scientific article; zbMATH DE number 1343500

From MaRDI portal
Publication:4267192





Tension spline of order \(k\) is a function that, for given partition \(x_{0}<x_{1}<\ldots <x_{n}\), on each interval \([x_{i},x_{i+1}]\) satisfies the differential equation \((D^{k}-\rho^{2}_{i}D^{k-2})u=0\), where \(\rho_{i}\)'s are prescribed nonnegative real numbers. There are many applications of such splines of order four; higher-order splines might be also applied in numerous problems of numerical analysis but for lack of an appropriate algorithm for their evaluation. The authors present an explicit algorithm doing this. As for the case of polynomial splines, the authors use a B-spline representation for tension splines. They start from the recursive definition of B-splines of order \(k\) [see \textit{P. E. Koch} and \textit{T. Lyche}, in: Curves and surfaces, pp. 255-258 (Academic Press, N.Y.) (1991; Zbl 0736.41013)] where each \(B_{j,k}\) is defined using \(B_{j,k-1}\) and \(B_{j+1,k-1}\) for \(j=m-k+1,\ldots,n\), with given function \(B_{i,2}\) involving hyperbolic sine. Next they represent locally the B-spline \(B_{j,k}\) on the subinterval \((t_{i},t_{i+1})\) as the linear combination of functions of the form \(\varphi_{k}(x,\rho_{i},\Delta_{i})\) , where \(\Delta_{i}=t_{i+1}-t_{i}\) and of a polynomial of the \(k-3\)rd degree. The authors present formulas to calculate the coefficients of this combination and of the polynomial and discuss the problem of computing the functions \(\varphi_{k}\); these functions are expressed by the hyperbolic sine and cosine and may be considered as depending on two parameters: \(t=x/\Delta\) and \(p=\rho_{i} \Delta\) (\(t\) is from interval \([0,1]\) and \(p\) is nonnegative). Computing these function using explicit formula is very sensitive: for \(p\) or \(pt\) small the formula is numerically unstable, while for \(p\) or \(pt\) large (e.g. over 700 for double precision FORTRAN 77 arithmetic) machine overflow occurs. Using for each of these cases appropriate methods of approximation, (e.g. finite sums of Taylor series for subexpressions of the formula for \(\varphi_{k}\)), the authors avoid the unstability and machine overflow.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4267192)