| 摘要: |
| 基于带有割线条件的DL方法，提出了两个满足改进的割线条件的修正共轭梯度方法——MDDL1方法与WMDDL1方法.在步长满足Wolfe线搜索的条件下，证明了MDDL1方法具有充分下降性；进一步地证明了WMDDL1方法不依赖任何线搜索具有充分下降性；最后分析和证明了两个方法在步长满足强Wolfe线搜索的条件下对一般函数均具有全局收敛性. |
| 关键词: 无约束优化 共轭梯度法 修正的DL共轭梯度法 强Wolfe线搜索 全局收敛性 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Two Modified DL Conjugate Gradient Methods |
|
LIU Xin
|
| Abstract: |
| Based on the DL conjugate gradient method with secant equation,two modified DL conjugate gradient methods with the modified secant equation are proposed,which are called MDDL1 method and WMDDL1 method.The MDDL1 method can be proved to satisfy the sufficient descent condition when step size is obtained by Wolfe line search.Furthermore,the WMDDL1 method satisfies sufficient descent condition independent of any line search.Alternatively,the two methods can be proved to possess global convergence for general functions when step size is obtained by strong Wolfe line search. |
| Key words: unconstrained optimization conjugate gradient modified DL conjugate gradient method strong Wolfe line search global convergence |
