A hybrid steepest descent method for constrained convex optimization

This paper describes a hybrid steepest descent method to decrease over time any given convex cost function while keeping the optimization variables in any given convex set. The method takes advantage of the properties of hybrid systems to avoid the computation of projections or of a dual optimum. The convergence to a global optimum is analyzed using Lyapunov stability arguments. A discretized implementation and simulation results are presented and analyzed. This method is of practical interest to integrate real-time convex optimization into embedded controllers thanks to its implementation as a dynamical system, its simplicity, and its low computation cost.