Regularized smoothing approximations to vertical nonlinear complementarity problems (Q1284038): Difference between revisions

This article is concerned with the vertical nonlinear complementarity problem \(\text{VNCP}(F^1,...,F^m)\), in short: \(VNCP\), of finding an \(x\in\mathbb{R}^n\) such that componentwise \[ R(x):= \min\{x,F^1(x), \dots, F^m(x)\}=0. \] Motivation comes from algorithmic applications of the mountain pass theorem (for nonlinear complementarity problems) on smoothing Newton methods (proof of global convergence), and from a corresponding interior point method in some extended linear case. The authors approximate VNCP with the help of Tikhonov regularization, replacing \(R(x)\) by \(R(t,x):= \min\{x,F^1 (x)+tx\), \(F^2(x)+tx, \dots,F^m(x)+ tx\}\) \((t\in (0,\overline t]\), following the solution trajector \(x(t)\) for \(t\to 0\) under mild conditions. They ask \(\{F^1, \dots,F^m\}\) to have the \(P_0\)-property, meaning a generalized componentwise monotonicity. Then, \(x(\cdot)\) is bounded, differentiable and has an accumulation point lying in the solution set of VNCP. Moreover, the approximation concept of \(\varepsilon\)-solutions is introduced, and any zero of some auxiliary function is shown to represent a solution of VNCP. Finally, the authors guarantee existence of the solution under the uniform \(P\)-property on quadratic growth, and they present error bounds. The presented worthwhile approach can be the basis of a numerical algorithm.