No label defined (Q4239649)

Let \(F:\mathbb{R}^n\to\mathbb{R}^n\) be a function. The corresponding system equations \(F(x)=0\) can be written in the following form \[ f_i(x_i, x_{q+1})=0,\;i=1,\dots,q; \quad f_{q+1}(x_1, \dots,x_q, x_{q+1})=0. \] The Jacobian matrix can be represented as that of block bordered type \[ J(x)=\left( \begin{matrix} A_1 & \cdots & \cdots & B_1\\ \vdots & \ddots & & \vdots\\ \vdots & & A_q & B_q\\ C_1 & \cdots & C_q & P\end{matrix} \right). \] This paper announces some theoretical extensions on the numerical problem of solving the system \(F(x)=0\). One result is the following. If \(\| J(x^k)\| \leq\eta\) for some \(\eta>0\), \(f_i\in C^1\), \(\nabla f_i\in Lip_1\), \(x^*\) is a solution of \(F(x)=0\) such that all the diagonal blocks \(A_j\), \(P\) are non singular, then the parallel implicit corrected robust Newton (PICRN) algorithm asymptotically behaves as not of Newton standard one. Concerning the non monotone stability it is mentioned that all the directions which are performed by the PICRN algorithm are uniformly gradient-connected. One of the author conclusions is that the non monotone parallel implicit approach seems to be suitable in dealing with block bordered systems.