Minimum effort decentralized control design for contracting network systems

Ron Ofir, Francesco Bullo, Michael Margaliot

Abstract: We consider the problem of making a networked system contracting by designing minimal effort local controllers. Our method combines a hierarchical contraction characterization and a matrix-balancing approach to stabilizing a Metzler matrix via minimal diagonal perturbations. We demonstrate our approach by designing local controllers that render contractive a network of FitzHugh-Nagumo neurons with a general topology of interactions.

Introduction

In large-scale networks such as power grids, social networks, or biological systems, ensuring predictable global behavior is a significant challenge. A powerful property for achieving this is contraction. A contractive system is one where any two trajectories eventually converge toward each other, meaning the system "forgets" its initial conditions and settles into a unique equilibrium or follows a periodic excitation—a property known as entrainment.

However, designing controllers to achieve contraction in massive networks is difficult. Centralized control is often impractical due to communication constraints, and over-controlling can be wasteful or harmful. This paper addresses the challenge of finding the smallest possible local control gains that guarantee global network contraction.

Methodology

The proposed approach bridges network stability and efficient optimization by combining three mathematical frameworks:

  1. Hierarchical Contraction: Instead of analyzing the entire system of differential equations simultaneously, the authors use a comparison approach. Network interactions are represented as a smaller matrix where each entry denotes the interaction strength between subsystems.
  2. Metzler Matrices: The resulting comparison matrix is a Metzler matrix (all off-diagonal entries are non-negative). The stability of Metzler matrices is well-understood and easier to analyze than general matrices.
  3. Matrix Balancing: The core innovation lies in showing that finding the minimum effort diagonal perturbation to stabilize a Metzler matrix is mathematically equivalent to a matrix balancing problem. Matrix balancing—scaling a matrix so its row sums equal its column sums—is a well-studied problem in numerical linear algebra solvable in nearly linear time.

The resulting algorithm involves bounding subsystem interactions to create a Metzler "interaction matrix," using matrix balancing to find the optimal scaling, and then calculating the exact minimum local control gain for each subsystem.

Application and Results

The method was validated using a high-fidelity simulation of a FitzHugh–Nagumo (FHN) neuron network, a standard benchmark for modeling neural interactions. The theoretical results, specifically Theorem 4, provide an explicit formula for the optimal controller based on the Perron eigenvector of the balanced interaction matrix.

The authors demonstrated the method on a non-symmetric network of six neurons. With the "minimum effort" controllers, all neurons successfully entrained to a periodic input signal, as shown in the simulation results below.

Entrainment of FitzHugh-Nagumo neurons to a periodic input signal.

Key insights from the results include:

  • Optimal Efficiency: The optimization problem for minimum effort control has a unique, global solution found via matrix balancing, turning a non-convex problem into a fast calculation.
  • Network Awareness: While the control is decentralized, the design accounts for the network topology.
  • Automatic Entrainment: The method guarantees that a network of oscillators will synchronize to an external periodic signal with minimal energy expenditure.

Conclusion and Future Work

This research provides a computationally efficient recipe for stabilizing large-scale systems. Because matrix balancing scales well, the method is applicable to very large networks where traditional Lyapunov-based designs might fail.

Future work may explore time-varying gains to further reduce total effort and methods for accurately estimating Jacobian bounds in uncertain or "black-box" systems. Additionally, extending the framework to space- and time-dependent norms could allow for even more precise control interventions.

Summary updated

Abstract

We consider the problem of making a networked system contracting by designing minimal effort local controllers. Our method combines a hierarchical contraction characterization and a matrix-balancing approach to stabilizing a Metzler matrix via minimal diagonal perturbations. We demonstrate our approach by designing local controllers that render contractive a network of FitzHugh-Nagumo neurons with a general topology of interactions.
BibTeX
@article{ofir2022minimum,
  author = {Ron Ofir and Francesco Bullo and Michael Margaliot},
  title = {{Minimum effort decentralized control design for contracting network systems}},
  journal = {IEEE Control. Syst. Lett},
  year = {2022},
  doi = {10.1109/lcsys.2022.3176196},
}
Figure