Non-Monotonic Lyapunov Stability for Efficient Intermittent Control

This direction is merged into another direction: event-triggered control, non-monotonic Lyapunov stability. Its last version stays readable and citable here.

Version 1 of 128 papers50 conceptsfrozen 2026-10-04

In classical control theory, stability is often treated as a strict, relentless descent. To prove a system is stable, engineers typically find a Lyapunov function—a mathematical measure of the system's "energy" or "error"—and show that this energy decreases at every single moment in time. While this monotonic requirement makes the math elegant, it is incredibly expensive in practice. It forces controllers to update frequently and sensors to stream data continuously just to ensure the energy never ticks upward, even for a millisecond.

A growing body of research suggests that this strictness is unnecessary. By allowing the system's energy to fluctuate—rising temporarily as long as it trends downward over time—we can drastically reduce the communication and computation costs of controlling complex systems.

The Lyapunov Bottleneck

The traditional requirement for monotonic decay creates a "Lyapunov bottleneck." In networked systems, where a controller might be miles away from the actuator, every update consumes bandwidth and battery. If we insist that the energy must decrease at every step, we are forced to sample the system state at a high frequency.

Manuel Mazo and Adolfo Anta argue in Self-Triggered Control: trading actuation for computation that we can trade onboard computation for communication efficiency. Instead of following a rigid periodic schedule, the controller predicts the system's future state and decides the next time it needs to run. This "self-triggered" approach relies on relaxing the stability constraints: as long as the system remains within a predefined "decay envelope" at specific checkpoints, it can safely be left alone in between.

Stability in the Average Sense

When dealing with unpredictable environments—like erasure channels where packets are frequently lost—monotonicity is often impossible to guarantee. Serdar Yuksel and Sean P. Meyn addressed this in Random-Time, State-Dependent Stochastic Drift for Markov Chains, showing that systems can remain stable even if they deviate or grow between successful updates.

This leads to the concept of stability in the average sense. Rather than requiring a decrease at every $t$, we require a decrease over a window of time or across a sequence of events. Borna Sayedana, Aditya Mahajan, and Peter E. Caines demonstrate in Strong Consistency and Rate of Convergence of Switched Least Squares System Identification that this relaxed condition is sufficient even for identifying the parameters of complex switching systems. If the energy grows no faster than the time horizon, the system remains manageable.

Performance Barriers and Decay Envelopes

How do we practically implement this "slack" without the system spiraling out of control? One powerful mechanism is the performance-barrier.

In Performance-Barrier Event-Triggered Control of a Class of Reaction-Diffusion PDEs, Bhathiya Rathnayake and Jorge Cortes define a "performance residual"—the gap between the current energy and a predefined exponential decay envelope. The controller only triggers an update when the system's energy threatens to cross this barrier. By allowing the energy to increase temporarily within this safe zone, they reduced control updates by a factor of 20 in some scenarios.

Similarly, Michael Hertneck and Steffen Linsenmayer show in Nonlinear Dynamic Periodic Event-Triggered Control that using non-monotonic Lyapunov functions allows for sampling periods more than twice as long as traditional methods. The system is allowed to "breathe" between samples, provided it settles down on average over successful updates.

Multi-Step Lookahead for Complex Policies

This relaxation is particularly vital for modern AI-driven control. Reinforcement Learning (RL) policies often exhibit transient instability—they might take a "wrong" step to set up a better maneuver later. Traditional Lyapunov functions would reject these policies as unstable.

To solve this, Kehan Long and Nikolay Atanasov introduced Generalized Lyapunov Functions in Certifying Stability of Reinforcement Learning Policies. By enforcing a decrease in a weighted average over a fixed future horizon (multi-step stability), they can certify the safety of neural network controllers that a single-step check would fail. This same multi-step logic is used by Shaoru Chen and George J. Pappas in Learning Lyapunov Functions for Hybrid Systems to find stability certificates for systems that mix continuous motion with discrete logic.

Why It Works: The Intuition of Slack

The core intuition behind this research direction is that stability is a destination, not a path.

If you are driving a car to a destination, you don't need to be moving closer to it at every single millisecond (you might need to turn away to follow a curve in the road). As long as you have a "barrier" (the road boundaries) and a "trend" (your average velocity toward the goal), you will arrive safely.

By formalizing this "slack" through non-monotonic Lyapunov functions and multi-step drift criteria, researchers like Kazumune Hashimoto and Dimos V. Dimarogonas are enabling a new generation of networked and autonomous systems that are both mathematically guaranteed to be stable and practically efficient enough to run on limited hardware and bandwidth.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_non_monotonic_lyapunov_stability_for_eff_e70ff7da,
  title = {Non-Monotonic Lyapunov Stability for Efficient Intermittent Control},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/non-monotonic-lyapunov-stability-for-efficient-intermittent-control/rdr_e70ff7da57bd4488bc3455224301e505},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_non_monotonic_lyapunov_stability_for_eff_e70ff7da_v1,
  title = {Non-Monotonic Lyapunov Stability for Efficient Intermittent Control},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/non-monotonic-lyapunov-stability-for-efficient-intermittent-control/rdr_e70ff7da57bd4488bc3455224301e505/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}