×

Approximation on anisotropic Besov classes with mixed norms by standard information. (English) Zbl 1078.41027

Let \(\mathbb{T}^d\) be the \(d\)-dimensional torus, let \(X\) be a normed linear space of real functions defined on \(\mathbb{T}^d\), and let \(C(\mathbb{T}^d)\) be the space of all real, continuous and periodic functions on \(\mathbb{T}^d\). In order to compare different approximation algorithms that are applicable to the functions belonging to a given set \(K\subset C(\mathbb{T}^d)\cap X\) and are based on standard information, i.e. the values of the involved functions at \(n\) points chosen from \(\mathbb{T}^d\), the \(n\)th minimum intrinsic error \(e_m\) and the \(n\)th minimum linear intrinsic error \(e^L_n\) are relevant.
In the present paper the authors estimate these errors in the case when \(X:= L^{{\mathbf s}}_{{\mathbf q}}(\mathbb{T}^d)\) and \(K\) is the unit ball of the anisotropic Besov space \(B^{{\mathbf r}}_{{\mathbf p}\theta}(\mathbb{T}^d)\), where \({\mathbf s}: =(s_1,\dots, s_d)\), \({\mathbf q}:= (q_1,\dots, q_d)\), \({\mathbf r}= (r_1,\dots, r_d)\) and \({\mathbf p}:= (p_1,\dots, p_d)\) are vectors satisfying \(s_i\in \mathbb{Z}_+\), \(1\leq q_i\), \(1< r_i\), \(1\leq p_i\) for \(i:= 1,\dots, d\), and \(\theta\) is a number satisfying \(1\leq\theta\leq\infty\). By using the Dirichlet approximation algorithm they reveal cases (depending on \({\mathbf p}\) and \({\mathbf q}\)) when the asymptotic equalities \[ e_n\asymp e^L_n\asymp n^{-g(r)(1-\beta)}\quad \text{and} \quad e_n\asymp e^L_n\asymp n^{-g(r)(1-\beta)+ g(r)\gamma} \] hold, where \[ \begin{aligned} g(r)&:= 1/\Biggl(\sum^d_{i=1} 1/r_i\Biggr),\\ \beta&:= \max(s_1/r_1,\dots, s_d/r_d),\;\gamma:= \sum^d_{i=1} (1/p_i- 1/q_i)1/r_i. \end{aligned} \]

MSC:

41A65 Abstract approximation theory (approximation in normed linear spaces and other abstract spaces)
41A46 Approximation by arbitrary nonlinear expressions; widths and entropy
30E10 Approximation in the complex plane
65K10 Numerical optimization and variational techniques
PDFBibTeX XMLCite
Full Text: DOI

References:

[1] Barron, A. R., Universal approximation bounds for superposition of \(n\) sigmoidal functions, IEEE Trans. Inform. Theory, 39, 930-945 (1993) · Zbl 0818.68126
[2] DeVore, R. A., Nonlinear approximation, Acta Numerica, 7, 51-150 (1998) · Zbl 0931.65007
[3] DeVore, R. A.; Temlyakov, V. N., Some remarks on greedy algorithms, Adv. Comp. Math., 5, 173-187 (1996) · Zbl 0857.65016
[4] Donahue, M.; Gurvits, L.; Darken, C.; Sontag, E., Rate of convex approximation in non-Hilbert spaces, Constr. Approx., 13, 187-220 (1997) · Zbl 0876.41016
[5] Jones, L., On a conjecture of Huber concerning the convergence of projection pursuit regression, Ann. Statist., 15, 880-882 (1987) · Zbl 0664.62061
[6] Jones, L., A simple lemma on greedy approximation in Hilbert space and convergence rates for projection pursuit regression and neural network training, Ann. Statist., 20, 608-613 (1992) · Zbl 0746.62060
[7] D. Leviatan, V.N. Temlyakov, Simultaneous approximation by greedy algorithms, Adv. Comp. Math., to appear.; D. Leviatan, V.N. Temlyakov, Simultaneous approximation by greedy algorithms, Adv. Comp. Math., to appear. · Zbl 1096.41026
[8] Lutoborski, A.; Temlyakov, V. N., Vector greedy algorithms, J. Complexity, 19, 458-473 (2003) · Zbl 1234.41024
[9] Gilles Pisier, Martingales with values in uniformly convex spaces, Israel J. Math., 20, 249-326 (1975) · Zbl 0344.46030
[10] Temlyakov, V. N., Weak greedy algorithms, Adv. Comp. Math., 12, 213-227 (2000) · Zbl 0964.65009
[11] Temlyakov, V. N., Greedy algorithms in Banach spaces, Adv. Comp. Math., 14, 277-292 (2001) · Zbl 0988.41022
[12] Temlyakov, V. N., Nonlinear methods of approximation, Found. Comput. Math., 3, 33-107 (2003) · Zbl 1039.41012
[13] Temlyakov, V. N., A remark on simultaneous greedy approximation, East J. Approx., 10, 17-25 (2004) · Zbl 1113.41026
This reference list is based on information provided by the publisher or from digital mathematics libraries. Its items are heuristically matched to zbMATH identifiers and may contain data conversion errors. In some cases that data have been complemented/enhanced by data from zbMATH Open. This attempts to reflect the references listed in the original paper as accurately as possible without claiming completeness or a perfect matching.