Approximating a function by choosing a covering of its domain and k points from its range (Q1103320)

This paper is concerned with an approximation procedure of a real-valued function \(\phi\) for every point \(\theta\) that might be selected from some given set E. Let \(\Sigma_ M\) be a covering of the given set E, with cells indexed by the elements of the set \(M: \Sigma_ M=\{\sigma_ m\subseteq E:\) \(m\in M\}\). To each cell of \(\Sigma\) we assign a real number, called an outcome, from a k-element set \(A\subset {\mathbb{R}}\) of possible outcomes. The authors study the following problem: (P) Given \(\phi\), E, and a particular j-cell covering \(\Sigma\) of E, find a set A of k outcomes that minimizes the error when outcomes are assigned to cells according to \(h_ A\), the closest-to-midvalue function for (\(\phi\),\(\Sigma\),A). The main results of the paper consist of two alternative sufficient conditions on a k-element set A, called equal- error conditions, that guarantee A to be a solution to problem (P). The authors apply these conditions to obtain an explicit solution problem (P) in the special case when E is a rectangle in \({\mathbb{R}}^ n,\Sigma\) is a uniform ``grid'', and \(\phi\) is linear. Finally, an open question is formulated.