On the power of adaptive information for functions with singularities

We study from a probabilistic viewpoint the problem of locating singularities of functions using function evaluations. We show that, under the assumption of a Wiener-like probability distribution on the class of singular functions, an adaptive algorithm can locate a singular point accurately with only a small probability of failure. As an application, we show that an integration algorithm that adaptively locates a singular point is probabilistically superior to nonadaptive algorithms.