First exit times for Lévy-driven diffusions with exponentially light jumps (Q1019089)

The authors study the first exit time from the interval \((-1,1)\) for the trajectories of the solution of the stochastic differential equation \[ dX_t^\varepsilon = -U'(X_{t-}^\varepsilon) dt + \varepsilon dL_t, \] where \(L\) is a symmetric Lévy process having exponentially light tails. That is, if \(\nu\) denotes the associated jump measure, \(\nu([u,+\infty)) \sim e^{-u^\alpha}\), \(\alpha>0\) and \(u\rightarrow +\infty\). The potential \(U\) is such that, among other assumptions, the corresponding deterministic dynamical system has a unique stable point at the origin. Exit problems for dynamical systems with Brownian perturbations are by far well understood and the main reference is [\textit{M. I. Freidlin} and \textit{A. D. Wentzell}, Random Perturbations of Dynamical Systems. Grundlehren der Mathematischen Wissenschaften. 260. New York, NY: Springer (1998; Zbl 0922.60006)]. On the other hand, random dynamical systems driven by a heavy-tailed Lévy noise exhibit a different behavior in comparison to the situation described above, for big jumps play the essential role in the exit time dynamics and the paths of the perturbed system, indeed, leave the interval in one big jump. Moreover, exit time problems for jump diffusions driven by an \(\alpha\)-stable Lévy process whose jump measure has regularly varying tails have been tackled e.g. in [\textit{P. Imkeller} and \textit{I. Pavlyukevich}, ESAIM Probab. Stat. 12, 412--437 (2008; Zbl 1205.60113)]. As it is commented in the introduction of the paper under review, one of the main motivations in the study of such problems is the attempt to model paleoclimatic time series from the Greenland ice core by dynamical systems perturbed by noise. Going back to the problem under consideration, the authors study the behavior of the exit time \(\sigma\) of \(X^\varepsilon\) from \((-1,1)\) as \(\varepsilon\rightarrow 0\). More precisely, they study the asymptotic law and mean value of the first exit time of the jump-diffusion and they prove that the latter exhibits a phase transition at \(\alpha=1\) of the form: \(\ln E(\sigma) \sim \varepsilon^{-\alpha}\) in the sub-exponential case (\(0<\alpha<1\)) and \(\ln E(\sigma) \sim \varepsilon^{-1} |\ln \varepsilon|^{1-\frac 1\alpha}\) in the super-exponential case (\(\alpha>1\)). Quoting from the authors, this behavior can be roughly explained as follows: ``Big jumps of the Lévy process govern the asymptotic behavior in the sub- and super-exponential regimes, but in the latter one the cumulative action of several large jumps, reminding the climbing of the potential well in the Gaussian regime, becomes important, while for \(\alpha<1\) the biggest jump alone governs the exit.'' Eventually, the same type of techniques are also applied to study random perturbations given by Lévy processes with bounded jumps.