Computational aspects of Kolmogorov's superposition theorem (Q1338265)

According to a theorem of \textit{D. A. Sprecher} [Neural Netw. 6, No. 8, 1089-1094 (1993; Zbl 0788.26010)] an arbitrary continuous function \(f: E^ n\to \mathbb{R}\) (\(E= [0, 1]\)) has a representation \(f(x_ 1,\dots, x_ n)= \sum^{2n}_{q= 0} F_ q\left( \sum^ n_{p= 1} z_ p\cdot g(x_ p+ q\cdot e_ n)\right)\) in which \(g\) is a predetermined continuous function (independent of the number of variables \(n\)) and \(e_ n\) is a suitable constant. The paper is devoted to computational aspects of the implementation of the functions \(g\) and \(F_ q\). The function \(g\) is defined as a limit of iterated contraction mappings. \(F_ q\) are the uniform limits of sequences of computable functions.