On asymptotic behavior of the prediction error for a class of deterministic stationary sequences (Q2095127)

One of the main problems in prediction theory of second-order stationary processes is to describe the asymptotic behavior of the prediction error. This behavior depends on the regularity nature (deterministic or nondeterministic) of the observed process. From the prediction point of view it is natural to distinguish the class of processes for which there is an error-free prediction by the entire infinite past. Such processes are called deterministic or singular. In this paper, the authors focus on deterministic processes. This case is not only of theoretical interest, but is also important from the point of view of applications. For example, situations of this type arise in Neumann's theoretical model of storm-generated ocean waves. Describing the results of the paper more precisely, the authors study the prediction problem for deterministic stationary processes \(X(t)\) possessing spectral density \(f\). They describe the asymptotic behavior of the best linear mean squared prediction error \(\sigma^2_n(f)\) in predicting of \(X(0)\) given \(X(t), -n \le t \le -1\), as \(n\) goes to infinity. They consider a class of spectral densities of the form \(f = f_dg\), where \(f_d\) is the spectral density of a deterministic process that has a very high order contact with zero due to which the Szegő condition is violated, while \(g\) is a nonnegative function that can have arbitrary power type singularities. It is established that for spectral densities \(f\) from this class the prediction error \(\sigma^2_n(f)\) behaves like a power as \(n\rightarrow \infty\). Three examples of spectral densities illustrate the obtained results.