A stable differential equation approach for inequality constrained optimization problems (Q2379039)

The author proposes the solution of systems of differential equations to solve a nonlinear constrained optimization problem. For this purpose, an augmented Lagrangian is derived. From this function two systems of differential equations are obtained, where the first one corresponds to a gradient-based optimization and the second one to a Newton-like methods. For both approaches, it is shown using usual assumptions that a Karush-Kuhn-Tucker point of the constrained optimization problem is an equilibrium point of the systems of differential equations, where the systems are asymptotically stable. For the numerical solution of the differential equations, it is show that versions of the Euler method are locally convergent. Iteration counts are shown for three problems of the Hock-Schittkowski test suite.