A Modified BFGS Trust Region Method

摘要

We propose a new trust region method that employs both the modified BFGS update and Amijio line search. The method exploits the information of function and gradient, and ensures the Hessian matrix of trust region subproblem positive-definite. At some assumptions, the global convergence and superlinear convergence property are proposed. Finally, numerical experiments show that the method is efficient.