摘要
根据非线性最小二乘问题的分裂拟牛顿法提出了无记忆型拟ＧＮ法。它的近似Ｈｅｓｓｅ矩阵在下降方向上满足拟牛顿方程，比ＧＮ法近似性好；它又比一般的分裂拟牛顿法节省存储空间，减少计算量；结合信赖域技术，算法具有良好的理论性态。文后给出一些有代表性的数值结果以说明算法的有效性。
Abstract A memoryless type quasi Gauss Newton method for nonlinear least squares problems was proposed based on the factorized quasi Newton methods. The new updating formula also satisfies the quasi Newton equation, it gives a better approximation than the GN update, and it saves much storage and algebraic operations than the orginal factorized quasi Newton methods. Trust region type algorithm possesses very good theoretical properties. Some of numerical results are presented.
出处
《工程数学学报》
CSCD
北大核心
1999年第2期104-108,126,共6页
Chinese Journal of Engineering Mathematics
基金
鞍山钢铁学院董事会基金
