On the convergence of parallel chaotic nonlinear multisplitting Newton-type methods (Q1360176)

The aim of this paper is to solve large, sparse nonlinear problems \(F(x)= 0\), exploiting the capabilities of a parallel computer. To this end the problem is split into a number \(\alpha\) of problems of the form \(f^{(i)}(x;y)=0\) which are to be solved for \(y\) with given \(x\); the functions \(f^{(i)}\) must satisfy \(f^{(i)}(x;x)= F(x)\). The computations for \(i=1,\dots,\alpha\) can be done independently (parallellism) and the obtained solutions combined with approximate weights to obtain a better approximation in an iterative scheme to solve \(F(x)= 0\). The novelty in the present paper (chaotic\dots) is to let the number of Newton iterations in the solution of \(f^{(i)}(x;y)=0\) depend on \(i\) and on the Newton step. So it extends an idea from the linear theory to the nonlinear theory. The paper mainly discusses the local convergence results. No numerical application is presented.