H∞-Optimal Control and Related

Tamer Başar, Pierre BernhardSystems 1991 · DOI · OpenAlex

Abstract: One of the major concentrated activities of the past decade in control theory has been the development of the so-called "H∞-optimal control theory," which addresses the issue of worst-case controller design. This work provides a unified framework for this theory by reframing it as a minimax dynamic game, offering a state-space approach that bridges the gap between complex frequency-domain methods and modern control engineering.

Introduction

In the 1980s, the field of control theory underwent a significant shift with the emergence of $H^\infty$-optimal control. This paradigm focuses on "worst-case" design, aiming to develop controllers for linear plants that maintain stability and performance even in the presence of unknown, additive disturbances—such as atmospheric turbulence affecting aircraft or electrical noise in precision circuitry.

Initially, $H^\infty$ problems were primarily addressed using frequency-domain tools, including spectral factorization and operator theory. While mathematically rigorous, these methods were often dense and resulted in high-dimensional controllers that were difficult to implement. Furthermore, frequency-domain approaches struggled with time-varying systems or finite-time horizons. The seminal work by Tamer Başar and Pierre Bernhard (1991) revolutionized this field by providing a direct, state-space (time-domain) derivation, reframing the problem through the lens of dynamic games.

The Dynamic Game Framework

The core innovation of this approach is the conceptualization of $H^\infty$ control as a zero-sum dynamic game between two competing players:

  1. The Controller (Minimizer): Tasked with minimizing a specific performance index or cost function.
  2. The Disturbance (Maximizer): An intelligent adversary attempting to maximize that same cost.

By seeking the "saddle-point" solution to this minimax problem, the resulting controller is guaranteed to handle the worst possible disturbance within a defined energy bound. This adversarial mindset ensures robustness; a controller designed to defeat an "intelligent" adversary is inherently resilient against any random or "dumb" noise.

Technical Methodology

The transition to a state-space framework allowed for the use of several key technical tools:

  • Generalized Riccati Equations: These are used to characterize optimal strategies in the time domain, providing a clear path to implementation.
  • State-Space Modeling: By moving from transfer functions to differential or difference equations, the theory became more accessible for modern computational tools.
  • Certainty Equivalence Principle: This principle was extended to handle cases with imperfect information (where the state is not perfectly measured), allowing for the separation of estimation and control in a minimax context.

Key Insights and Theoretical Contributions

One of the most significant insights provided by Başar and Bernhard is the simplification of the McMillan degree. They demonstrated that $H^\infty$ controllers do not need to be more complex than the plants they control; the controller's dimension is directly tied to the plant's dimension.

Furthermore, the work provides a unified theory for both continuous-time and discrete-time systems. It shows that the underlying game-theoretic logic remains consistent across different temporal domains. The text also meticulously explores how various "information structures"—such as delayed measurements or sampled data—influence the existence and specific form of the optimal controller.

Results and Impact

The state-space approach yielded explicit formulas for $H^\infty$ controllers using standard state-space matrices, significantly easing software implementation. The work established necessary and sufficient conditions for achieving specific levels of disturbance attenuation, typically involving the existence of positive-definite solutions to Riccati equations.

While primarily focused on linear systems, the authors also laid the groundwork for extending these minimax techniques to nonlinear plants and non-quadratic cost functions, opening new avenues for robust control research.

Limitations and Future Directions

Despite its foundational status, the 1991 framework had limitations. The rigorous proofs were largely confined to linear, time-invariant (LTI) or time-varying systems. A generalized theory for fully nonlinear $H^\infty$ control remained an open frontier at the time. Additionally, solving the required Riccati equations can be computationally intensive for very high-dimensional systems, such as large-scale power grids.

Future research directions identified by the authors included extending these results to distributed-parameter systems (governed by partial differential equations) and further refining minimax estimation techniques under imperfect information.

Conclusion

The dynamic game approach to $H^\infty$-optimal control remains a cornerstone of modern robust control theory. By transforming a complex frequency-domain problem into an intuitive adversarial game, Başar and Bernhard provided engineers and mathematicians with a powerful, unified toolkit for designing systems that can withstand the worst the world has to throw at them.

Summary updated

Abstract

One of the major concentrated activities of the past decade in control theory has been the development of the so-called "HOO-optimal control theory," which addresses the issue of worst-case controller
BibTeX
@inproceedings{basar1991h,
  author = {Tamer Başar and Pierre Bernhard},
  title = {{H∞-Optimal Control and Related}},
  booktitle = {Systems},
  year = {1991},
  doi = {10.1007/978-1-4899-3561-8},
}

Loading related content...

Figure