Communication Melting in Graphs and Complex Networks

TL;DR

This paper introduces a universal phase transition in the communicability of complex networks, analogous to melting in solids, revealing how network structure and stress influence connectivity loss and diffusive processes.

Contribution

It analytically demonstrates a universal melting-like phase transition in the communicability of all simple graphs, linking network topology to dynamical phase changes under stress.

Findings

Regular-like graphs melt at lower temperatures with sharper transitions.

Random graphs show more gradual melting, resembling amorphous solids.

Node melting is driven by eigenvector centrality, especially near critical points.

Abstract

Complex networks are the representative graphs of interactions in many complex systems. Usually, these interactions are abstractions of the communication/diffusion channels between the units of the system. Real complex networks, e.g. traffic networks, reveal different operation phases governed by the dynamical stress of the system. Here we show how, communicability, a topological descriptor that reveals the efficiency of the network functionality in terms of these diffusive paths, could be used to reveal the transitions mentioned. By considering a vibrational model of nodes and edges in a graph/network at a given temperature (stress), we show that the communicability function plays the role of the thermal Green's function of a network of harmonic oscillators. After, we prove analytically the existence of a universal phase transition in the communicability structure of every simple graph. This transition resembles the melting process occurring in solids. For instance, regular-like graphs resembling crystals, melts at lower temperatures and display a sharper transition between connected to disconnected structures than the random spatial graphs, which resemble amorphous solids. Finally, we study computationally this graph melting process in some real-world networks and observe that the rate of melting of graphs changes either as an exponential or as a power-law with the inverse temperature. At the local level we discover that the main driver for node melting is the eigenvector centrality of the corresponding node, particularly when the critical value of the inverse temperature approaches zero. These universal results sheds light on many dynamical diffusive-like processes on networks that present transitions as traffic jams, communication lost or failure cascades.