A parallel projection method for overdetermined nonlinear systems of equations (Q1208707)

The paper studies an issue of finding an approximate solution to a system \(F(x) = 0\), where \(F: \mathbb{R}^ n \to \mathbb{R}^ m\), \(m \geq n\), is a nonlinear map. Assume that the components of \(F\) are divided into \(N\) blocks and \(J_ i(x) = F_ i'(x)\) be the Jacobian matrix of the \(i\)-th block. If \(V_ i(x) = \{y \in \mathbb{R}^ n| y = {\underset {z}{\text{Argmin}}}\| F_ i(x) + J_ i(x)(z-x)\|_ 2\}\), \(\text{dist}(y,V_ i(x)) = \inf\{\| y-z\|_ 2| z \in V_ i(x)\}\), \(\lambda_ 1,\dots,\lambda_ N > 0\) be such that \(\sum^ N_{i=1}\lambda_ i = 1\) then it is shown that the problem of solving the overdetermined system can be reduced to a minimization problem \[ \varphi(x):{\underset {y}{\text{Minimize}}} \sum^ N_{i=1}\lambda_ i[\text{dist}(y_ ,V_ i(x))]^ 2 \] and a solution is called a weighted least square distance solution (WLSD solution). Algorithms for computing a WLSD solution are presented and numerical examples as well as a comparison to results from a Gauss-Newton method are discussed.