Stability of Oseledets spaces is equivalent to stability of Lyapunov exponents (Q4257744)

Consider a product of stationary random matrices \(\Phi(n)= A_{n-1} \dots A_0\), \(n\in\mathbb{N}\), given by a probability measure \(P\) on \(\text{Gl} (d,\mathbb{R})^\mathbb{Z}\), which is invariant with respect to the shift \(\sigma\). A random perturbation of \(\Phi\) is given by \(\Psi(n)= A_{n-1}B_{n-1} \dots A_0B_0\), where \(B_j\) are random matrices whose distributions are allowed to depend on \(A_j\) in a stationary way. This can be achieved by considering a measure \(\widehat P\) on \(\text{Gl} (d,\mathbb{R})^\mathbb{Z} \times\text{Gl}(d,\mathbb{R})^\mathbb{Z}\) which is invariant under \(\sigma\times\sigma\), and letting \((A_n)\) be distributed according to \(P=\pi_1\widehat P\), \(\pi_1\) denoting the projection to the first component. The measure \(\widehat P\) is said to define an \(\varepsilon\)-perturtation of \(P\) if \[ \widehat P\bigl\{(\omega,\eta) \in\text{Gl}(d,\mathbb{R})^\mathbb{Z} \times\text{Gl}(d,\mathbb{R})^\mathbb{Z}: |\eta_0-\text{Id}|>\varepsilon\bigr\} <\varepsilon. \] Assuming the usual integrability conditions for the multiplicative ergodic theorem and an additional uniform integrability condition for a sequence of \(\varepsilon_i\)-perturbations \(\Phi_{\varepsilon_i}\) of \(\Phi\) with \(\varepsilon_i\) tending to zero for \(i\to \infty\), it is shown that convergence of the Lyapunov exponents of \(\Phi_{\varepsilon_i}(n)\) to those of \(\Phi(n)\) is equivalent to convergence of the Oseledets spaces of \(\Phi_{\varepsilon_i}(n)\) to those of \(\Phi(n)\) for \(i\to\infty\).