Innite horizon optimal control and stabilizability for linear descriptor systems

Sergiy ZhukarXiv 2014 · arXiv

Abstract

This paper presents a novel construction for an infinite horizon Linear-Quadratic (LQ) optimal controller applicable to linear stationary differential-algebraic equations (DAEs). By leveraging linear geometric control theory, the authors demonstrate that the existence of such a controller is fundamentally tied to the stabilizability of the system from its initial state. Unlike traditional methods that require strict regularity assumptions, this approach maps DAE solutions to the outputs of standard Linear Time-Invariant (LTI) systems, providing a robust framework for generic, non-regular descriptor systems.

Introduction

Differential-Algebraic Equations (DAEs), also known as descriptor systems, are indispensable in the modeling of complex physical phenomena, including electrical circuits, constrained mechanical systems, and discretized partial differential equations (PDEs). Despite their utility, DAEs present significant theoretical hurdles compared to standard Ordinary Differential Equations (ODEs). Most classical control theories rely on the assumption of "regularity"—the guarantee of a unique solution for any given initial state and input.

In reality, many physical systems are non-regular, potentially exhibiting no solutions or an infinite set of solutions for certain conditions. Previous research often bypassed these issues by assuming perfectly smooth inputs or allowing for "distributional" solutions involving Dirac deltas, which often lack physical meaning. This paper addresses these gaps by introducing a framework that handles physically meaningful, locally integrable solutions for generic, rectangular DAE systems.

Methodology: A Geometric Approach

The core innovation of this work lies in the synthesis of geometric control theory with the behavioral approach to system modeling. The authors propose a multi-step methodology to bridge the gap between DAEs and standard LTI theory.

1. The Associated ODE-LTI Mapping

The authors prove that for any DAE defined by the equation: $$\frac{d(Ex)}{dt} = Ax + Bu$$ there exists an associated standard ODE system. In this framework, the state and input trajectories of the original DAE are represented as the outputs of the associated ODE. This mapping effectively "hides" the complex algebraic constraints within the state-space of the simpler system.

2. Feedback Equivalence and Consistency

A critical theoretical hurdle is ensuring that the choice of the associated ODE does not bias the results. The paper demonstrates that while multiple ODE representations may exist for a single DAE, all minimal representations are "feedback equivalent." This ensures that the derived control laws are consistent and intrinsic to the original DAE's behavior.

3. LQ Control and the Riccati Equation

By transforming the DAE into an associated ODE, the authors can map the DAE’s cost function into a standard quadratic cost. This allows for the application of the classical Algebraic Riccati Equation (ARE) to derive the optimal controller.

4. Behavioral Controllers

Traditional state feedback ($u = Kx$) often fails in the context of non-regular DAEs. To circumvent this, the authors propose a "behavioral" controller defined by an algebraic constraint: $$K_1x + K_2u = 0$$ This generalized approach allows for the stabilization of systems that would be considered uncontrollable under standard state-feedback paradigms.

Experimental Validation: Heat Transfer Simulation

To validate the theoretical framework, the authors conducted a numerical simulation based on a heat transfer PDE. The PDE was approximated using a DAE-LTI model and compared against a traditional Galerkin projection—a common engineering shortcut that simplifies PDEs into ODEs.

The results highlighted the precision of the DAE-based approach:

  • Theoretical Optimal Cost: 3.94
  • DAE-based Method Cost: 3.89
  • Galerkin Projection Cost: 6.13

The DAE-based method significantly outperformed the Galerkin projection, achieving a cost very close to the theoretical optimum. This suggests that the DAE-LTI framework captures the underlying physics of the PDE with much higher fidelity than standard ODE approximations.

Key Insights and Theoretical Contributions

The paper provides several landmark contributions to the field of descriptor systems:

  • Stabilizability as a Necessary Condition: The authors define "behavioral stabilizability" and prove it is the exact necessary and sufficient condition for the existence of an infinite-horizon optimal controller.
  • Constructive Algorithm: The research provides a concrete algorithm, implemented in MATLAB, to generate the associated ODE for any given DAE.
  • Handling Non-Regularity: By treating DAE solutions as output trajectories, the framework successfully manages systems previously deemed too "ill-behaved" for standard optimal control theory.

Limitations and Future Directions

While the framework represents a significant advancement, it is currently limited to Linear Time-Invariant (LTI) systems. Extending these geometric methods to time-varying or nonlinear DAEs remains an open challenge. Additionally, the authors intentionally exclude distributional solutions (impulses) to maintain physical relevance, though certain applications like impact mechanics might eventually require their inclusion. Future work may also explore the deeper mathematical connections between this framework and the "quasi-Weierstrass forms" found in matrix pencil theory.

Conclusion

This paper provides a rigorous and constructive solution to the LQ optimal control problem for linear descriptor systems. By shifting the perspective from direct state feedback to a behavioral, geometric mapping, the authors have opened a path for the optimal control of complex, non-regular systems across various engineering and physical domains.

Summary updated

Abstract

In this paper we construct an innite horizon LQ optimal controller for a linear stationary dierential-algeb raic equation (DAE) and prove that stabilizability from the given initial state is necessary and sucient for the controller existence. We also solve the problem of nite horizon LQ optimal control. Our approach is based on ideas from linear geometric control theory that allows one to represent solutions of DAEs as outputs of LTI system. It is applicable for generic DAEs without imposing additional regularity assumptions.
BibTeX
@misc{zhuk2014innite,
  author = {Sergiy Zhuk},
  title = {{Innite horizon optimal control and stabilizability for linear descriptor systems}},
  year = {2014},
  eprint = {1312.7547},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1312.7547},
}
Figure