A computable convex programming characterization of optimal interpolatory quadratic splines with free knots (Q1079925): Difference between revisions

A constructive approach is developed for computing solutions of the following minimization problem for \(n=2\). Given points \(a\leq x_ 1\leq...\leq x_{n+r}\leq b\) and real numbers \(y_ 1,...,y_{n+r}\), let \(F_{\infty}^{(n)}=\{feW_{\infty}^{(n)}[a,b]:\quad f(x_ i)=y_ i,\quad i=1,...,n+r\},\) where \(W_{\infty}^{(n)}[a,b]\) denotes the Sobolev space in \(C^{n-1}[a,b]\). \textit{S. Karlin} [Trans. Am. Math. Soc. 206, 25-66 (1975; Zbl 0303.41011)] showed that there exists a perfect spline of degree n with less than r interior knots (i.e. the n-th derivative of the spline is constant in absolute value and changes sign at each knot) which solves the minimization problem. For \(n=2\) the author gives a characterization of the spline solution in terms of a convex optimization problem. By using this result standard optimization methods can be applied to compute an optimal spline. Two numerical examples are given for illustrating this approach.