No label defined (Q4378697)

Let \(U\subset\mathbb{R}^n\) and let \(1\leq k<n\) be fixed. The author defines a pattern as a finite union of subsets \(\Phi_i\) of \(U\), for each of which there exists a \(C^1\) function \(g_i:U\to\mathbb{R}^k\) so that \(\Phi_i\subset g_i^{-1}(0)\), and such that all points in \(\Phi_i\) are regular points of \(g_i\). Regularity of \(a\in U\) with respect to \(f:U\to\mathbb{R}^k\), is defined as the Jacobian \(J_f(a)\) having maximal rank, that is, \(k\). Also, given a point \(a\in U\), and a pattern \(\Phi=f^{-1}(0)\), such that \(a\) is a regular point of \(f\), the author defines the distance \(d(a,\Phi)\) as the Euclidean norm of \(J_f(a)^+f(a)\), where \(A^+\) denotes the right inverse of the matrix \(A\). Now the distance between a finite set of distinct (regular) points (points of surveillance) in \(U\) and the pattern \(\Phi=f^{-1}(0)\) is defined as the sum of squares of the distances to the individual points. While the author is interested in minimizing that distance in various situations and for that purpose it does not matter whether one takes the above sum of its square root, the reviewer suggests that to take the square root in order for the distance to satisfy the basic properties of distance. The author suggests that minimizing the distance may have applications to pattern recognition however he does not elaborate or give an example. The paper contains some algorithm for the minimization problem and a Newton-Raphson type result.