Analysis of a smoothing Newton method for second-order cone complementarity problem (Q1034984)

This paper considers the second-order cone complementarity problem which consists in finding \(z\in\mathbb{R}^n\) such that \[ \langle f(z), z\rangle= 0\quad\text{and}\quad f(z)\in K,\quad z\in K, \] where \(\langle\cdot,\cdot\rangle\) stands for the Euclidean inner product, \(f: \mathbb{R}^n\to \mathbb{R}^n\) is a \(C^1\)-function, and \(K\) is the Cartesian product of second-order cones, that is, \(K= K^{n_1}\times K^{n_2}\times\cdots\times K^{n_m}\) with \(n_1+ n_2+\cdots+ n_m= n\). The \(n_i\)-dimensional second-order cone \(K^{n_i}\) is defined by \[ K^{n_i}= \{(z_1,z_2)\in \mathbb{R}\times \mathbb{R}^{n_i- 1}: z_1\geq\| z_2\|\}. \] The study is developed under \(P_0\)-property. By introducing a smoothing parameter into the Fischer-Burmeister function, a smoothing Newton method for the second-order cone complementarity problem is presented. The proposed algorithm solves only a linear system of equations and performs only one line search at each iteration, and no restriction on its starting point is imposed. In addition, the algorithm has global convergence, and under non-singularity, the locally quadratic convergence without strict complementarity is established.