An affine scaling interior trust-region method for \(LC^{1}\) minimization subject to bounds on variables (Q2489419)

The following optimization problem is considered: Minimize \(f(x)\) subject to \(Ax = b\), \(l\leq x\leq u\), where \(f : \mathbb{R}^n\mapsto \mathbb{R}\) is a \(LC^1\) differentiable not necessarily convex function with locally Lipschitzian derivatives, and \(l_j\in \mathbb{R}\cup \{-\infty\}\), \(u_j\in \mathbb{R}\cup\{+\infty\}\) \(\forall j =1,\dots,n\), \(l\leq u\), \(l\neq u\) are given lower and upper bounds on the variables \(x_j\). The classical affine scaling interior trust-region algorithm for smooth nonlinear programming problems is extended to nonsmooth problems with a locally Lipschitzian objective function. The proposed algorithm combines the technique of trust-region, interior point, affine scaling, backtracking and nonmonotonic search. Convergence properties of the algorithm and applications to various nonsmooth problems are discussed.