结合滤子技术的牛顿折线法及其实现

A Filter Newton Dogleg Method and Its Implementation

成功将多维滤子技术应用到牛顿折线法,提出了多维滤子牛顿折线法.新算法增加了牛顿点以及信赖域的试探点被接收作为下一步迭代点的几率.在一定的假设条件下证明了算法的全局收敛性.数值试验表明,滤子牛顿折线法适合于求解等势线呈峡谷状的函数.

This paper gives an implementation of the multidimensional filter technique,we employ this technique into the Newton Dogleg method,which makes the Newton point and the trial point of the trust region subproblem to be taken more often.Global convergence is promoted through the use of the filter.Numerical results demonstrate the filter Newton Dogleg method suitable for solving the function of curved valleys.