Revealing dynamics of non-autonomous complex systems from data
Chengzuo Zhuge, Zheng Jiang, Zhefan Xu, Wei Chen · 2026
Read Revealing dynamics of non-autonomous complex systems from data by Chengzuo Zhuge, Zheng Jiang, Zhefan Xu, Wei Chen (2026) 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-2605-13878.txt.
Introduction
Zhuge, Jiang, Xu and Chen build a data-driven method that recovers the governing equations of systems whose rules change over time. The approach picks an optimal set of basis functions, so it needs only simple prior assumptions. Tests on synthetic cases, including coupled Kuramoto oscillators, succeed, and the method also predicts cases such as drone navigation, chick-heart aggregates and a marine fish community.
Catalog record
| CREATOR | Chengzuo Zhuge, Zheng Jiang, Zhefan Xu, Wei Chen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63617 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.13878
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Revealing dynamics of non-autonomous complex systems from data Chengzuo Zhuge, Zheng Jiang, Zhefan Xu, Wei Chen Abstract Discovering governing equations from data is crucial for understanding complex systems in many diverse fields from science to engineering. Yet, there still is a lack of versatile computational toolbox to deal with this long standing challenge due to the inherent non-autonomicity and unknowability of the underlying dynamics. Here, we introduce a data-driven approach for inferring non-autonomous dynamical equations by identifying an optimal set of basis functions within the model space, enabling the reconstruction of complex systems behavior under simplified prior specifications. Our method demonstrates effectiveness in equation discovery on canonical synthetic systems such as cusp bifurcation and coupled Kuramoto oscillators. Furthermore, we extend the application of this approach to leaf cellular energy, unmanned aerial vehicle navigation, chick-heart aggregates, and marine fish community under simple basis function libraries. Leveraging the inferred equations, we accurately predict the evolution of these empirical systems and further uncover their governing laws. Our approach offers a novel paradigm to reveal the underlying dynamics of a wide range of real-world systems. Introduction From cellular metabolism ( 15 , 62 ) , to animal population migration ( 33 , 48 ) , to autonomous driving of vehicles ( 42 , 65 ) , the intricate evolution of numerous real-world systems is fundamentally governed by unknown system dynamics. Given the critical importance of the governing equations underlying a system’s dynamical behavior for understanding and predicting its evolution, several advanced methodologies have been proposed in recent years to uncover the dynamical equations from observed data ( 7 , 47 , 43 , 19 , 59 , 14 , 25 , 24 , 44 , 22 ) . These methods typically define a model space that potentially contains the system’s latent dynamics through the linear combinations of predefined basis functions. Then the optimal combination in this model space is learned from observed data to capture the governing dynamics, for example, by applying sparse regression to find parsimonious model ( 7 ) , or by integrating local fine-tuning with global regression to derive compact equations ( 25 ) . Despite these advancements, most existing methodologies have focused on autonomous systems, there still is a lack of systematic investigation of non-autonomous systems, particularly those involving network topologies ( 60 , 40 , 32 , 54 , 66 , 11 ) . However, real-world systems are predominantly non-autonomous, with the evolution of their states often being subject to external time-varying parameters ( 13 ) . For instance, population dynamics are not only governed by internal interactions such as mutualism and competition, but also driven by external factors like temperature fluctuations ( 37 , 23 ) . Such non-autonomicity leads to the emergence of particularly noteworthy dynamical behaviors, such as critical transitions ( 55 , 58 ) and seasonal patterns ( 28 , 45 ) . To extend existing methodologies to non-autonomous systems, current approaches typically augment the candidate library to include not only functions of the system’s state variables but also those of the forcing parameters and their coupled terms ( 5 , 35 , 26 ) . However, this framework faces several fundamental challenges when applied to real-world non-autonomous dynamics. First, the forcing parameters in real-world systems often exhibit unobservability. Consequently, their values are not available for assigning to the basis functions, and thus the system’s governing equations cannot be inferred in most cases. As a remedy, an intuitive strategy for modeling non-autonomous dynamics is to replace forcing parameters with a readily observable time variable within the basis functions ( 7 ) . Yet, the validity of such a simplification in the basis functions has not been systematically verified. It remains unclear what forms of the time variable in the basis functions are valid, while invalid basis functions can directly lead to biased dynamics inference. Second, the dynamics of real-world systems are highly complex and not fully knowable, especially for non-autonomous dynamics with intricate signal couplings. Current methods address this issue by identifying the optimal combination of basis functions from an extensive candidate library to reconstruct system dynamics ( 25 , 24 ) . Such a library comprises a set of basis functions whose selection is heuristic and often relies on expert knowledge. However, the search for the optimal combination of basis functions within a broad model space incurs expensive computational costs ( 25 , 24 , 44 ) . Meanwhile, the heuristically selected basis functions may still lack sufficient completeness to span a model space that contains the true governing equation of real-world system. As a consequence, existing methodologies struggle to extend their applicability to non-autonomous dynamics inference in real-world settings. These challenges necessitate a framework capable of inferring dynamical equations without rigid dependence on the completeness of predefined libraries or accessibility of forcing parameters. In this work, we introduce a novel paradigm that shifts the focus from merely finding the optimal linear combination of predefined basis functions to adaptively constructing the basis functions themselves within the given model space. Specifically, we first prove a theorem for the basis functions of non-autonomous systems, which provides a theoretical guidance for developing a method to infer equations for complex system dynamics. The theorem states that, within a given model space, there exist infinitely many sets of basis functions, each parameterized by distinct driving variables, to represent the multi-parameter driven equations hidden in this space. Based on the theorem, we propose an approach for accurately inferring equations by adaptively identifying an optimal set of basis functions within a given model space. These optimal basis functions, identified through a unique driving variable detected from the observed data, introduce minimal fitting and numerical errors in the process of equation inference. Such improvements in the basis functions enable the uncovering of non-autonomous dynamics even when the forcing parameters are inaccessible, and promise accurate reconstruction of the true dynamics without requiring rigid prior knowledge of the system dynamics. To validate our method’s equation discovery capability, we employ two canonical synthetic systems: the cusp bifurcation normal form ( 39 ) and the coupled Kuramoto oscillators ( 38 ) . The governing equations obtained via the optimal basis functions achieve higher fidelity to the ground truth compared to those obtained using basis functions that depend on forcing parameters or time variable. This improvement, observed in both the recovered functional forms and coefficient estimates, remains robust as the size of the basis function library increases. Notably, even if key functional terms of the true equations are omitted in the basis function library, our method consistently recovers the system dynamics through a surrogate model, where comparable approaches fail. Moreover, when dealing with real-world systems where the true underlying equations are highly intricate and not fully knowable, our method can still accurately infer representations of the underlying dynamics under simple basis function libraries. We demonstrate this capability on non-autonomous complex systems from diverse real-world domains, including leaf cellular energy status ( 62 ) , unmanned aerial vehicle (UAV) navigation, spontaneously beating chick-heart aggregates ( 8 ) , and natural marine fish community ( 60 ) . Leveraging the inferred equations, we not only generate accurate predictions for the evolution of these empirical systems, but our further analysis