On topological properties of closed attractors
Wouter Jongeneel · 2025
Read On topological properties of closed attractors by Wouter Jongeneel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-10429.txt.
Introduction
Wouter Jongeneel writes on attractors in dynamical systems, the sets toward which a system's motion settles over time. When attractors are compact, several rival definitions agree and the theory is fairly complete. Without compactness the picture blurs, and basic structural questions about closed attractors have stayed open.
Jongeneel characterizes when a closed, asymptotically stable attractor is a strong deformation retract of its domain of attraction. In plain terms, the whole basin can be continuously shrunk onto the attractor without tearing. The result works on locally compact metric spaces and needs no compactness of the attractor itself. It is meant to support a deeper structural study of feedback stabilization problems, where such attractors arise naturally.
Catalog record
| CREATOR | Wouter Jongeneel |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 69 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-10429
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
On topological properties of closed attractors Wouter Jongeneel Abstract The notion of an attractor has various definitions in the theory of dynamical systems. Under compactness assumptions, several of those definitions coincide and the theory is rather complete. However, without compactness, the picture becomes blurry. To improve our understanding, we characterize in this work when a closed, not necessarily compact, asymptotically stable attractor on a locally compact metric space is homotopy equivalent to its domain of attraction. This enables a further structural study of the corresponding feedback stabilization problem. Keywords —attractors, dynamical systems, stability AMS Subject Classification (2020) — 37B25, 55P10, 93D20 1 Introduction The (topological) dynamical systems theory of (asymptotically stable) compact attractors is rather mature, e.g. , see [ 11 , 14 , 19 , 1 ] . When an attractor is merely closed , however, our theory is significantly less complete and unified. In this work, we aim to contribute to improving our understanding here by answering the following question: Question: “ When is a closed attractor A , on a metric space (X,d) , homotopy equivalent to its basin of attraction B(A) ? ” We will make this question more precise in the remainder of the introduction, but first we elaborate on its relevance. The study of topological relations between attractors and their domain of attraction is a classical one and of importance in control theory. A reason being, due to insufficiently accurate models and computational obstructions, we are typically unable to find an explicit expression for the domain of attraction, especially when feedback is introduced. However, applications demand an understanding of this set, e.g. , what are all the perturbations a cruising aeroplane, oscillating genetic regulatory network or walking robot can recover from? Or differently put, given a control system and some desirable domain of attraction B^{\prime} , can we find continuous feedback such that the resulting basin of attraction B(A) equals B^{\prime} ? If such a feedback does not exist, how to overcome this? The topological viewpoint provides for coarse, but general answers to these questions. A basic example to have in mind is that a pendulum cannot be globally stabilized upright using continuous feedback, essentially due to the circle \mathbb{S}^{1} not being contractible. However, introducing a discontinuity in the feedback, that is, a single jump or switch, does allow for global stabilization. Topologically speaking, one needs to cut the circle. We cannot do justice to the rich history of this line of work, but we highlight the seminal contributions [ 74 ] , [ 14 ] , [ 30 ] , [ 72 , Thm. 21] , [ 10 ] and adjacent [ 18 , 43 , 77 , 20 ] . We highlight [ 56 , 57 , 64 , 9 , 45 , 75 , 76 , 46 , 47 , 40 ] as more recent contributions and we point the reader to [ 37 ] for a recent overview. The early focus on compact attractors can be understood as their study encapsulates equilibrium points and limit cycles [ 5 ] . Nevertheless, the focus on closed, but non-compact attractors in particular, is also of great theoretical and practical interest. From a practical perspective, we may provide the following examples. (i) The kernel of output maps . Suppose we have a continuous nonlinear control system of the canonical form \dot{x}=f(x,u) , y=h(x) , with 0\in\mathrm{dom}(h) , and we would like to globally zero the output y using an appropriate choice of static state-feedback x\mapstochar\rightarrow\mu(x) that enters as the input u . That means that we aim to render h^{-1}(0)=\{x\,|\,h(x)=0\} a global attractor. Since h is continuous, h^{-1}(0) is closed, but not necessarily compact. (ii) Synchronization and estimation . In the context of, for instance, observer design on a state space X , we generally aim to stabilize diagonal sets of the form \Delta_{X}=\{(x,x)\,|\,x\in X\} . If X is not compact, so is \Delta_{X} . (iii) Time-varying systems . Consider the time-varying ODE \dot{x}=f(x,t) on a space X . Suppose we want to understand stability of the set A\subseteq X under f . In that case, we could study stability of A\times\mathbb{R}_{\geq 0} under the autonomous ODE \dot{x}=f(x,s) , \dot{s}=1 . (iv) Overparametrized learning . In the context of machine learning, the majority of problems are formulated as optimization problems of the form \inf_{\theta\in\mathbb{R}^{d}}L(\theta) , where L:\mathbb{R}^{d}\rightarrow\mathbb{R} is some differentiable loss function. For instance, L(\theta)=\frac{1}{n}\sum^{n}_{i=1}\ell(h(x_{i},\theta),y_{i})^{2} , where (x_{i},y_{i}) are pairs of data points, \ell is a (local) loss function and h is a predictor parametrized by \theta , e.g. , x_{i} might be an image and y_{i} a binary classification variable that indicates if the image contains a dog or not. The map h is usually a neural network and the optimal \theta is typically sought through some approximation of \dot{\theta}=-\nabla L(\theta) . When the network is overparametrized , that is, d\gg n , the set of optimizers can be shown to be closed and unbounded under appropriate assumptions [ 67 ] . This means we study a closed attractor. From a theoretical perspective, closed attractors are challenging as compactness is a highly convenient structure exploited in a lot of proofs, e.g. , consult [ 11 , 14 , 19 , 1 , 29 ] , more concretely, see for instance [ 45 , Thm. 1] for a result where several of these constructions come together. To elaborate on our question, we will assume that (X,d) is a locally compact metric space and that A is uniformly asymptotically stable under some continuous dynamical system. Then, we will largely address our question using Borsuk’s retraction theory [ 35 , 15 , 21 ] . Moreover, we are particularly inspired by Auslander’s work towards unifying stability through filters [ 6 ] , that is, for compact attractors there is no difference between metrical- and topological definitions of stability, but for closed attractors this difference is non-trivial and neatly captured by neighbourhood filters. In general, the line of work by Bhatia and coworkers [ 5 , 12 , 11 , 14 ] provides us with the right foundations. We remark that their work builds upon the seminal monographs by Nemytskii and Stepanov [ 66 ] and Zubov [ 78 ] . Other noteworthy developments are Hájek’s para-stability [ 32 ] and Hurley’s work on exploiting locally compact \sigma -compact spaces in the context of non-compact attractors under maps [ 36 ] . We highlight that if a space X is locally compact and \sigma -compact, then there is a countable set of compact sets K_{1},K_{2},\dots, such that X=\cup_{i\in\mathbb{N}_{>0}}K_{i} , with K_{i}\subseteq\mathrm{int}\,K_{i+1} [ 16 , p. 94] . This structure is exploited in the majority of work concerned with closed attractors, e.g. , below we assume that (X,d) is not only locally compact, but also separable, this to appeal to [ 14 , Lem. V.4.26] . Indeed, for metric spaces this is equivalent to assuming local compactness and \sigma -compactness, e.g. , see [ 66 , Ch. V.1] . Most works appeal to a metric structure, we highlight one exception. There, the price to pay is that assumptions on A are arguably stronger. Specifically, building upon the likes of Zubov and Ura, Bhatia and Hájek provide a