Superlinearly convergent exact penalty projected structured hessian updating schemes for constrained nonlinear least squares: asymptotic analysis (Q2445078)

The paper is concerned with the problem \(\min \phi(x)=F(x)^T F(x)\) subject to \(c_i(x)=0\), \(i \in M_1\) and \(c_j(x) \leq 0\), \(j \in M_2\), where \(F(x)=[f_1(x),\dots,f_l(x)]^T\), \(x \in R^n\). The functions \(f_s\), \(c_i\) and \(c_j\) are assumed to be twice continuously differentiable. The proposed iterative method uses the penalty function \(\psi(x,\mu)=\mu \phi(x)+\sum_{i\in M_1} | c_i(x)|-\sum_{j\in M_2} \min\{ 0, c_j(x) \}\) with an adaptive choice of the penalty parameter \(\mu=\mu_k>0\). The authors establish local \(Q\)--superlinear convergence of the method.