Best error bounds for quartic spline interpolation

From MaRDI portal





The existence and uniqueness of a certain interpolating quartic spline function is proved and error estimates are given for smooth functions involving their fifth order derivatives. In detail, let \(S_{k,4}\) denote the class of quartic spline functions s, i.e. \(s\in C^ 2[0,1]\) and \(s_{| [x_ i,x_{i+1}]}\) be a quartic polynomial, \(i=0,1,...,k-1\), for a given partition \(0=x_ 0<x_ 1<...<x_{k- 1}<x_ k=1\). The two main results of the paper are Theorem 1: Given arbitrary numbers \(f(x_ i)\), \(i=0,1,...,k\), \(f(z_ i)\), \(i=1,2,...,k\) \((2z_ i=x_ i+x_{i-1})\), \(f'(x_ 0)\), \(f'(x_ k)\), there exists a unique \(s\in S_{k,4}\) such that \(\quad s(x_ i)=f(x_ i),\quad i=0,1,...,k,\quad s'(x_ 0)=f'(x_ 0),s'(x_ k)=f'(x_ k);\) \(s(z_ i)=f(z_ i),\quad i=1,2,...,k.\) Theorem 2: Let \(f\in C^ 5[0,1]\). Let s be the spline function of Theorem 1. Then it holds that \[ | f(x)- s(x)| \leq \frac{C_ 0h^ 5}{5!}\max_{0\leq x\leq 1}| f^{(5)}(x)|,\quad where\quad C_ 0=(1/30+\sqrt{30}/3)(1/4- 1/\sqrt{30})^{1/2}=\max_{0\leq t\leq 1}| c(t)| \quad and\quad c(t)\quad is\quad given\quad as\quad \] \[ c(t)=(3t^ 2(1-2t)(1-t)^ 2+t(1-t)(1-2t))/6. \] Furthermore \[ | f'(x_ i)-s'(x_ i)| \leq \frac{h^ 4}{6!}\max_{0\leq x\leq 1}| f^{(5)}(\quad x)|,\quad i=1,2,...,k-1, \] which is the best possible estimate. Also \(C_ 0\) cannot be improved for an equally spaced partition. Finally, \(| f'(x)-s'(x)| \leq c_ 1\frac{h^ 4}{6!}\max_{0\leq x\leq 1}| f^{(5)}\quad (x)|.\)As an additional result, an error estimate is stated without proof for the approximation of continuous functions by suitably modified interpolating quartic splines involving the ordinary modulus of smoothness.











This page was built for publication: Best error bounds for quartic spline interpolation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1263025)