On Newton-like methods of ``bounded deterioration'' using recurrent functions (Q623385): Difference between revisions

This article deals with the following Newton-like method \[ x_{n+1} = x_n - A(x_n)^{-1}F(x_n) \qquad (n = 0,1,2,\dots, \;x_0 \in X), \] where \(F:\;X \to Y\) is a Fréchet-differentiable operator defined on an open subset of a Banach space \(X\) and taking values in a Banach space \(Y\), \(A(x):\;X \to L(X,Y)\) is an approximation to the Fréchet derivative \(F'(x)\). The authors present conditions guaranteeing that the sequence \(x_n\) is well defined and converges to a (unique) solution \(x^*\) to the equation \(F(x) = 0\). These conditions are described in terms of a scalar sequence \[ t_0 = 0, \quad t_1 = \eta, \quad t_{n+2} = t_{n+1} + \frac{K(t_{n+1} - t_n) + 2(b_n - \Delta + K_1t_n)}{2(1 - (d + b_{n+1} - \Delta + \frac{K_2}2 t_{n+1}))}(t_{n+1} - t_n), \] where \(\eta > 0\), \(d \geq 0\), \(K_0 > 0\), \(K > 0\), \(K_1 > 0\), \(b_n \geq 0\), \(\Delta \geq 0\) are some numerical characteristics of \(F\) and \(A\); moreover, the authors state that the usual error estimates \(\|x_{n+1} - x_n\| \leq t_{n+1} - t_n\) and \(\|x_n - x^*\| \leq t^* - t_n\) \ (\(n = 0,1,2,\dots\)) are hold. However, it seems to be strange that among the conditions mentioned above one can find the following inequalities \[ \|A(x_0)^{-1}(A(x_n) - F'(x_n))\| \leq b_n - \Delta + K_1 \sum_{j=1}^n \|x_j - x_{j-1}\|; \] in these inequalities the approximations \(x_n\) (whose existence is proved after this assumption) are presented in the explicit form (before this proof!). At the end of the article three examples are given.