A CONIC AFFINE SCALING DOGLEG METHOD FOR NONLINEAR OPTIMIZATION WITH BOUND CONSTRAINTS

The conic model proposed by Davidon (1980) [Davidon, WC (1980). Conic approximations and collinear scalings for optimizers. SIAM Journal on Numerical Analysis, 17, 268–281.] is an extension of quadratic model incorporating more information than quadratic model. In this paper, we propose a conic affine scaling dogleg method for bound constrained optimization. Under some mild conditions, the sequence generated by our algorithm converges to a stationary point. Numerical tests on a series of standard testing problems confirm the efficiency of the method.