An ODE-based nonmonotone method for unconstrained optimization problems

This paper proposes an ODE-based nonmonotone method for unconstrained optimization problems, which combines the idea of IMPBOT with the nonmonotone technique. The main characteristic of this method is that at each iteration, a system of linear equations is solved only once to obtain a trial step, via a modified L-BFGS two loop recursion that requires only vector inner products, thus reducing the matrix computation and storage. Then a modified nonmonotone line search is performed to generate next iterative point instead of resolving the linear system. Under some reasonable assumptions, the method is proven to be globally and superlinearly convergent. Numerical results show the efficiency of this proposed method in practical computation.