基于简单锥模型的带线搜索的新信赖域算法

A New Trust Region Algorithm with Simple Conic Models and Line Search

对于无约束优化问题,提出了一类基于简单锥模型的带线搜索的新信赖域算法。该算法采用大步长Armijo线搜索技术获得迭代步长,克服了每次迭代求解信赖域子问题时计算量较大的缺点,适用于求解大型优化问题。在适当的条件下,证明了该算法的全局收敛性。

This paper proposes a new non - monotone trust region algorithm on the basis of simple conic model for unconstrained optimization. It takes larger Armijo line search technique and overcomes the weakness of large amount of calculation while resolving the sub - problems at each iteration. Therefore, it is more suitable for large scale optimization problems. Under suitable conditions, the global convergence of the algorithm is proved.