Asymptotically minimax separation of two simple hypotheses (Q1280862)

We consider the problem of asymptotic separation of two simple hypotheses when a general stochastic process is observed. Optimal statistical criteria can be constructed by two strategies: the strategy of the most powerful criteria, and that of minimax criteria. The first strategy involves minimizing the probability of type II error when the level of probability of type I error is given. The second one involves minimizing the maximum of the probabilities of type I error and type II error. At the present time, asymptotic properties of the most powerful criteria are investigated for general statistical models as well as for particular ones. Asymptotic properties of minimax criteria are only investigated for observations which are independent, identically distributed random variables. Here we give conditions under which the asymptotic behaviour can be defined for maximal probabilities of errors of minimax criterion for general statistical models when oberservations are of an arbitrary nature. In the first section the main results are proved. In the second section, the representation of the Hellinger integral is given and, as examples, multivariate point processes, nonhomogeneous Poisson processes, renewal processes and processes of diffusion type are analyzed.