Data-driven Control of Hypergraphs: Leveraging THIS to Damp Noise in Diffusive Hypergraphs
Robin Delabays, Yuanzhao Zhang, Florian Dörfler, Giulia De Pasquale · 2025
Read Data-driven Control of Hypergraphs: Leveraging THIS to Damp Noise in Diffusive Hypergraphs by Robin Delabays, Yuanzhao Zhang, Florian Dörfler, Giulia De Pasquale (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-2511-08647.txt.
Introduction
Robin Delabays and four co-authors ask how to steer systems whose members interact in groups, not just pairs. Graphs miss such higher-order links, so the work uses hypergraphs, and it starts from a harder spot: the interaction structure itself may be unknown and hard to measure.
The pipeline first infers the hypergraph from observed behavior, then picks which nodes can be controlled, and finally drives the system toward a chosen state. For polynomial dynamics the authors prove analytical controllability guarantees. For other dynamics they give a heuristic for choosing control nodes and test it on Kuramoto oscillators, the standard model of coupled rhythms. The approach targets settings from biological collectives to engineered infrastructure.
Catalog record
| CREATOR | Robin Delabays, Yuanzhao Zhang, Florian Dörfler, Giulia De Pasquale |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-08647
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Data-driven Control of Hypergraphs: Leveraging THIS to Damp Noise in Diffusive Hypergraphs Robin Delabays, Yuanzhao Zhang, Florian Dörfler, Giulia De Pasquale Abstract Controllability determines whether a system’s state can be guided toward any desired configuration, making it a fundamental prerequisite for designing effective control strategies. In the context of networked systems, controllability is a well-established concept. However, many real-world systems, from biological collectives to engineered infrastructures, exhibit higher-order interactions that cannot be captured by simple graphs. Moreover, the way in which agents interact and influence one another is often unknown and must be inferred from partial observations of the system. Here, we close the loop between a hypergraph representation and our recently developed hypergraph inference algorithm, THIS, to infer the underlying multibody couplings. Building on the inferred structure, we design a parsimonious controller that, given a minimal set of controllable nodes, steers the system toward a desired configuration. We validate the proposed system identification and control framework on a network of Kuramoto oscillators evolving over a hypergraph. I Introduction Motivation. Networked systems provide a powerful framework for describing the collective dynamics of interacting agents, from power grids [ 1 ] to biological [ 2 ] and social systems [ 3 ] . However, in many real systems, interactions that naturally occur among groups of agents cannot be reduced to a set of pairwise interactions, for example, in neuronal assemblies [ 4 ] , ecological communities [ 5 ] , and social groups [ 6 ] where the state of an individual depends simultaneously on multiple others. These higher-order interactions are more accurately captured by hypergraphs or simplicial complexes, which generalize networks by allowing multi-body connections [ 7 ] . The ability to understand and control such systems is crucial for ensuring stability and robustness. Yet, in many practical scenarios, the underlying interaction structure is not directly observable and must be inferred from limited data. This lack of full structural knowledge challenges traditional control approaches, which typically assume a known network topology. By integrating structure inference and control design within a unified framework, we aim to develop principled tools for identifying and steering higher-order dynamical systems, enabling effective interventions even when the governing interactions are only partially known. More concretely, extensions of the well studied Kuramoto model to include three-body or higher-order interactions have revealed that these couplings can make the synchronized state’s basin of attraction smaller yet deeper, raising the critical coupling threshold while enhancing robustness once synchronization is achieved [ 8 ] . These findings highlight that higher-order structures fundamentally reshape the landscape of collective dynamics. The coupling strength plays a central role in determining the collective behavior of oscillator networks [ 9 ] . It governs the transition from incoherence to synchrony and critically shapes the stability, robustness, and transient dynamics of the