Greedy algorithm with gaps (Q1685952)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Greedy algorithm with gaps |
scientific article |
Statements
Greedy algorithm with gaps (English)
0 references
20 December 2017
0 references
The aim of the present paper is to extend some results on greedy approximation algorithms to greedy algorithms allowing gaps in the approximating sequences. Let \(X\) be a Banach space and \((e_i,e^*_i)\in X\times X^*,\, i\in I,\) a biorthogonal system such that span\(\{e_i\}\) is dense in \(X\), span\(\{e^*_i\}\) is \(w^*\)-dense in \(X^*\) and \(\sup_i\max\{\|e_i\|,\|e^*_i\|\}<\infty\). For a finite subset \(A\) of \(I\) and \(x\in X\) put \(Ax=\sum_{i\in A}\langle e_i^*,x\rangle e_i\,.\) Let \( 0 < t\leq 1\). A set \(A\subset I\) is called \(t\)-greedy (greedy if \(t=1\)) for \(x\in X\) if \(\inf_{i\in A}|\langle e_i^*,x\rangle| \geq t\sup_{i\in I\setminus A}|\langle e_i^*,x\rangle|.\) A \(t\)-greedy projection of \(x\) is \(\mathbf{G}^t_m(x)=Ax\) where \(A\) is a \(t\)-greedy subset of \(I\) of cardinality \(m\). For a sequence \( \mathbf{n}=(n_1 < n_2 < \dots)\) of natural numbers one says that the system \((e_i)\) is \(\mathbf{n}\)-quasi-greedy with parameter \(t\) (\(\mathbf{n}\)-\(t\)-QG) if for every \(x\in X,\, \lim_k \mathbf{G}^t_{n_k}x=x\) for any choice of greedy projections \(\mathbf{G}^t_{n_k}\). If \(\mathbf{n}=\mathbb{N}\) one says that \((e_i)\) is \(t\)-quasi-greedy (quasi-greedy for \(t=1\)). The authors characterizes the \(\mathbf{n}\)-\(t\)-QG bases as those bases \(\{(e_i,e^*_i)\}\) for which there exists a constant \(C\geq 0\) such that for all \(x\in X\), \(\|\mathbf{G}^t_n(x)\|\leq C\|x\|\) for any \(n \in \mathbf{n}\) and any \(t\)-greedy projection \(\mathbf{G}^t_n\) (Theorem 2.1). He gives examples of \(\mathbf{n}\)-QG-bases which are not QG. He also shows that that any basis which is \(\mathbf{n}\)-QG with constant 1 must be 1-suppression unconditional (meaning that \(\|\sum_{i\in B}\alpha_i e_i\|\leq \|\sum_{i\in A}\alpha_i e_i\|,\) whenever \(B\subset A\) are finite subsets of \(\mathbb{N}\)), extending a result of \textit{F. Albiac} and \textit{J. L. Ansorena} [J. Approx. Theory 201, 7--12 (2016; Zbl 1335.46010)]. In Section 5, one shows that certain classical bases -- as the Haar basis in \(L_1\), the trigonometric basis in a weighted \(L_p\) space -- are not \(\mathbf{n}\)-QG, for any sequence \(\mathbf{n}\).
0 references
greedy basis
0 references
quasi-greedy basis
0 references
greedy algorithm
0 references
Haar basis
0 references
0 references