针对无约束优化问题,提出一种新的锥模型信赖域算法。该方法组合了线搜索技术、截断拟牛顿法和锥信赖域法。当试探步不被接受时,采用非单调线搜索原则产生下一次迭代点,无需重解锥信赖域子问题。在适当的条件下,证明算法的全局收敛性和超线性收敛性,数值结果表明算法是可行的和有效的。
A conic trust region algorithm is proposed for unconstrained optimization problems. The method can be regard as a combination of nonmonotone line search technique, truncated Quasi-Newton method and conic trust re- gion method. When trail step is not accepted, a nonmonotone line search rule is used to obtain a suitable step length and generate next iterative point. It need not resolve the conic trust region subproblem. The theoretical anal- ysis shows that the algorithm is not only global convergence but also super linearly convergence under some suitable conditions. Numerical results show that this algorithm is effective and applicable.