Lebesgue-type inequalities for quasi-greedy bases (Q387546): Difference between revisions

If \(\mathbb{X}\) is a Banach space and \(\mathbb{B}=\{{\mathbf e}_j\}_{j=1}^{\infty}\) is a normalized basis of \(\mathbb{X}\), then for \(N\in\mathbb{N}\) denote by \(\Sigma_N\) the set of elements of \(\mathbb{X}\) with at most \(N\) non-null coefficients in the basis representation. Recall that a thresholding greedy operator of order \(N\) is any operator \(G_N:\mathbb{X}\to\Sigma_N\) which assigns to any element \(x=\sum_{j=1}^{\infty}a_j{\mathbf e}_j\) the value \(G_N(x)=\sum_{\lambda\in\Lambda(x)}a_{\lambda}{\mathbf e}_{\lambda}\), where \(\Lambda(x)\) has cardinality \(N\) and \(\min_{\lambda\in\Lambda(x)}|a_{\lambda}|\geq\max _{\lambda\not\in\Lambda(x)}|a_{\lambda}|\). The basis \(\mathbb{B}\) is named quasi-greedy if there is a constant \(\tilde{K}\), such that \(\sup \| G_N(x)\|\leq\tilde{K}\| x\|\) for all \(x\in \mathbb{X}\), where the sup is taken over all \(N\) and all thresholding operators of order \(N\). The basis is named greedy if \(C_N=O(1)\). The Lebesgue-type inequality for the greedy algorithm \(G_N\) is given by \(\| x-G_N(x)\|\leq C_N\sigma_N(x)\) for all \(x\in\mathbb{X}\), where \(\sigma_N(x)=\inf_{y\in\Sigma_N}\| x-y\|\) and \(C_N\) is the smallest possible constant. The first main result of the paper asserts that if the basis \(\mathbb{B}\) is quasi-greedy, then the constant \(C_N\) is equivalent when \(N\to\infty\) with \(\max\{k_N, \mu(N)\}\), where the quantities \(k_N\) and \(\mu(N)\) measure the conditionality and democracy of the basis, respectively. This result is a generalization and a precised quantitative version of the result of \textit{S.~V.~Konyagin} and \textit{V.~N.~Temlyakov} in [East J. Approx.\ 5, No. 3, 365--379 (1999; Zbl 1084.46509)] concerning greedy bases. Also in the paper there are given other two quantitative estimates for the approximation with finite combination of elements of a basis. Then, more detailed discussions about the order of growth of the given quantities and the optimality of the estimates are made using particular Banach spaces.