Passage times of random walks and Lévy processes across power law boundaries (Q2570834)

Consider a random walk \(\{S_n,n\geq 1\}\) (thus, with i.i.d. increments). The first passage time \(\min\{n:S_n>an^\kappa\}\) for \(a>0\) has attracted great attention over the last half-century. When the mean is positive, the natural parameter set is \(0\leq\kappa<1\). When the mean equals zero, there is a sharp boundary at \(\kappa=1/2\). The present paper is devoted to the mean zero case, in particular on ``the remaining case'' \(\kappa\in(1/2,1)\), \(E(X^+)^{1/\kappa}< \infty\), \(E(X^-)^{1/\kappa}=\infty\), where \(X\) is a generic increment variable. The main result is an integral test for deciding precisely when the first passage time is finite a.s., and when not (it may be worth while recalling that only the integrability of the negative tail matters for finiteness and integrability of the first passage time when the mean is positive). The results, as well as those of an earlier paper by H. Kesten and the second author, are then extended to Lévy processes.