A sufficient condition for the uniform exponential stability of time-varying systems with noise. (Q1428422)

Consider the multi-scale initial value problems of ordinary differential equations of the form \[ \dot{x}_\varepsilon = f \left(\frac{t}{\varepsilon},x_\varepsilon(t),\xi(t)\right), \quad x_\varepsilon(0)=x_0 \in \mathbb R^n , \] where \(\xi : [0,+\infty) \to M\) is a measurable, deterministic function representing noise with values in a compact metric space \(M\) and \(\varepsilon\) is a small perturbation parameter. The authors adapt a multi-valued averaging approach which has originally been introduced for differential inclusions to this noisy system. They are interested in conditions under which a transfer of exponential stability from the related averaged system to the above formulated original system takes place for small parameters \(\varepsilon > 0\). In contrast to the standard Lyapunov-techniques requiring differentiability of the averaged system, their approach relies on a direct computation of the distance between trajectories of the originally perturbed system and those of the related averaged system without frequently met smoothness assumptions. In fact, the authors extend the results of \textit{G. Grammel} [J. Dyn. Control Syst. 7, No.3, 327-338 (2001; Zbl 1038.34047)] to systems with not necessarily periodic fast time variable. For this purpose, they introduce set-valued finite time averages that converge as time tends to infinity, even if the vector field \(f\) is not of Krylov-Bogoliubov-Mitropolskii type. Counter examples are discussed, too.