Local Lyapunov exponents. Sublimiting growth rates of linear random differential equations (Q947712)

These lecture notes originate from the author's dissertation and provide a self-contained introduction to the notion of local Lyapunov exponents. This concept combines the theories of Lyapunov exponents and large deviations for the study of the stochastic differential system \[ \begin{aligned} d Z^{\epsilon}_t &= A(X^\epsilon_t) ~Z^\epsilon_t ~d t\\ d X^{\epsilon}_t &= b(X^\epsilon_t)~d t + \sqrt{\epsilon} ~\sigma(X^\epsilon_t)~d W_t. \end{aligned} \] The second equation is a usual Brownian diffusion in \(\mathbb{R}^n\), which drives the first one, a linear autonomous equation in \(\mathbb{R}^d\), the so-called variational equation of \(X^\epsilon\). Oseledet's multiplicative ergodic theorem states that due to stationarity the Lyapunov exponents of \(Z^\epsilon_t\) do not depend on the initial value \(x\) of \(X^\epsilon\). On the other hand it is known from Freidlin-Wentzell theory that in the limit of small noise intensity \(\sqrt{\epsilon}\) the process \(X^\epsilon\) exhibits a quasi-deterministic or meta-stable behaviour, which actually depends on the initial position. The author's new conceptual idea is now to introduce a Lyapunov-type number which does depend on the respective initial domain of attraction of the deterministic equation \(X^0\). This is obtained by choosing an appropriate scaling connection of \(\epsilon = \epsilon(t)\) and \(t\) for \(t\rightarrow \infty\), which up to a factor corresponds to the Freidlin Wentzell-scaling. In this case the author obtains sublimiting distributions supported on the scale-dependent meta-stable states. It is shown that the local Lyapunov exponents are always bounded from by the smallest and the largest real part of the eigenvalues of \(A\) evaluated at the respective meta-stable states. For the special case of \(A\) taking values in the diagonal matrices it is shown that its eigenvalues evaluated at the respective meta-stable states are precisely the possible local Lyapunov exponents. The monograph is structured as follows. After a short introduction, the author develops in the first chapter the theory of linear differential systems with parameter excitation. The second chapter is dedicated to the large deviation principle, Freidlin-Wentzell theory, in particular the asymptotic exit times and the meta-stable behaviour. This is illustrated by several examples in economic time series, climate modelling and biology. The third part consists of an introduction to the study of asymptotic exit times for degenerated systems, the so-called Hernández-Lerma theory, and respective meta-stability results. The last chapter as the core of this book gives a concise exposition of the theory of local Lyapunov exponents and its application to the stochastic differential system. In particular the author obtains conditions for convergence in the general two-dimensional case, where the previously described results by Hernándes-Lerma finally enter. This method is based on the observation that the decomposition of \(Z^\epsilon_t\) by means of spherical coordinates demands for the control of the time the angle process \(Z^\epsilon_t/|Z^\epsilon_t|\) stays close to the so-called switching curve of drift. These lecture notes provide a beautiful exposition to this partially subtle subject for readers acquainted with the theory of stochastic differential equations. The three preliminary chapters about linear differential systems and Lyapunov exponents, Freidlin-Wentzell and Hernández-Lerma theory offer to the interested reader the broad background necessary to fully appreciate the obtained results. Great qualities of this book are also the ample bibliography giving a representative state of the large literature in this field and the great amount of instructively worked out examples. The author explains deep results to the reader intuitively without omitting the sometimes technical details this field has to face. The composition of the text is throughout clear, carefully thought through and harmonic.