本文讨论Rn空间上的无约束极大极小问题.通过Rn+1空间上的广义梯度投影技术产生Rn上的下降搜索方向,进而结合Armijo非精确线搜索建立了原问题Rn上的一个广义梯度投影型算法.算法在仿射线性无关条件下,具有全局收敛性和强收敛性.文中对算法进行了初步的数值试验.
In this paper, the unconstrained minimax problems on Rn are discussed. The search direction of descent in Rn is obtained by a generalized gradient projection on Rnn+1, then, with Armjio non-exact line search, a generalized gradient projection algorithm on Rn for the discussed minimax problems is presented. The proposed algorithm possesses global and strong convergence under affine linearly independent condition. Some preliminary numerical experiments are carried out.