A modification of krawczyk's algorithm

From MaRDI portal





In order to enclose a simple zero \(\rho\) of a continuously differentiable function \(f:\mathbb{R}^n \to\mathbb{R}^n\) by a tight compact interval vector the authors consider the iteration \[ \begin{cases} KM(X_k) & =x_k-Y_k f(x_k)+ {1\over \gamma} \bigl(I-Y_kF'(X_k) \bigr) (X_k-x_k) \\ X_{k+1} & = X_k\cap KM (X_k) \end{cases} \quad k=0,1, \dots \] where \(x_k\) is the midpoint of the interval vector \(X_k\), \(X_0\) is given, \(Y_k\) is some preconditioning real nonsingular \(n \times n\) matrix and \(F'(X_k)\) is an inclusion monotone interval extension of the Fréchet derivative of \(f\) at \(X_k\). The parameter \(\gamma\) is newly introduced; its choice depends on \(k\). For \(\gamma=1\) the method contains the well-known Krawczyk method. By choosing \(\gamma\) such that \[ {2\over \gamma} \biggl\| \bigl| I-Y_kF'(X_k) \bigr| \biggr \|_\infty \leq q, \] \[ q\in (0,1) \text{ being a preselected real number}, \] the authors show that \(\rho\in X_0\) implies \(\rho\in X_k \subseteq X_{k-1} \subseteq \dots \subseteq X_0\) and \(\lim_{k \to \infty} X_k= \rho\). (Some assumptions seem to be missing the paper in order to verify this result; note that \(KM(X_0) \approx x_0- f(x_0)\) for \(Y_0=I\) and for \(\gamma\) sufficiently large which can yield to \(\rho \notin X_1\).) Conditions for \(\gamma\) are presented which guarantee the quadratic convergence of the method. The one-dimensional case is handled separately. Numerical examples conclude the paper.











This page was built for publication: A modification of krawczyk's algorithm

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4375411)