Network robustness assessed within a dual connectivity perspective

Network robustness against attacks has been widely studied in fields as diverse as the Internet, power grids and human societies. Typically, in these studies, robustness is assessed only in terms of the connectivity of the nodes unaffected by the attack. Here we put forward the idea that the connectivity of the affected nodes can play a crucial role in properly evaluating the overall network robustness and its future recovery from the attack. Specifically, we propose a dual perspective approach wherein at any instant in the network evolution under attack, two distinct networks are defined: (i) the Active Network (AN) composed of the unaffected nodes and (ii) the Idle Network (IN) composed of the affected nodes. The proposed robustness metric considers both the efficiency of destroying the AN and the efficiency of building-up the IN. We show, via analysis of both prototype networks and real world data, that trade-offs between the efficiency of Active and Idle network dynamics give rise to surprising crossovers and re-ranking of different attack strategies, pointing to significant implications for decision making.

💡 Research Summary

The paper addresses a fundamental shortcoming in most network‑robustness studies: the exclusive focus on the connectivity of the nodes that remain functional after an attack (the “Active Network”, AN). The authors argue that the structure of the nodes that have already been removed or disabled (the “Idle Network”, IN) can be just as important for assessing overall robustness and for planning recovery actions. To capture this dual perspective, they introduce a framework in which, at any moment during an attack, the original graph is partitioned into two disjoint sub‑graphs: AN, consisting of all still‑operational nodes and the edges among them, and IN, consisting of all affected nodes together with any edges that still exist between them.

Two efficiency measures are defined. The first, (E_A), quantifies how quickly the AN deteriorates; it aggregates the relative changes in average shortest‑path length, average clustering coefficient, and the number of connected components as nodes are removed. The second, (E_I), quantifies how rapidly the IN becomes internally cohesive; it tracks the increase in edge density and clustering within the IN as each new node joins it. A single robustness metric is then constructed as a weighted sum
\