Topological State Space Inference for Dynamical Systems
Mishal Assif P K, Yuliy Baryshnikov · 2025
Read Topological State Space Inference for Dynamical Systems by Mishal Assif P K, Yuliy Baryshnikov (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-2505-00276.txt.
Introduction
Mishal Assif and Yuliy Baryshnikov present a computational pipeline to recover the shape of a dynamical system's state space from observations, in work from 2025. A system is watched through one output function along many paths. The hidden space in which the system moves has a topology that shapes all its behavior. Recovering that shape from data alone is the goal. The method works from samples of trajectories rather than from equations. Such recovery would help analysts infer hidden structure in systems known only through measured outputs.
Catalog record
| CREATOR | Mishal Assif P K, Yuliy Baryshnikov |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-00276
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Topological State Space Inference for Dynamical Systems Mishal Assif P K, Yuliy Baryshnikov Abstract We present a computational pipe aiming at recovery of the topology of the underlying phase space from observation of an output function along a sample of trajectories of a dynamical system. keywords State space realization, topological data analysis, nonlinear dynamics. 1 Introduction The problem of state space realization, that is, the reconstruction of the underlying space of a dynamical system and an observation function, remains foundational in systems theory. At its core, state space realization aims to reconstruct or identify a state space model that accurately describes the internal dynamics of a system based solely on observed data or system outputs. This problem originated with the work of Richard Bellman and Rudolf Kalman, whose pioneering research in the 1960s laid the groundwork for modern systems theory ( Schutter 2000 ) . The state space realization problem for linear models is well studied. Classical elegant solutions are available in this case relying, in essence on the theory of modules over the polynomial rings. Linearity enables a clear and tractable description of system dynamics, where system dynamics can be modeled as linear transformations, providing well-established methods for state space realization. Extensions to nonlinear systems introduce, however, substantial complexity. Nonlinear systems often exhibit intricate behaviors, including bifurcations and chaos. Consequently, realizing or reconstructing a state space for such systems is far less straightforward. In this note, we present an attempt to address the nonlinear version of the state space realization problem using a mathematical theory as a guiding light, much as the LTI realization theory relied on the structure theory of modules over polynomial rings in one variable. 1.1 Setup Consider a dynamical system (M,\phi) where M is a topological space and \phi:\mathbb{R}\times M\to M is a one-parametric transformation group, that is, \phi is a continuous function \mathbb{R}\times M\ni(t,m)\rightarrow\phi(t,m):=\phi_{t}(m)\in M\text{ such that }\phi_{0}(m)=m,\ \phi_{t}(\phi_{s}(m))=\phi_{t+s}(m) In most of our examples, M will be a smooth compact manifold and the transformation group will be generated by flows of a vector field v:M\rightarrow TM . We will denote by \pi_{x}:\mathbb{R}\rightarrow M the trajectory of v such that \pi_{x}(0)=x,\ \frac{d\pi_{x}(t)}{dt}=v(\pi_{x}(t)). (1) Note that in such cases, the transformation group \phi is at least as smooth as the vector field v . The question we investigate here is whether the topology of the space M can be recovered from a finite number of trajectories \{\pi_{x_{k}}\}_{k=1,..,n} of such a dynamical system on M . We will see that this is indeed possible and provide a principled computational pipeline and a methodology to extract topological information of M from such trajectories. 1.2 Data Analytic setup We are interested in building an algorithmic setup towards inference of the topology from observation data. To make the general problem more amenable to such data analysis, we add the following constraints to the problem setting: 1. The data are collected at discrete time intervals, whether it be actual readings from sensors or the result of numerical algorithms or simulations. In addition, such readings are collected only up to a finite time limit. We cannot hope to get entire trajectories \pi_{x} starting from a point x . Instead we will get discrete trajectories obtained by performing a finite sampling along the trajectory at regular time intervals. The sampling interval \delta will be assumed to be constant across all trajectories and will be dropped from the notation hereafter. 2. One observes only a finite sample of such trajectories, initialized according to some distribution, which may or not be under our control. 3. One observes not the points of the state space M (which is unknown), but rather a smooth output function g:M\rightarrow V\cong\mathbb{R}^{p} such that only g^{n}(x):=(g(\pi_{x}(-l\delta)),g(\pi_{x}(-(l-1)\delta)),...,g(\pi_{x}(k\delta)))\in V^{n},n=k+l+1 is available to us. The dimension of V can be much smaller than that of M , making this problem more challenging. Hence the overall problem we investigate here is: What topological information of M can be extracted from a finite set of observations of uniformly sampled trajectories \{g^{n}(\pi_{x_{\alpha}})\}_{\alpha=1,..,N} of a dynamical system on M ? 1.3 Takens’ Embedding There is a well-known answer to the question of the state realization due to the Takens embedding theorem ( Takens 2006 ) . Theorem 1.1 . Let M be a compact smooth manifold, and n\geq 2\dim(M)+1 an integer. Then, for a generic pair (\phi,g) , where \phi:M\to M is a C^{2} -diffeomorphism, and g a function in C^{2}(M,\mathbb{R}) , the map M\ni x\mapsto(g(x),g(\phi(x)),...,g(\phi^{n-1}(x))\in\mathbb{R}^{k} is an embedding of M . Genericity here is understood in the standard way: the pairs (\phi,g) which satisfy this condition form an open, everywhere dense subset in the Banach manifold of such pairs. Takens embedding theorem guarantees that for generic dynamical systems and observations, given the infinite collection of observations of uniformly sampled trajectories starting at any point on the manifold M with length more than twice the dimension of M , we can exactly recover M up to a diffeomorphism. While Takens theorem is encouraging and suggests that our problem is not completely intractable, it does not address our problem, being a purely existence result, not providing a viable strategy for reconstructing M from the trajectories, even in the ideal case where trajectories starting from all possible initial points are available. In our setting, the nature of the sample one observes, - a finite sample, - is intrinsically different from the smooth manifold M . Nevertheless, topological invariants , namely, the homology groups of the state space, can be recovered from finite observation of the trajectories, as we argue below. 1.4 Related Results The problem of the state space realization of nonlinear dynamical systems from observations has been addressed by many authors. Thus, ( Jakubczyk 1980 ) develops an analogue of the Taken’s embedding theorem for nonlinear control systems, while ( van der Schaft 1986 ) looks at the action of the dynamics on the functions in a similar manner, leading to, essentially, a version of the Koopman formalism ( Budišić et al. 2012 ) . The recent work ( Otto et al. 2023 ) considers data driven state space recovery using Machine Learning methods. Among other works exploring such model-free approaches one should mention ( Schmidt and Lipson 2009 ; Brunton et al. 2016 ; Champion et al. 2019 ) . 2 Mathematical Background The motivation for our theory comes from the beautiful work of Cohen, Jones, and Segal ( Cohen et al. 1995 ) , which we then couple with the well-established apparatus of the Topological Data Analysis (TDA) ( Edelsbrunner and Harer 2010 ) . 2.1 Basic Definitions Here is their setup. Assume that M is a smooth closed Riemannian manifold and f:M\rightarrow\mathbb{R} a smooth Morse function on M . The key idea of ( Cohen et al. 1995 ) is to describe a combinatorial construction that reconstructs the topology of M from the space of flow lines of the gradient vector field \nabla f . To this end, they introduced the enriched category 1 1 1 The language of category theory (which can be looked up in, say, ( Riehl 2017 ) ) can be avoided, but for the space reasons we just follow the original presentation. Enriched category here just means that the morphisms between