ensemble. However, in most practical scenarios, coupling strengths are not directly measurable, as the interactions between oscillators are often mediated by latent physical, chemical, or informational processes. Instead, only phase measurements are typically accessible. This limitation motivated us to design a closed loop algorithm between our proposed Taylor-based Hypergraph Inference using SINDy (THIS) algorithm [ 10 ] , able to estimate effective coupling parameters and interaction structures from time series data together with a control law in order to drive the system towards a desired state. Such approach, allows to identify hidden connectivity patterns, assess synchronization regimes, and design control strategies in real-world networks where the underlying coupling mechanisms remain partially unknown. Literature Review. The notion of structural controllability for linear time invariant systems evolving over network traces back to the 70s [ 11 ] with the work of Lin on ”structural controllability”. Both algebraic and graph theoretic conditions were provided to ensure that a system can be driven towards any point of the state space based solely on topological considerations. Further analytical tools to certify structural controllability of a network have been provided in [ 12 ] , where the authors identify the set of driver nodes with time-dependent control that can drive the system dynamics. The extension of the notion of controllability to hypergraphs is much more recent with Refs. [ 13 , 14 , 15 ] , which relies on tensor algebra and polynomial control theory [ 16 ] . They provide theoretical results about the minimum number of control nodes to be controlled given specific hypergraph structures. Contributions. We leverage our recently proposed identification algorithm, Taylor-based Hypergraph Inference using SINDy (THIS) [ 10 ] , to infer coupling strengths in a network of coupled oscillators. Building on this inferred model, we design a simple yet effective proportional control strategy that enables any system evolving over a hypergraph to reach a desired fixed point. In a similar spirit as Ref. [ 12 ] , we introduce a notion of a minimal set of controllable nodes , allowing us to design a parsimonious controller that acts only on a minimal subset of nodes to drive the system’s dynamics. Finally, we demonstrate the proposed identification and control framework on a hypergraph of Kuramoto oscillators, successfully steering them toward the synchronous state. Article Organization. The paper is organized as follows. Section II introduces the notation and preliminary concepts used throughout the paper. Section III presents the problem formulation for the design of a parsimonious state-feedback controller. Section IV details the proposed algorithm design, while Section V demonstrates its effectiveness on a hypergraph of Kuramoto oscillators. Finally, Section VI concludes the manuscript. II Notation and Preliminaries We provide the notation and preliminary concepts that will be used throughout the manuscript. II-A Hypergraphs For a matrix A\in\mathbb{R}^{n\times n} , let a_{i,j} be its (i,j) -th entry. A p -th order hypergraph \mathcal{H} is a tuple \mathcal{H}(V,E,A^{(2)},...,A^{(p)}) , where V=\{1,\dots,n\} indicates the set of nodes, E^{(k)}\subset V^{k} is the set of k -th order (hyper)edges and E=\bigcup_{k}E^{(k)} is the set of all (hyper)edges, and A^{(k)}\in\mathbb{R}^{n\times\dots\times n} is the k -th order adjacency tensor , such that a k -hyperedge (i_{1},i_{2},\dots,i_{k}) is present in the hypergraph if and only if A^{(k)}_{i_{1},i_{2},\dots,i_{k}}\neq 0 . In this work, we consider directed hypergraphs, meaning that each permutation of k indices (i_{1},\dots,i_{k}) denotes a distinct directed hyperedge. The tuple e=(i_{1},i_{2},\dots,i_{k}) indicates that nodes i_{2},\dots,i_{k} all influence node i_{1} ’s dynamics. We say that the node at the first index is the head of a hyperedge and the other nodes are the tails , e.g., in e=(i_{1},i_{2},\dots,i_{k}) , node i_{1} is the head of e , and i_{2},\dots,i_{k} are the tails. We define a sequence of edges (e_{1},...,e_{\ell}) is a directed path from node i to node j if i is the head of e_{1} , j is a tail of e_{\ell} , and the head of e_{k} is one of the tails of e_{k-1} , for each k\in\{2,...,\ell\} . A node is said to be a leaf if it is not the head of any edge. For instance, in the middle panel of Figure 1 , node 2 is a leaf, while nodes 1 and 3 are not. A hypergraph is said to be connected if for any two nodes i and j , there is either a path from i to j or a path from j to i . Fig. 1: