Empirical estimates in stochastic optimization and identification (Q1396170)

This book presents a high theoretical study of the use and properties of empirical methods in stochastic optimisation, and related problems that appear in statistics and random processes. It is divided into 5 chapters. The first is an introduction, which mainly consists in the fixation of the notation used and to review basic probability theory aspects. The second one presents some variants of stochastic programs problems in which empirical functions are minimised. The needed convergence in probability results of the estimates is presented and different models are studied in deep. Asymptotic results of empirical distribution based methods are derived in the case of independent observations in a model, under the boundness of the expectations of the extremes, with or without inequality constraints. Some of them establish the convergence in distribution to a certain Gaussian. When the model is measurable stationary ergodic (in the strict sense) random process with discrete time, a similar treatment yields a set of convergence theorems, lemmas and corollaries. The other chapters deal with particular models of practical interest. Chapter 3 considers Parametric Regression models and the consistency of a proposed modification of Least Squares Estimates is proved. The convergence of empirical moment based estimates of the parameters of a linear regression (when dealing with Gaussian and random fields) and of a nonlinear, as well as the nonstationary regression models and the Gaussian regression for a random field represent the rest of the content. Chapter 4 presents a similar study of the periodogram's estimates. They establish that the empirical estimates are strongly convergent and, under some additional conditions, also asymptotically normally distributed. The last chapter tackles the discussion of non-parametric models for stochastic optimisation and identification (estimation). A list of 134 documents covers the references. A considerable number of them are due to the authors. This is a remarkable book that should be of interest both for optimisers and theoretical statisticians.