A nonmonotone ODE-based method for unconstrained optimization

摘要: This paper presents a new hybrid algorithm for unconstrained optimization problems, which combines the idea of IMPBOT with nonmonotone line search technique. A feature proposed method is that at each iteration, system linear equations solved only once to obtain trial step, via modified limited-memory BFGS two loop recursion requires matrix–vector products, thus reducing computations and storage. Furthermore, when step not accepted, performs along it using scheme, larger stepsize can be yielded in procedure. Under some reasonable assumptions, convergence properties are analysed. Numerical results also reported show efficiency this method.