论文部分内容阅读
变尺度法是优化方法中应用较广的一个分枝。Davidon提出了一个方案。在每个迭代步中只须计算一次目标函数及其梯度而省去对目标函数的一维寻查。这样的方法对容易用解析法计算梯度而目标函数计算量较大的问题尤其有利。作者近年已将此法先后应用于CNDO/2几何优化、MNDO几何优化、Mossbauer谱分解、流体状态方程中常数的确定、键能与键长的非线性相关等工作。这些实践表明此法有对初值依赖不大、收敛性
Variable scale method is a branch widely used in optimization methods. Davidon proposed a solution. In each iteration step, the objective function and its gradient need to be calculated only once and the one-dimensional search of the objective function is omitted. Such a method is particularly advantageous for the problem of easily calculating the gradient by analytic method and the calculation of the objective function is large. In recent years, the author has applied this method to CNDO / 2 geometric optimization, MNDO geometric optimization, Mossbauer spectral decomposition, the determination of constants in fluid equation of state, and the nonlinear correlation between bond energy and bond length. These practices show that this method has little dependence on the initial value, astringency