Convergence Properties of the Spline Fit

Let \(f(x) C^2\) have the period unity. Let \(\Delta_n: 0=x_{n,0} < x_{n,1}<\cdots< x_{n,m_n} = 1\) be a division of \([0,1]\) of mesh gauge \(H_n = \max_i (h_{n,i})\), where \(h_{n,i}=x_{n,i} - x_{n,i-1}\). A (periodic) spline function \(y_n(x)\) is a function composed of a cubic polynomial in each of the intervals of \(\Delta_n\) with the requirement that \(y_n(x)\in C^2[0,1]\) and \(y =y_n^{(i)} (1)\) \((i=0,1,2)\). It was observed by Walsh and the authors in a paper reviewed above that there exists a unique periodic spline function \(y_n(x)\) which interpolates \(f(x)\) at the points \(x_{n,i}\). They also discovered the remarkable fact that the same interpolating spline function is uniquely characterized (up to an additive constant) among all periodic spline functions with junction points in the points of \(\Delta_n\) by the property of minimizing \(\int_0^1(f(x) - y_n(x))^2\,dx\). It was also shown there that \(y_n(x)\) and \(y'_n(x)\) converge uniformly to \(f(x)\) and \(f'(x)\), respectively, if \(H_n\to 0\). In the present paper the authors study the more delicate question of the convergence of \(y_n(x) \) to \(f(x)\). Writing \(\lambda_{n,i} = h_{n,i+1}/(h_{n,i} +h_{n, i+1})\) \((i=1,\ldots,m_n - 1)\), \(\Lambda_n = \max_i \vert \lambda_{n,i} - \tfrac12\vert\), they show that \(\lim y_n(x) = f(x) \) uniformly, provided that \(H_n\to 0\) and \(\Lambda_n\to 0\). Two further ``local convergence theorems are established. It would be very interesting to find out to what extent the condition \(\Lambda_n\to 0\) is really necessary in the main theorem quoted.

Recommendations

Cited in

(64)

This page was built for publication: Convergence Properties of the Spline Fit

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5593546)