一种充分下降的CD共轭梯度法及其收敛性

A Sufficiently Descent CD Conjugate Gradient Method and Its Convergence

基于已有的CD方法,提出了一种改进的CD共轭梯度法(MCD算法).该算法产生的搜索方向为充分下降方向,且这一性质与所采用的线搜索方法无关;并在一定的条件下证明了该算法基于Wolfe线搜索求解非凸优化问题的全局收敛性.

Based on CD method,this paper proposes a modified CD conjugate gradient method（ MCD method）,the search direction generated by this algorithm is sufficiently descent direction and this property is nothing to do with the used line search method,and under certain condition,the solution to the global convergence of non-convex-optimization problem based on Wolfe line searching is proved.