Stable processes conditioned to avoid an interval

Abstract

Conditioning Markov processes to avoid a domain is a classical problem that has been studied in many settings. Ingredients for standard arguments involve the leading order tail asymptotics of the distribution of the first hitting time of the domain of interest and its relation to an underlying harmonic function. In the present article we condition stable processes to avoid intervals. The required tail asymptotics in the stable setting for $\alpha\geq 1$ go back to classical work of Blumenthal et al. and Port from the 1960s. For $\alpha<1$, we appeal to recent results centred around the so-called deep factorisation of the stable process to compute hitting probabilities and, moreover, to identify the associated harmonic functions for all $\alpha\in (0,2)$. With these in hand, we thus prove that conditioning to avoid an interval is possible in the classical sense and that the resulting process is a Doob $h$-transform of the stable process killed on entering the aforesaid interval. Appealing to the representation of the conditioned process as a Doob $h$-transform, we verify that the conditioned process is transient.