On Irregular Linear Quadratic Control: Stochastic Case

TL;DR

This paper addresses stochastic linear quadratic control problems with irregular Riccati equations, proposing a multi-layer optimization approach to derive controllers, which differs from classical methods that rely on regular Riccati equations.

Contribution

It introduces a novel multi-layer optimization method for irregular stochastic LQ control problems, enabling controller design without regular Riccati equation solutions.

Findings

Different controller entries require separate equilibrium conditions.

Controller design involves multiple layers with terminal constraints.

The approach clarifies differences between open-loop and closed-loop control.

Abstract

As it is popular known, Riccati equation is the key basic tool for optimal control in the modern control theory. The solvability conditions of optimal control, stabilization conditions and controller design are all based on the Riccati equation. However, these results highly depends on a strictly assumption that the Riccati equation is regular. If the Riccati equation is irregular, the controller could not be derived from the equilibrium condition. This paper is concerned with the general stochastic linear quadratic (LQ) with irregular Riccati equation. Different from the classical control theory for regular LQ problems, a new approach of `multi-layer optimization' is proposed. With the approach, we show that different controller entries of irregular-LQ controller need to be derived from different equilibrium conditions and a specified terminal constraint condition in different layers, which is much different from the classical regular LQ control where all the controller entries can be obtained from equilibrium condition in one layer based on regular Riccati equation. The presented results clarify the differences of open-loop control from closed-loop control in the aspects of solvability and controller design and also explores in essentially the differences of regular control from irregular control. Several examples are presented to show the effectiveness of the proposed approach.