Abstract
In this paper, we introduce the notion of recurrence entropy in the context of nonlinear control systems. A set is said to be ($\tau$-)recurrent if every trajectory that starts in the set returns to it (within at most $\tau$ units of time). Recurrence entropy quantifies the complexity of making a set $\tau$-recurrent measured by the average rate of growth, as time increases, of the number of control signals required to achieve this goal. Our analysis reveals that, compared to invariance, recurrence is quantitatively less complex, meaning that the recurrence entropy of a set is no larger than, and often strictly smaller than, the invariance entropy. Our results further offer insights into the minimum data rate required for achieving recurrence. We also present an algorithm for achieving recurrence asymptotically.
Introduction
In control theory, the concept of invariance is a cornerstone: ensuring a system stays within a "safe" set $Q$ indefinitely. However, strict invariance is often an excessively high bar, requiring high-precision control and high-bandwidth communication that may be mathematically or practically impossible for complex nonlinear systems.
This research explores a more flexible alternative: recurrence. A system is $\tau$-recurrent if it is guaranteed to return to the target set $Q$ within a fixed time interval $\tau$, even if it temporarily leaves the set. This paper introduces recurrence entropy as a formal metric to quantify the informational cost of maintaining this behavior. The central finding is that recurrence is fundamentally "cheaper" than invariance, requiring lower data rates to achieve.
Methodology and Theoretical Framework
The authors develop a framework for recurrence entropy by extending the established theory of invariance entropy. The methodology involves several key components:
- Formal Definitions: The study defines $(\tau, \epsilon, Q)$-recurrent trajectories—paths that return to an $\epsilon$-neighborhood of a set $Q$ at least once every $\tau$ seconds.
- Containment Lemma: A critical technical result that bounds the maximum drift of a recurrent trajectory before it is forced back to the target set.
- Entropy Bounds: Utilizing Grönwall’s inequality to manage error growth and Liouville’s trace formula to account for state-space volume expansion, the authors derive analytical upper and lower bounds for recurrence entropy.
- Algorithmic Implementation: To bridge theory and practice, the authors propose a sensor-controller pair that utilizes state quantization and a dynamic "moving grid" to enforce recurrence asymptotically over limited-bandwidth channels.
Key Insights
The analysis yields several significant insights into the nature of controlled recurrence:
- Recurrence vs. Invariance: The headline result proves that recurrence entropy $h_{rec}(\tau, Q)$ is always less than or equal to invariance entropy $h_{inv}(Q)$. This confirms that allowing a system "breathing room" to exit a set reduces the information processing load.
- Convergence at the Limit: As the return time $\tau$ approaches zero, recurrence entropy converges to invariance entropy, positioning recurrence as a generalized, flexible version of invariance.
- Information-Control Duality: A direct link is established between the system's inherent dynamics (e.g., divergence or "chaos") and the minimum bit rate required for control. The required bit rate scales linearly with the system's dimension and the desired convergence rate.
Results
The theoretical findings are crystallized in several key theorems:
- Theorem 1: Formally establishes the inequality $h_{rec}(\tau, Q) \le h_{inv}(Q)$.
- Theorem 3 & 5: Provide upper and lower bounds based on the system's Lipschitz constant, fractal dimension, and the average divergence (trace of the Jacobian) of the vector field.
- Algorithmic Efficiency: The proposed algorithm (Algorithm 1) is shown to achieve recurrence with a bit rate matching the derived theoretical upper bounds, demonstrating near-optimal performance.
The findings were validated using an illustrative two-dimensional linear system, demonstrating that while the cost of strict invariance can be infinite, the cost of recurrence remains finite and manageable.
Limitations and Future Directions
While providing a robust theoretical foundation, the work identifies areas for further development:
- Computational Efficiency: The current algorithm's reliance on real-time "moving grids" presents a computational burden. Future research may focus on fixed grids or symbolic abstractions.
- Numerical Estimation: Developing robust numerical methods to estimate recurrence entropy for black-box or high-dimensional systems remains an open challenge.
- Discrete-Time Integration: Connecting these continuous-time results to discrete-time safety properties is a promising avenue for future study.
Summary updated