When do Trajectories have Bounded Sensitivity to Cumulative Perturbations

Arsalan Sharifnassab, S. Jamaloddin GolestaniarXiv 2019 · arXiv

Introduction

In the fields of control theory and robotics, understanding how disturbances affect a system's trajectory is paramount. Traditionally, robustness is measured against the total absolute magnitude of noise over time. However, in many stochastic environments, positive and negative disturbances often cancel each other out. This paper investigates "bounded sensitivity," a property where trajectory error is constrained by the cumulative sum (the net integral) of disturbances rather than their absolute sum. Because cumulative noise grows much more slowly (typically $\sqrt{t}$) than absolute noise (linear with $t$), systems with this property are exceptionally robust.

The Linear System Constraint

A primary finding of this research is that simple stability is insufficient for a linear system ($\dot{x} = Ax$) to exhibit bounded sensitivity. The authors demonstrate that the system must be Stable and Orbit-Free (SOF).

A 2D spiral illustrating how small perturbations in a system with orbits can lead to unbounded trajectory deviations over time.

As shown in Figure 1, even a simple circular orbit allows a tiny, bounded cumulative perturbation to push a trajectory infinitely far away. Theorem 2 formalizes this: a linear system has bounded sensitivity if and only if its eigenvalues have non-positive real parts, and those with zero real parts have no imaginary components (preventing oscillations) and a simple Jordan structure.

The "Finiteness" Requirement in Convex Systems

Previous literature suggested that "Finitely Piecewise Constant Subgradient" (FPCS) systems—those resembling a convex roof of finite flat tiles—possess bounded sensitivity. This paper asks if this generalizes to all convex systems. The answer is a definitive "no."

The authors prove that the "finiteness" of pieces is essential. They constructed a strictly convex, smooth potential function and an infinitely piecewise linear convex function that both exhibit unbounded sensitivity.

The "rotating book" counter-example: (a) shows the complex surface construction, and (b) illustrates the resulting diverging spiral trajectory under specific perturbations.

To visualize this "convex fail," the authors use a "rotating book" analogy (Figure 2). Imagine a surface like an open book where the "spine" rotates as you move along the vertical axis. This rotation forces trajectories into diverging spirals, amplifying small net disturbances into large deviations.

Methodology and Theoretical Framework

The research utilizes rigorous mathematical analysis centered on differential inclusions, which model systems where movement can occur in a range of possible directions. Key techniques include:

  • Spectral Theory: Used to identify the specific eigenvalue conditions for linear systems.
  • Transfinite Recursion: Advanced set-theoretic tools were employed to prove that "spreading" a system (allowing it to drift among nearby points) preserves sensitivity by partitioning time into infinitesimal intervals.
  • Discretization Analysis: The authors proved that bounded sensitivity in continuous-time systems is inherited by their discrete-time implementations, a critical finding for computer-controlled robotics.

Conclusion and Future Directions

This work clarifies the boundaries of trajectory robustness. While symmetric linear systems are naturally robust (with a sensitivity constant $C = n+1$), the presence of oscillations or infinite complexity in convex surfaces can destroy this property.

Future research will explore the gap between "strong" and "weak" sensitivity and extend these findings to non-convex systems and systems with hard boundary conditions, such as queueing networks where state variables cannot be negative.

Summary updated

Abstract

We investigate sensitivity to cumulative perturbations for a few dynamical system classes of practical interest. A system is said to have bounded sensitivity to cumulative perturbations (bounded sensitivity, for short) if an additive disturbance leads to a change in the state trajectory that is bounded by a constant multiple of the size of the cumulative disturbance. As our main result, we show that there exist dynamical systems in the form of (negative) gradient field of a convex function that have unbounded sensitivity. We show that the result holds even when the convex potential function is piecewise linear. This resolves a question raised in [1], wherein it was shown that the (negative) (sub)gradient field of a piecewise linear and convex function has bounded sensitivity if the number of linear pieces is finite. Our results establish that the finiteness assumption is indeed necessary. Among our other results, we provide a necessary and sufficient condition for a linear dynamical system to have bounded sensitivity to cumulative perturbations. We also establish that the bounded sensitivity property is preserved, when a dynamical system with bounded sensitivity undergoes certain transformations. These transformations include convolution, time discretization, and spreading of a system (a transformation that captures approximate solutions of a system).
BibTeX
@misc{sharifnassab2019when,
  author = {Arsalan Sharifnassab and Arsalan Sharif-Nassab and S. Jamaloddin Golestani},
  title = {{When do Trajectories have Bounded Sensitivity to Cumulative Perturbations}},
  year = {2019},
  eprint = {1905.11746},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1905.11746},
}
Figure