针对只含有不等式约束的优化问题,本文首先给出了其Karush-Kuhn-Tucker (KKT)条件,并利用光滑互补函数将KKT系统转化为一类光滑的方程组问题;其次,将光滑方程组问题转化为无约束优化问题;最后,本文提出一类二阶微分方程系统求解无约束...针对只含有不等式约束的优化问题,本文首先给出了其Karush-Kuhn-Tucker (KKT)条件,并利用光滑互补函数将KKT系统转化为一类光滑的方程组问题;其次,将光滑方程组问题转化为无约束优化问题;最后,本文提出一类二阶微分方程系统求解无约束优化问题,并讨论了二阶微分方程系统的解的稳定性及收敛速度。For optimization problems with only inequality constraints, this paper first presents their Karush-Kuhn-Tucker (KKT) conditions, and uses smooth complementarity functions to transform the KKT system into a class of smooth system of equations problems. Secondly, this article transforms the problem of smooth equation systems into an unconstrained optimization problem. Finally, this article proposes a class of second-order differential equation systems for solving unconstrained optimization problems, and discusses the stability and convergence speed of the solutions of second-order differential equation systems.展开更多
本文运用光滑化的自然残差函数建立了具有不等式约束条件的变分不等式问题的光滑化KKT方程组,并建立了与其等价的无约束优化问题。建立了具有阻尼惯性参数和时间尺度参数的二阶微分方程系统来求解该无约束优化问题,并证明了该二阶微分...本文运用光滑化的自然残差函数建立了具有不等式约束条件的变分不等式问题的光滑化KKT方程组,并建立了与其等价的无约束优化问题。建立了具有阻尼惯性参数和时间尺度参数的二阶微分方程系统来求解该无约束优化问题,并证明了该二阶微分方程系统的稳定性,从而得到了具有不等式约束的变分不等式问题的KKT点的收敛性。并将二阶微分方程方法与已有的一阶微分方程方法进行了理论条件和数值结果的对比。在理论条件的要求上,二阶微分方程方法的条件要更容易实现,而在数值结果上,一阶微分方程方法的收敛速度要快,但是两种方法的差距可以忽略不计。In this paper, a smoothing KKT equation system with inequality constraints is established by using the smoothing natural residual function, and the unconstrained optimization problem is established. A second order differential equation system with damping coefficient and time scale coefficient is established to solve the unconstrained optimization problem, and the stability of the second-order differential equation system is proved, then the convergence of the KKT points of the variational inequality problem with inequality constraints is obtained. The theoretical conditions and numerical results of the second-order differential equation method are compared with the existing first order differential equation method. In terms of theoretical conditions, the conditions of the second-order differential equation method are easier to implement, while in the numerical results, the convergence speed of the first order differential equation method is faster, but the difference between the two methods is negligible.展开更多
文摘针对只含有不等式约束的优化问题,本文首先给出了其Karush-Kuhn-Tucker (KKT)条件,并利用光滑互补函数将KKT系统转化为一类光滑的方程组问题;其次,将光滑方程组问题转化为无约束优化问题;最后,本文提出一类二阶微分方程系统求解无约束优化问题,并讨论了二阶微分方程系统的解的稳定性及收敛速度。For optimization problems with only inequality constraints, this paper first presents their Karush-Kuhn-Tucker (KKT) conditions, and uses smooth complementarity functions to transform the KKT system into a class of smooth system of equations problems. Secondly, this article transforms the problem of smooth equation systems into an unconstrained optimization problem. Finally, this article proposes a class of second-order differential equation systems for solving unconstrained optimization problems, and discusses the stability and convergence speed of the solutions of second-order differential equation systems.
文摘本文运用光滑化的自然残差函数建立了具有不等式约束条件的变分不等式问题的光滑化KKT方程组,并建立了与其等价的无约束优化问题。建立了具有阻尼惯性参数和时间尺度参数的二阶微分方程系统来求解该无约束优化问题,并证明了该二阶微分方程系统的稳定性,从而得到了具有不等式约束的变分不等式问题的KKT点的收敛性。并将二阶微分方程方法与已有的一阶微分方程方法进行了理论条件和数值结果的对比。在理论条件的要求上,二阶微分方程方法的条件要更容易实现,而在数值结果上,一阶微分方程方法的收敛速度要快,但是两种方法的差距可以忽略不计。In this paper, a smoothing KKT equation system with inequality constraints is established by using the smoothing natural residual function, and the unconstrained optimization problem is established. A second order differential equation system with damping coefficient and time scale coefficient is established to solve the unconstrained optimization problem, and the stability of the second-order differential equation system is proved, then the convergence of the KKT points of the variational inequality problem with inequality constraints is obtained. The theoretical conditions and numerical results of the second-order differential equation method are compared with the existing first order differential equation method. In terms of theoretical conditions, the conditions of the second-order differential equation method are easier to implement, while in the numerical results, the convergence speed of the first order differential equation method is faster, but the difference between the two methods is negligible.
