Networks beyond pairwise interactions: structure and dynamics

Federico Battiston, Giulia Cencetti, Iacopo Iacopini, Vito Latora, Maxime Lucas, Alice Patania, Jean-Gabriel Young, Giovanni PetriarXiv 2020 · arXiv

Introduction

For decades, network science has been dominated by the study of graphs where edges connect exactly two nodes. While this approach has yielded significant insights, it creates a "dyadic bottleneck" that oversimplifies the true nature of interactions in real-world systems. In social groups, ecological communities, and biochemical pathways, interactions often involve three or more entities simultaneously. This review paper serves as a comprehensive "state of the union" for higher-order network science, arguing that capturing these multi-body interactions is essential for understanding the complex behaviors of biological, social, and technological systems.

Mathematical Frameworks for Higher-Order Systems

To move beyond pairwise links, researchers have adopted several mathematical frameworks that can represent group interactions. The two primary approaches are simplicial complexes and hypergraphs.

Comparison of pairwise and higher-order network representations.

Simplicial Complexes are derived from algebraic topology. They represent interactions as simplices: a 0-simplex is a node, a 1-simplex is a link, a 2-simplex is a filled triangle, and so on. A defining feature of simplicial complexes is the requirement of "downward closure"—if a group of nodes forms a simplex, all its subsets must also be present as simplices. This structure allows for the use of powerful tools like Topological Data Analysis (TDA) and persistent homology to identify "holes" or cavities in high-dimensional data.

Hypergraphs offer a more flexible alternative. In a hypergraph, a "hyperedge" can connect any number of nodes without the restriction that all sub-groups must also interact. This makes hypergraphs particularly useful for modeling systems where group interactions are primary and do not necessarily imply lower-order relationships, such as group chats or specific chemical reactions.

Dynamical Phenomena: Beyond Smooth Transitions

One of the most profound insights from higher-order network science is how these structures fundamentally alter system dynamics. In traditional pairwise networks, phase transitions—such as the onset of an epidemic or synchronization—are typically smooth and continuous. However, in higher-order systems, these transitions often become "explosive" or discontinuous.

For example, in models of simplicial contagion, 3-body "social reinforcement" can create bistable regions. Unlike standard models where an infection spreads gradually, a higher-order system might remain healthy until a "critical mass" of group-based spreading is reached, at which point the infection explodes into an epidemic. Similarly, in higher-order Kuramoto models of coupled oscillators, the transition to global synchronization can occur instantly, jumping from total disorder to perfect coherence.

Empirical Evidence and Applications

The paper reviews a vast array of empirical datasets where higher-order interactions are critical:

  • Social Systems: Co-authorship networks and social affiliation data are naturally represented as hypergraphs, where a single hyperedge represents a paper with multiple authors or an event with multiple attendees.
  • Neuroscience: Higher-order methods have been used to analyze brain connectivity and co-activation patterns in fMRI and EEG data, revealing topological features that pairwise methods miss.
  • Ecology: In "web of life" models, higher-order interactions (e.g., a predator's behavior changing in the presence of a third species) can stabilize ecosystems that would otherwise be unstable under purely pairwise competition.

Challenges and Future Outlook

Despite the promise of higher-order network science, several hurdles remain. Inference is a primary challenge: it is often difficult to determine from observational data whether a group interaction is truly higher-order or merely a collection of simultaneous pairwise links. Furthermore, the computational complexity of working with tensors and high-dimensional topological features currently limits the scalability of these methods to massive datasets.

As the field moves forward, the focus is shifting toward "Dynamical Topology," where the high-dimensional "shape" of a system is treated as an evolving entity. By breaking the dyadic bottleneck, researchers are uncovering a richer landscape of complexity that more accurately reflects the interconnected world.

Summary updated

BibTeX
@misc{battiston2020networks,
  author = {Federico Battiston and Giulia Cencetti and Iacopo Iacopini and Vito Latora and Maxime Lucas and Alice Patania and Jean-Gabriel Young and Giovanni Petri},
  title = {{Networks beyond pairwise interactions: structure and dynamics}},
  year = {2020},
  eprint = {2006.01764},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2006.01764},
}
Figure