【作者】 朱见广；

【导师】 刘红卫；

【作者基本信息】 西安电子科技大学 ， 应用数学， 2011， 博士

【摘要】 设计有效的算法是数值优化中的重要研究课题.本论文研究了非线性互补问题和非线性不等式系统这两类有着广泛应用背景的问题,主要从算法的设计,收敛性分析,数值效果等方面进行研究.主要内容概括总结如下：1.光滑互补函数在互补问题的光滑算法重构理论中起着重要作用.本章首先提出了一族新的光滑互补函数,该光滑函数包含了众多流行的光滑函数作为特例.这族光滑函数具有一些良好的性质,即：保证了与之相关的光滑路径的存在性和连续性,光滑算法所产生的迭代序列的有界性以及Jacobian相容性.利用这族光滑函数,讨论了一个光滑算法,实验结果表明新提出的光滑函数是有价值的.2.基于Fischer-Burmeister光滑函数,提出了求解P0非线性互补问题的一种正则化非精确非单调光滑牛顿算法.在较弱的条件下,我们证明了水平集是有界的以及算法具有全局收敛性和局部二次收敛性.数值实验结果也表明了算法的有效性,尤其在求解大规模的非线性互补问题时优势更明显.3.通过将信赖域技巧与线性搜索技巧相结合,提出了求解一般的(即：不要求是P0函数)非线性互补问题的一种新的半光滑Levenberg-Marquardt算法.这使得该算法可以去掉F至少是一个P0函数的假设.在适当的条件下,得到了算法的全局收敛性和局部超线性收敛性.数值实验结果表明该算法比一些现存方法更有效.4.对于互补问题的一些无导数下降算法,现存的通常都是基于单调线搜索进行分析的,而实际计算中都用了非单调线搜索,缺乏相应的理论分析.本章基于p范数,引入了一种广义的惩罚Fischer-Burmeister价值函数并证明了其具有很多好的性质.利用这个新价值函数,提出了具有非单调线搜索的无导数算法并证明了其全局收敛性与局部收敛性.使用测试题库MCPLIB进行了数值实验,实验结果表明：提出的算法是有效的以及新提出的价值函数是有意义的.5.利用加函数将非线性不等式系统转化为一非光滑的非线性方程组,再通过加函数的CHKS光滑函数,建立起非光滑方程组的近似光滑方程组.而后,提出了一个正则光滑牛顿算法来求解近似光滑方程组,从而得到原非线性不等式系统的解.数值实验结果表明了提出的算法是有效的.更多还原

【Abstract】 The design of an efficient algorithm is one of the most important research areas in numerical optimization community. This dissertation considers two class of wide appli-cation background problems, such as nonlinear complementarity problems and system of nonlinear inequalities, and mainly concentrates on the design of the algorithm, con-vergence analysis, numerical implementation. The main contributions are listed as follows:1. Smoothing functions play a key role in reformulation for the smoothing method. Firstly, a new smoothing function is proposed, which include many smoothing functions as special cases. The new smoothing functions possess a system of favorite properties, such as, the existence and continuity of a smooth path, boundedness of the iteration sequence and Jacobian consistency. Based on the new smoothing functions, we inves-tigate a smoothing Newton algorithm for the NCP, numerical results demonstrate that the new smoothing function introduced in this chapter is worth investigating.2. Based on the Fischer-Burmeister smoothing function, we propose a nonmonotone inexact regularized smoothing Newton method for nonlinear complementarity problems with a P0-function. Under the weaker condition, the boundedness of level set, global convergence and locally super linear convergence of the algorithm are established. Nu-merical results indicate that the newly proposed algorithm is effective, especially for the large scale nonlinear complementarity problems.3. By combining the trust region techniques and line searches, we propose a new semismooth Levenberg-Marquardt method for solving general (not necessarily P0) non-linear complementarity problems. Under suitable assumptions, global convergence and locally super linear convergence of the algorithm are established. Numerical results demonstrate that the proposed method is more efficient than some existing algorithms.4. Since the existing derivative-free descent algorithms for complementarity prob-lems are based on monotone line search, while their numerical implementations are based on nonmonotone line search, there lack theoretical analysis. Based on p-norm, we introduce a new generalized penalized Fischer-Burmeister merit function, and show that the function possesses a system of favorite properties. We propose a derivative-free algorithm for nonlinear comple-mentarity problems with a nonmonotone line search. The global convergence and locally convergence of the algorithm are established. Nu-merical results indicate that the newly proposed algorithm is effective and the new merit function is meaningful. 5. The system of nonlinear inequalities is reformulated as a nonsmooth equation by using a plus function. By using the Chen-Harker-Kanzow-Smale smoothing function of plus function, the nonsmooth equation is approximated by a family of parameterized smooth equations. A regularized smoothing Newton algorithm is proposed to solve the smooth equations and the solution of the system of nonlinear inequalities is founded. Numerical results indicate that the proposed algorithm is efficient.更多还原