摘要
提出了一种新的修正拟牛顿信赖域算法.算法同时利用函数值信息和梯度信息构造信赖域子问题,既可保持信赖域子问题海森矩阵的正定性,又能改善算法的数值执行.在一定假设的条件下,证明了算法的全局收敛性,并通过数值实验表明了提出算法的有效性.
A new trust region method based on the modified quasi-Newton equation are proposed.The presented algorithm uses available gradient and function information to build trust region subproblem,which can not only make sure the positive-definite for Hessian matrix of trust region subproblem but also improve numerical implementation.Under some suitable conditions,a global convergence theorem is established.Numerical experiments show that the presented algorithm is efficient.
出处
《北华大学学报(自然科学版)》
CAS
2009年第5期385-388,共4页
Journal of Beihua University(Natural Science)
基金
吉林省教育厅“十一五”科学研究项目(2008-129)
关键词
信赖域方法
修正拟牛顿方法
全局收敛性
Trust region
Modified quasi-Newton equation
Global convergence