Wavelet optimal estimations for a density with some additive noises (Q2450939): Difference between revisions

Wavelet estimations for a density with moderately ill-posed additive noises over a Besov ball \(B^s_{r,q} (L)\) (\(q, r \in [1,\infty]\)) and over \(L^p\) (\(1 \leq p \leq \infty\)) risk are investigated in this paper. A lower bound of \(L^p\) risk is given first, and then linear and nonlinear wavelet estimators are defined. The linear estimator is rate optimal for \(r \geq p\). For \(r < p\), the nonlinear estimator is suboptimal (optimal up to a logarithmic factor), and it is adaptive to the indices \(s, r , q\) and \(L\). The optimal wavelet estimators over \(L^2\) risk and \(B^s_{r,q} (\mathbb R)\) with \(1 \leq r \leq 2\) by Fan and Koo, over \(L^\infty\) risk and \(B^s_{\infty,\infty} (\mathbb R)\) by Lounici and Nickl both are special cases of this work. That is to say, this study is motivated by the previous researchers' work but generalizes their results considerably.