给出无约束最优化的一类带线搜索的非单调信赖域算法.在一定条件下证明了算法的全局收敛性和Q-二阶收敛速度.这类算法与通常的非单调信赖域算法不同.当试探步不成功时,采用线搜索技术得到下一个迭代点.这样不仅减少了计算量,而且避免了下参考函数值远大于实际函数值的问题.
A nonmonotonic trust region algorithm with line search for unconstrained optimization problems is presented in this paper. Its global convergence and Q- quadratic convergence are proved under suitable conditions. Differing from usual nonmonotonic trust region method, the algorithm takes line search technique to get the next iterative point when the trail step is not accepted. This method may not only reduce a considerable saving, but also avoid the possibility that the reference function value may be much larger than the real one.