A PRIMAL-DUAL TRUST REGION ALGORITHM FOR NONLINEAR

摘要: This paper concerns general (nonconvex) nonlinear optimization when rst and second derivatives of the objective constraint functions are available. The proposed method is based on nding an approximate solution a sequence unconstrained subproblems pa- rameterized by scalar parameter. function each subproblem augmented penalty-barrier that involves both primal dual variables. Each solved using second-derivative Newton-type employs combined trust-region line search strategy to ensure global convergence. It shown trust- region step can be computed factorizing systems with diagonally-modie d primal-dual structure, where inertia these determined without recourse special factorization method. has benet o-the-shelf linear system software used at all times, allowing straightforward extension large-scale problems.