Discrete weighted cubic splines (Q478201)

The paper presents methods for spline interpolation preserving the shape of the data (monotonicity and convexity) by using discrete weighted cubic splines. One considers the data \((x_i,f_i),\, i=0,1,\dots,N+1,\) \(a=x_0<x_1<\dots<x_{N+1}=b,\) and a function \(w:[a,b]\to\mathbb R,\, 0<m\leq w(x)\leq M\) (weight function) that is piecewise constant on subdivisions: \(w(x)=w_i,\, x\in [x_i,x_{i+1}],\, i=0,1,\dots,N.\) One denotes by \(f[x_i,x_{i+1}]=(f_{i+1}-f_i)/h_i,\, h_i=x_{i+1}-x_i,\, i=0,\dots,N,\) the divided differences. The data are called monotonically increasing if \(f[x_i,x_{i+1}]\geq 0,\, i=0,\dots,N,\) and convex if \(f[x_i,x_{i+1}]-f[x_{i-1},x_{i}]\geq 0,\, i=1,\dots,N.\) The author defines the weighted cubic spline \(S\) by the following conditions: (i)\; \(S\) is a cubic polynomial on each interval \( [x_i,x_{i+1}],\, S(x)=S_i(x)=\sum_{j=0}^3a_{i,j}(x-x_i)^j,\, i=0,1,\dots,N; \) (ii)\; the first and respectively the second central divided differences of \(S\) with respect to the points \(x+j\tau_{i-1},\, x+j\tau_{i},\, j=-1,0,1, \) (for some \(\tau_i>0\)) coincide. For \(\tau_i=0\) one obtains cubic splines of class \(C^1.\) In Theorem 1, imposing the boundary conditions \(S'(x_0)=f'_0,\, S'(x_{N+1})=f'_{N+1}\) to a function \(f\in C^4[a,b]\), one obtains error estimations for \(\|S^{(r)}(x)-f^{(r)}(x)\|_\infty,\, r=0,1.\) In Theorems 3 and 4 one obtains sufficient conditions for preserving the monotonicity, respectively the convexity, of the data \(\{f_i\}.\) One gives two algorithms with automatic selection of the shape control parameters. Discrete weighted cubic \(B\)-splines, control point approximation and graphical examples are considered as well.