Summary
Curvature is a way to understand shapes and connections, not just in physical spaces but in networks like social media or transport systems. This paper compares two types of curvature called Steinerberger curvature (SBC) and resistance curvature (RC) to see how they relate to other known measures that describe network structure. The authors find that SBC behaves differently from other measures, capturing unique features, while RC is closely related to common curvature methods and is good at identifying important nodes that keep a network connected. These insights can help detect critical points in networks that, if removed, would greatly affect communication or connectivity.
What this means in practice
- •For network engineers: Identify critical nodes in communication networks whose failure would rapidly reduce network efficiency.
- •For infrastructure planners: Design more resilient transportation or utility networks by understanding nodes that strongly affect overall connectivity.
Authors
Madhumita Mondal, Priyank Vasu, Areejit Samal, Jürgen Jost
Abstract
Discrete curvature extends the concept of curvature from continuous spaces to discrete structures such as graphs and networks. It provides a mathematical framework for characterising local and global network geometry, that offers insights into structural properties such as connectivity and clustering. In this study, we investigate Steinerberger curvature (SBC) and resistance curvature (RC) and examine their relationships with established graph curvature measures, namely Forman-Ricci, Augmented Forman-Ricci, and Ollivier-Ricci curvatures, a notion of scalar sectional curvature, as well as conventional vertex-based measures. A systematic comparative analysis is conducted across various synthetic network models and real-world networks, focusing on curvature distributions, correlations with structural measures, and the identification of structurally important vertices through network disintegration experiments based on global efficiency. Our results show that SBC generally exhibits weaker correlations with other measures. On the other hand, RC is strongly positively correlated with established discrete Ricci curvature measures, whereas it is predominantly negatively correlated with conventional vertex-based measures like degree and centrality-based measures. Furthermore, RC achieves a reduction in communication efficiency comparable to or faster than established Ricci curvature measures. These findings suggest that SBC captures structural characteristics that are largely distinct from those reflected by other measures, while RC exhibits distinct structural associations and shows significant potential for identifying vertices critical to network connectivity.