Knot insertion and totally positive systems (Q1567422): Difference between revisions

This paper deals with the construction of a knot insertion procedure in a given space of functions, using techniques related to total positivity. Let \(T_n=\{u_0,\dots,u_n\}\) be a basis of the space \({\mathcal U}\), where each \(u_i\) is a real function on the interval \(I\). \(T_n\) is normalized in the sense that \(u_0(t)+\cdots+u_n(t)=1\), \(\forall t\in I\). Shape preserving representation of curves by means of control polygons is associated with normalized totally positive (TP) bases. The system \(T_n\) is TP if all its collocation matrices \((u_j(t_i))\) are TP. A system \(T_n\) is a normalized B-basis if it is a normalized TP basis with optimal shape preserving properties among all normalized TP bases of the space. A B-algorithm is a corner cutting algorithm which satisfies evaluation and subdivision properties. The connection between symmetric B-algorithms and knot insertion algorithms is investigated in this paper. One of the main results is the following theorem, which expresses that elementary knot insertions are always associated to corner cutting algorithms. Theorem 4.9 Let \({\mathcal U}={\mathcal U}^{n+1}\) be an \((n+1)\)-dimensional space of functions on \(I\) with a normalized totally positive basis and \(t_0\in\text{Int}(I)\) whose potential knot multiplicity in \({\mathcal U}\) is \(k\). Then the B-algorithm corresponding to \(t_0\) provides a knot insertion algorithm if and only if we can perform \(k\) consecutive elementary knot insertions with \(t_0\). Symmetric B-algorithms are precisely the B-algorithms providing knot insertion. Theorem 5.1 Let \(T_n\) be the normalized B-basis of a vector space \({\mathcal U}\) of functions defined on \(I\). Let \(t_0\in \text{Int} (I)\) be a \(k\)-admissible parameter in \({\mathcal U}\). Then the B-algorithm is symmetric if and only if we can perform \(k\) consecutive knot insertions with \(t_0\) and \(t_0\) is a \((k-p)\)-admissible parameter in the corresponding spaces \({\mathcal U}^{n +p+1}\), \(p=1,\dots, k-1\). Examples of spaces with symmetric B-algorithms are given in the last section.