Slope tests for Newton-type methods (Q1207820)

Suppose we are given a function \(f: \mathbb{R}^ n\to\mathbb{R}^ n\), with \(f\in C^ 1(D)\), \(D\subseteq\mathbb{R}^ n\). \textit{M. C. Pandian} [SIAM J. Numer. Anal. 22, 779-791 (1985; Zbl 0586.65041)] gave existence and uniqueness tests for a zero \(x^*\) of \(f\) in a rectangular subset \(X\subset D\). The uniqueness test guarantees the convergence of the sequence \(x_ k\) generated from the simplified Newton iteration \(x_{k+1}=x_ k-A^{- 1}f(x_ k)\), \(k\geq 0\). In addition, componentwise error estimates of the \(x_ k\) are given. Pandian's paper makes use of the usual derivative form of the mean-value theorem. The current authors replace the derivatives of \(f\) by the slope \(f[x,z]\), introduced by \textit{R.Krawczyk} and \textit{A. Neumaier} [SIAM J. Numer. Anal. 22, 604-616 (1985; Zbl 0572.65007)] (in an interval analysis context) and generally defined by \(f(z)-f(x)=f[x,z](x-z)\). It is shown that Pandian's existence test can be weakened and that the sharper pointwise error bound on \(x^*\) may be obtained. Two numerical examples are used to illustrate their results.