:x\in S^1\}\) is integrable with respect to \(P\). Then (1) if \(\lambda(\mu)\leq 0\) then \(\omega\mapsto\mu_\omega\) is measurable with respect to \({\mathcal F}_{\leq 0}\). (2) if \(\lambda(\mu)\geq 0\) then \(\omega\mapsto\mu_\omega\) is measurable with respect to \({\mathcal F}_{\geq 0}\). Corollary 4.4: Let \(\omega\mapsto\mu_\omega\) be an ergodic invariant measure for a white noise \(C^1\) RDS on \(S^1\). Then (1) If \(\lambda(\mu)=0\), then \(\mu_\omega=\rho\) a.s. for some probability measure \(\rho\). Moreover, either \(\rho\) is Dirac or \(\text{ supp}\rho=S^1\). (2) If \(\lambda(\mu)\neq 0\), then \(\mu\) is measurable with respect to \({\mathcal F}_{\leq 0}\) or \({\mathcal F}_{\geq 0}\). (a) \(\mu\) is measurable with respect to both \({\mathcal F}_{\leq 0}\) and \({\mathcal F}_{\geq 0}\), if and only if \(\mu\) is a Dirac measure. (b) If \(\mu\) is measurable only for one of \({\mathcal F}_{\leq 0}\) and \({\mathcal F}_{\geq 0}\), then there exists a finite random set \(\omega\mapsto M(\omega)\subset S^1\) with \(\text{card }M=m\) a.s., which is measurable with respect to \({\mathcal F}_{\leq 0}\) (\({\mathcal F}_{\geq 0}\)) if \(\lambda(\mu)<0\) (\(\lambda(\mu)>0\)), respectively, such that \(\mu_\omega={1\over m}\sum_{p\in M(\omega)}\delta_p\).