A new norm-relaxed method of strongly sub-feasible direction for inequality constrained optimization (Q2572648)

Proceeding from the ideas of the generalized norm-relaxed method of feasible direction and the method of strongly-feasible direction, a new algorithm is proposed for solving nonlinear constrained optimization problems. At each iteration, the new algorithm solves one direction finding subproblem which always has a solution. The new algorithm not only unifies automatically the operations of initialzation (Phase I) and optimization (Phase II), but also guarantees that the number of functions satisfying the inequality constraints is nondecreasing. The advantages of the new method are: (i) the initial point is arbitrary; (ii) the number of constraints satisfying the inequality is nondecreasing; (iii) a feasible descent direction for the problem is obtained by solving only one subproblem whenever the iteration point enters the feasible set; (iv) the global and strong convergence are guranteed under some mild assumptions without the linear independence. The proposed method is tested on some practical problems. The numerical experiments are implemented on MATLAB 6.5.