Exact asymptotics of large deviations of stationary Ornstein-Uhlenbeck processes for \(L^p\)-functionals, \(p>0\) (Q2498322): Difference between revisions

Let \(\eta_\gamma(t)\), \(t\in\mathbb R\), be a stationary Ornstein-Uhlenbeck process of order \(\gamma>0\), i.e., it holds \[ \mathbb E \eta_\gamma (t) \eta_\gamma(s) = e^{-\gamma| t-s| }\;,\quad t,s\in\mathbb R\;. \] The aim of the present paper is to investigate the asymptotic behavior of \[ \mathbb P\left\{\int_0^1 | \eta_\gamma(t)| ^p d t >u^p\right\} \tag{1} \] as \(u\to\infty\) and where \(p>0\) is some given number. The author is able to describe the exact order of (1), even with precise constants. For example, if \(p=1\), the asymptotic of (1) is \(\exp\left(-\frac{u^2}{\sigma^2}\right) \frac{\sqrt 2\, \sigma}{u\sqrt\pi}(1+o(1))\) where \(\sigma^2 = \frac{2(\gamma-1+e^{-\gamma})}{\gamma^2}.\) One should compare this with the \(L_\infty\)-case where the order is known to be \(\exp\left(-\frac{u^2}{2}\right) \frac{2\gamma u}{u\sqrt{2\pi}}(1+o(1)).\) A basic tool for proving the asymptotic of (1) is the Laplace method for \(L^p\)-functionals. This seems to be very interesting own its own right and may be used to attack related questions.