The topology of synergy: linking topological and information-theoretic approaches to higher-order interactions in complex systems

Thomas F. Varley, Pedro A. M. Mediano, Alice Patania, Josh C. Bongard

Abstract

Understanding the "physics of emergence" requires a robust framework for characterizing higher-order interactions in complex systems. This paper, authored by Thomas Varley and colleagues, establishes a novel bridge between Topological Data Analysis (TDA) and Multivariate Information Theory. By analyzing both synthetic manifolds and real-world fMRI data, the researchers demonstrate that synergistic information—information present in a system as a whole but not in its individual parts—corresponds to specific geometric signatures, namely multi-dimensional cavities. Conversely, redundancy is associated with knotted, locally linear structures. These findings reveal a fundamental "blind spot" in standard dimensionality reduction techniques like Principal Component Analysis (PCA), which systematically overlook synergistic structures as noise.

Introduction

Complex systems, ranging from the neural architectures of the human brain to the intricate dynamics of global economies, are defined by interactions that transcend simple pairwise relationships. These higher-order interactions (HOIs) are the bedrock of emergent phenomena. Historically, two powerful but disparate frameworks have attempted to map these interactions: Topological Data Analysis (TDA), which identifies the "shape" of data through algebraic topology, and Multivariate Information Theory, which quantifies the flow and storage of information.

Despite their shared goal of understanding system-wide dependencies, these fields have largely evolved in isolation. This research seeks to unify them, testing the hypothesis that the abstract concept of "synergy" has a concrete, topological representation. The results suggest that synergy is not merely a statistical curiosity but is intrinsically linked to the multi-dimensional "voids" or cavities within a data manifold.

The Problem: The Pairwise Bias

Most standard statistical tools, including correlation networks and Principal Component Analysis (PCA), are built on pairwise logic. They excel at identifying redundancy—where information is duplicated across multiple variables—but are often "blind" to synergy. Synergy occurs when the relationship between three or more elements provides information that cannot be gleaned from any subset of those elements. Because our current tools treat these complex, non-linear structures as high-dimensional noise, we may be missing the most critical features of data in fields like neuroscience and systems biology.

Methodological Framework

To bridge the gap between topology and information, the authors conducted a "head-to-head" comparison using two primary metrics:

  1. O-information: A metric that quantifies the balance between redundancy and synergy. To handle continuous, non-linear data without assuming Gaussian distributions, the authors implemented a non-parametric K-nearest neighbor (KNN) estimator.
  2. Persistent Homology: A TDA tool used to identify 3D cavities (Betti numbers) in data point clouds via Vietoris-Rips filtration.

Crucially, the study employed a "Rotation Test" using PCA. By comparing data before and after rotation, the researchers could distinguish between "contextual" information (dependent on the data's orientation) and "intrinsic" information (inherent to the data's topological shape).

Experimental Design

The study utilized two distinct data sources to validate its findings:

  • Synthetic Manifolds: Toy models with known topologies, such as hollow spheres (representing voids), solid toroids, and complex knots (e.g., trefoil knots), allowed for controlled testing of topological signatures.
  • Neuroimaging Data: Resting-state fMRI BOLD signals from the Human Connectome Project (HCP). The authors analyzed thousands of "triads" (sets of three brain regions) to determine how their informational properties correlated with their topological structures.

Key Insights: The Shape of Information

Synergy as "Hollowness"

The most significant finding is the link between synergy and 3D cavities. A hollow sphere, for instance, is synergy-dominated, whereas a solid ball is not. This suggests that synergy arises when a system is "forbidden" from occupying certain states—represented by the empty center of the sphere—forcing a global coordination among variables that cannot be reduced to local interactions.

Redundancy as "Knottedness"

In contrast, redundancy is associated with "knotted" structures. In these manifolds, variables are highly constrained by local linearity. This creates high redundancy that persists even when the data is projected into lower dimensions, as the variables are essentially duplicating the same underlying signal.

The PCA Blind Spot

The research highlights a critical limitation of PCA. While PCA is the "gold standard" for dimensionality reduction, it is fundamentally biased toward redundancy. It treats synergistic structures—the very "holes" that define complex coordination—as random variance. When researchers use PCA to simplify high-dimensional data, they may be inadvertently "flattening" the topological features that contain the system's most vital synergistic information.

Results and Empirical Evidence

The analysis of fMRI data provided strong empirical support for these theoretical links. The authors observed a significant negative correlation between normalized O-information and the presence of 3D voids: as the number and size of cavities increased, so did the synergy of the brain region triads. Furthermore, while redundant triads were highly compressible (with a single principal component explaining most of the variance), synergistic triads appeared incompressible to PCA, despite possessing clear, non-random higher-order structures.

Discussion and Future Directions

While this study provides compelling evidence for a link between synergy and topology, it remains an exercise in "experimental mathematics." A formal analytical proof linking O-information to Betti numbers is still required to solidify this bridge. Additionally, the computational complexity of TDA and non-parametric estimators currently limits analysis to small sets of variables (triads).

The implications for future research are profound. There is an urgent need for new "topology-aware" manifold learning techniques—tools that can reduce the dimensionality of data while preserving the synergistic cavities that characterize complex system behavior. By moving beyond pairwise statistics, we can begin to see the true "shadow structure" of higher-order interactions that defines the natural world.

Summary updated

BibTeX
@article{varley2025topology,
  author = {Thomas F. Varley and Pedro A. M. Mediano and Alice Patania and Josh C. Bongard},
  title = {{The topology of synergy: linking topological and information-theoretic approaches to higher-order interactions in complex systems}},
  journal = {PLOS Computational Biology},
  year = {2025},
  doi = {10.1371/journal.pcbi.1013649},
}
Figure