High order structure preserving explicit methods for solving linear-quadratic optimal control problems and differential games

TL;DR

This paper introduces high order explicit geometric integrators for solving linear-quadratic optimal control problems and differential games, effectively preserving qualitative properties of solutions through splitting methods.

Contribution

The paper develops novel high order explicit splitting methods tailored for coupled Riccati and state equations, preserving qualitative solution features in control problems.

Findings

Methods effectively preserve qualitative properties of solutions

Performance improves when system is a perturbation of an exactly solvable problem

Numerical examples demonstrate the methods' effectiveness

Abstract

We present high order explicit geometric integrators to solve linear-quadratic optimal control problems and $N$-player differential games. These problems are described by a system coupled non-linear differential equations with boundary conditions. We propose first to integrate backward in time the non-autonomous matrix Riccati differential equations and next to integrate forward in time the coupled system of equations for the Riccati and the state vector. This can be achieved by using appropriate splitting methods, which we show they preserve most qualitative properties of the exact solution. Since the coupled system of equations is usually explicitly time dependent, a preliminary analysis has to be considered. We consider the time as two new coordinates, and this allows us to integrate the whole system forward in time using splitting methods while preserving the most relevant qualitative structure of the exact solution. If the system is a perturbation of an exactly solvable problem, the performance of the splitting methods considerably improves. Some numerical examples are also considered which show the performance of the proposed methods.