Asymptotics of minimax mean-square risk of statistical estimators of spectral density in the space \(L_ 2\) (Q1820536)

Let \(X_ t\), \(t=0,\pm 1,..\). be a real wide-sense stationary sequence with zero mean and unknown spectral density f(\(\lambda)\). It is required to estimate \(f(\lambda)\) on the basis of the observed realization \(X_ 1,...,X_ N\) and a priori information \(X\in {\mathcal X}\), where \({\mathcal X}\) is some subset of real random sequences which are wide-sense stationary with zero mean and spectral density from \(L_ 2(-\pi,\pi)\). The author studies the asymptotics of minimax risks \[ \Delta^ 2_ N=\inf_{\hat f}\sup_{X\in {\mathcal X}}E_ X\| \hat f-f\|^ 2_{L_ 2} \] and \[ \delta^ 2_ N=\inf_{\hat f}\sup_{X\in {\mathcal X}}E_ X\| \hat f-f\|^ 2_{L_ 2}/\| f\|^ 2_{L_ 2}, \] where \(E_ X\) is the mean with respect to the measure corresponding to \(X_ t\) and inf is taken over all estimators \(\hat f: [-\pi,\pi] \times {\mathbb{R}}^ N\to {\mathbb{R}}\), as \(N\to \infty\) in the case of two natural forms of the set \({\mathcal X}\), defined by restrictions on f(\(\lambda)\) and the fourth cumulants of \(X_ t\).