Approximation order to a function in \(\overline C({\mathbb R})\) by superposition of a sigmoidal function (Q1609519): Difference between revisions

The authors investigate the approximation error to a continuous function defined on the whole real line by superposition of a sigmoidal function. Definition. A sigmoidal function is a function \(\sigma: \mathbb{R} \to \mathbb{R}\) such that \[ \lim_{x\to\infty} \sigma(x)=1 \quad \text{and} \quad \lim_{x\to-\infty} \sigma(x)=0. \] Furthermore, there are introduced the following notations \[ \overline C(\mathbb{R})= \{f:\mathbb{R}\to \mathbb{R}\mid f\text{ is continuous and }\lim_{x\to\infty} f(x)\text{ and }\lim_{x\to-\infty} f(x)\text{ exist}\}; \] \[ \Phi_{n,\sigma} =\left\{\sum^n_{i=1} a_i\sigma(b_ix+c_i): a_i,b_i,c_i\in \mathbb{R} \right\} \] and \[ E_{n,\sigma, \mathbb{R}}^\Phi(f)= E^\Phi_{n,\sigma} (f)= \inf_{g\in \Phi_{n,\sigma}} \|f-g\|_{\mathbb{R}, \infty},\;f\in\overline C (\mathbb{R}). \] Fox the general reference see \textit{E. W. Cheney} and \textit{W. Light} [A course in approximation theory (Brooks/Cole, New York) (1999)]. In the main theorem, after some preliminary results, the authors compute the approximation error \(E^\Phi_{2 n,\sigma} (f)\) to a function \(f\) in \(\overline C(\mathbb{R})\) by \(3n\) superpositions of a sigmoidal function: Theorem 5. Let \(\sigma\) be a bounded sigmoidal function on \(\mathbb{R}\). If \(f\in\overline C(\mathbb{R})\) then for each positive integer \(n\), \[ E^\Phi_{3n,\sigma} (f)\leq\|\sigma\|\omega \left(f,{2A_1 \over n+1} \right)+ {c\over n}, \] where the constants \(c\) and \(A_1\) depend on \(f\) and \(\sigma\), \(\|\sigma \|=\sup_{x\in\mathbb{R}} |\sigma (x)|\) and \(\omega (f,\cdot)\) is the modulus of continuity of \(f\). Moreover the authors prove Corollary 6. Let \(\sigma\) be a bounded sigmoidal function on \(\mathbb{R}\). If \(f\in \overline C(\mathbb{R})\) satisfies the Lipschitz condition then for each positive integer \(n\), \(E^\Phi_{3n,\sigma} (f)=O ({1\over n})\).