On the tail asymptotics of the area swept under the Brownian storage graph (Q2448698): Difference between revisions

Let \(X(t)\) be a stochastic process. The paper deals with the corresponding workload process \[ Q(t) = \mathop {\sup }\limits_{s \leqslant t} \left( {X(t) - X(s) - c(t - s)} \right), \] where \(c > 0\) is the drain rate of the queue, and investigates probabilities \[ {\pi _{T(u)}}(u) = P\left( {\int_0^{T(u)} {Q(t)dt} > u} \right) \] for different types of interval lengths \(T(u)\), \(u \to \infty \). The authors assume that \(X(t)\) is a standard Brownian motion, and the workload is in stationary regime. In Section 3, they consider the case \(T(u) = o(\sqrt u )\) and give an explicit function \(\varphi (u)\) such that \({\pi _{T(u)}}(u)/\varphi (u) \to 1\) as \(u \to \infty \). In Section 4, the case \(T(u) = T\sqrt u \) is considered. Another explicit function \(\varphi (T,M)\) is proposed for the equality \[ - \mathop {\lim }\limits_{u \to \infty } \frac{1}{{\sqrt u }}\log {\text{P}}\left( {\int_0^{T\sqrt u } {Q(r)dr} > Mu} \right) = \varphi (T,M) \] to hold. Section 5 deals with the case \(\sqrt u = o(T(u)\) but \(T(u) = o(u)\). It is shown that \({\pi _{T(u)}}(u)\) behaves like \(\exp ( - \alpha \sqrt {u)} \) for some \(\alpha > 0\). Section 6 considers the integral over the remaining busy period. It gives an explicit expression of the Laplace transform of this integral in terms of the Airy function.