No label defined (Q4946111)

The aim of the paper is the presentation of the concept of approximator for a nonsmooth mapping, i.e., a first order local approximation of the mapping, which depends on the current point and the directions without any assumptions of homogeneity. This is done in order to overcome the tight connection, shown by the directional derivatives, between continuity with respect to the point of approximation and linearity with respect to the direction. Let \(X\) and \(Y\) be Banach spaces. Given a mapping \(f: X \to Y\) the authors say that \(f^{\odot}: X \times X \to Y\) is an approximator for \(f\) at the point \(x\) in the direction \(v\) if \(f(x+v)-f(x)=f^{\odot}(x,v)+o(v)\), where \(\lim_{t \to 0+} t^{-1}o(tv)=0\). In the case when \(Y\) is a Banach lattice it is proved that for a hypodifferentiable mapping \(f\) in the sence of Demyanov and Rubinov there exists an approximator \(f^{\odot}\) which is continuous as a function of \((x,v)\). The authors extend for the approximators classical theorems on the differentiation of sums, products and composite functions, the mean value theorem and Newton-Leibniz formula. They present applications of the developed calculus to the fixed point theory, nonsmooth optimization theory and Newton's method for nonsmooth operator equations. Specifically, the following Newton-type method for the equation \(f(x)=0\) is proposed: \(x_{k+1}=x_k+u_k\), where \(u_k\) solves the equation \(f(x_k)+f^{\odot}(x_k,u)=0\). Let \(\|f(x_0)\|\leq \mu\) and for all \(y \in Y\), \(\|y-x_0\|\leq \mu\) the equation \(f^{\odot}(x,u)=y\) possesses a solution \(u\) such that \(\|u\|\leq L \|y\|\) \((\mu, L>0)\). Suppose that there exist \(\gamma, \delta>0\) such that \(\|f(x+u)-f(x)-f^{\odot}(x,u) \|\leq \gamma \|u\|^2\) \(\forall x \in X\), \(\|x-x_0\|\leq \delta\). Then \(\{x_n \}\) converges locally to the solution of the original equation. For the entire collection see [Zbl 0927.00033].