一种全局收敛的杂交谱共轭梯度方法

The Global Convergence of a New Hybrid Spectral Conjugate Gradient Method

本文对于解决无约束优化问题提出了一种新的具有全局收敛性的CD法与DY法的杂交谱共轭梯度方法,新算法具有以下性质:(1)这种新方法在任何条件下满足充分下降性质;(2)证明了在Wolfe线搜索条件下的全局收敛性质.并且进行了数值实验,结果显示这种方法是非常有效的.

In this pager, we propose a new hybrid spectral conjugate gradient method to solve unconstrained optimization problems. It has the following properties：（1） the new method satisfies the sufficient descent condition with Wolfe line search; （2） under the Wolfe line search condition the method is globally convergence. Preliminary numerical results show that the method is very efficient.