Two differential equation systems for inequality constrained optimization
The inequality-constrained optimization problem \[ \min f(x)\text{ with respect to }g_i(x)\leq 0,\qquad i= 1,\dots, m \] is studied, where all real-valued functions are twice continuously differentiable. The idea is to construct a system of ordinary differential equations (ODEs), so that its equilibrium points coincide with the solutions to the inequality-constrained problem. This idea was used by \textit{Yu. G. Evtushenko} and \textit{V. G. Zhadan} [Stable barrier projection and barrier-Newton methods in nonlinear progrmming. Opt. Software 3, 237--256 (1994)] in the case of equality-constrained problems. The transformation proposed in the present paper is a new one. Then a nonlinear Lagrangian is defined by \[ F_\sigma(x, y)= f(x)+ \sigma\Sigma y^2_i(e^{Sg_i(x)}- 1),\quad S= \sigma^{-1}. \] Let \((x^*, \mu^*)\) be a Kuhn-Tucker point of the original optimization problem, necessary optimality conditions are derived using the Lagrangian. These points are asymptotically stable equilibrium points of two differential equations. The Euler discrete schemes with constant stepsizes for the two ODE-systems are given and their convergence theorems are demonstrated. An Euler discrete scheme with an Armijo line search has a quadratic convergence rate. An algorithm is given. The Runge-Kutta method for the second system has good stability and the Euler scheme with the Armijo line search rule has fast convergence. Numerical results of some classical examples published in a lecture notes of \textit{W. Hock} and \textit{K. Schittkowski} [Text examples for nonlinear programming codes. (1981; Zbl 0452.90038)] are added. Unfortunately, these examples are not repeated here.
- Two differential equation systems for equality-constrained optimization
- A stable differential equation approach for inequality constrained optimization problems
- A dynamical systems approach to constrained minimization
- Constrained optimization: Projected gradient flows
- Differential systems for constrained optimization via a nonlinear augmented Lagrangian
- A differential equation approach to nonlinear programming
- Barrier-projective methods for nonlinear programming
- scientific article; zbMATH DE number 3116768 (Why is no real title available?)
- scientific article; zbMATH DE number 4055390 (Why is no real title available?)
- scientific article; zbMATH DE number 37929 (Why is no real title available?)
- scientific article; zbMATH DE number 778134 (Why is no real title available?)
- scientific article; zbMATH DE number 3307153 (Why is no real title available?)
- scientific article; zbMATH DE number 3381785 (Why is no real title available?)
- Stable barrier-projection and barrier-Newton methods in linear programming
- Stable exponential-penalty algorithm with superlinear convergence
- Test examples for nonlinear programming codes
- Optimization of systems governed by hyperbolic partial differential equations with equality and inequality constraints
- Differential equation method based on approximate augmented Lagrangian for nonlinear programming
- A stable differential equation approach for inequality constrained optimization problems
- Two differential equation systems for equality-constrained optimization
- Differential systems for constrained optimization via a nonlinear augmented Lagrangian
- scientific article; zbMATH DE number 555729 (Why is no real title available?)
- Inequality systems and optimization
This page was built for publication: Two differential equation systems for inequality constrained optimization
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2372013)
