An active set quasi-Newton method with projected search for bound constrained minimization

We analyze an active set quasi-Newton method for large scale bound constrained problems. Our approach combines the accurate active set identification function and the projected search. Both of these strategies permit fast change in the working set. The limited memory method is employed to update the inactive variables, while the active variables are updated by simple rules. A further division of the active set enables the global convergence of the new algorithm. Numerical tests demonstrate the efficiency and performance of the present strategy and its comparison with some existing active set strategies.