Canonical primal-dual algorithm for solving fourth-order polynomial minimization problems

Abstract

This paper focuses on implementation of a general canonical primal–dual algorithm for solving a class of fourth-order polynomial minimization problems. A critical issue in the canonical duality theory has been addressed, i.e., in the case that the canonical dual problem has no interior critical point in its feasible space View the MathML source, a quadratic perturbation method is introduced to recover the global solution through a primal–dual iterative approach, and a gradient-based method is further used to refine the solution. A series of test problems, including the benchmark polynomials and several instances of the sensor network localization problems, have been used to testify the effectiveness of the proposed algorithm.