Random walks with killing (Q1102040)

Let P(x,dy) be the transition probability kernel of a random walk on \(R^ n\). Let \(k:R^ n\to [0,1]\) be a given function. Suppose that the random walk is ``killed'' according to k: if it is at position x, then it jumps to a special state (the cemetery) with probability k(x); otherwise, it executes a usual random walk step (with law given by P). We say that the killed process is ``long-lived'' if: (a) P is transient, and there is a non-zero probability of never jumping to the cemetery; or if (b) P is recurrent, and the process started ``from infinity'' has a chance to get close to the origin before being killed. This paper develops potential-theoretic characterizations of those functions k which permit long-lived behavior.