Source-linked AI summary
Catastrophic Cascade of Failures in Interdependent Networks
S. Havlin, N. A. M. Araujo, S. V. Buldyrev, C. S. Dias, R. Parshani, G. Paul, H. E. Stanley
TL;DR
Interdependent networks pose a vulnerability problem because failures can cascade across coupled systems, whose properties differ from single networks. This article reviews analytical and numerical results on their percolation properties, finding that coupling increases vulnerability and can change continuous transitions into abrupt first-order transitions.
Problem
Interdependent networks were only recently studied, leaving their percolation properties insufficiently characterized despite their technological and multidisciplinary relevance.
Method
The paper reviews analytical and numerical results on interdependent-network percolation, including bidirectionally coupled networks subject to random node removal and cascading failures.
Results
Coupling increases vulnerability by raising the percolation threshold and produces first-order, abrupt transitions instead of the continuous transitions found in single networks.
Takeaways & Limitations
The percolation behavior of coupled networks differs substantially from single networks, with stronger coupling requiring fewer removed nodes to fragment the mutually connected giant component.
Abstract
from arXiv · showhide
Modern network-like systems are usually coupled in such a way that failures in one network can affect the entire system. In infrastructures, biology, sociology, and economy, systems are interconnected and events taking place in one system can propagate to any other coupled system. Recent studies on such coupled systems show that the coupling increases their vulnerability to random failure. Properties for interdependent networks differ significantly from those of single-network systems. In this article, these results are reviewed and the main properties discussed.
G. Paul
G. Paul is affiliated with the Center for Polymer Studies and the Department of Physics at Boston University in Boston, Massachusetts, USA.
- G. Paul is affiliated with the Center for Polymer Studies.
- G. Paul is affiliated with Boston University’s Department of Physics.
- The affiliation is located in Boston, Massachusetts 02215, USA.
H. E. Stanley
This section identifies H. E. Stanley and gives his Boston University affiliation in Boston, Massachusetts, USA.
- H. E. Stanley is associated with the Center for Polymer Studies and Department of Physics at Boston University.The affiliation is listed in Boston, Massachusetts 02215, USA.
- The listed institutional affiliation is located in Boston, Massachusetts 02215, USA.
- The page carries a copyright mark attributed to Società Italiana di Fisica.
1. – Introduction
The introduction frames modern systems as interdependent networks whose failures can propagate across coupled infrastructures and other domains. It emphasizes that bidirectional coupling can produce catastrophic cascades rather than isolated disruptions.
- Motivation: Complex-network research increasingly examines robustness to random failures and malicious attacks, beyond the earlier focus on single isolated networks.Real-world examples include the Internet, airport connections, and power grids.
- Motivation: Infrastructure networks such as oil, transportation, electric power, natural gas, water, and telecommunications are bidirectionally interdependent.Electrical power both supports and depends on the other infrastructures.
- Motivation: A failure in one interdependent network can trigger a cascade in which dependent components become disconnected and inactive across the coupled system.The example describes inactive servers followed by shutdown and disconnection of power stations.
- Motivation: Interdependent networks also span economy and biology, including banks, insurance companies, firms, and mutually dependent human-body systems.These interconnections played a major role in the recent financial crisis.
2. – Single Network Robustness
Single-network robustness is characterized through percolation and differs between Erdős–Rényi and scale-free topologies. Both exhibit continuous second-order transitions, but scale-free networks with 2 ≤ γ ≤ 3 have no finite percolation threshold and are especially robust to random failure.
- Percolation framework: The section examines percolation by randomly removing a fraction 1−p of nodes and tracking the giant component as the order parameter.Percolation concerns the emergence or disappearance of a giant cluster as nodes or links are added or removed.
- Network models: The analysis focuses on Erdős–Rényi random graphs and Barabási–Albert scale-free networks, whose different degree distributions produce different critical properties.ER networks have Poissonian degree distributions, whereas BA networks have power-law degree distributions with highly connected nodes.
- Erdős–Rényi networks: pc = 1/<k> for ER networks, where the percolation transition is second order.The giant component vanishes when the removed-node fraction exceeds 1−pc in the thermodynamic limit.
- Scale-free networks: pc = 0 for scale-free networks with 2 ≤ γ ≤ 3, so no percolation transition occurs.For γ above 3, a second-order transition returns, but with critical exponents different from those of ER networks.
- Robustness comparison: Both network types undergo second-order transitions with smooth order-parameter decreases, and their low pc values indicate robustness to random failure.The paper contrasts this robustness with the behavior of interdependent systems discussed later.
3. – Interdependent Networks Robustness
Coupling creates percolation behavior unlike that of single networks: cascading failures make interdependent systems more vulnerable, with larger thresholds and abrupt fragmentation. For coupled Erdős–Rényi and scale-free networks, a mutually connected giant component can disappear discontinuously after limited node removal.
- Model and cascade mechanism: Bidirectional one-to-one dependencies require each coupled node pair to remain connected to both networks’ giant components, so failures propagate across networks.Removing an A-node also removes its links and coupled B-node; nodes disconnected from either giant cluster become inactive with their counterparts.
- Erdős–Rényi networks: Below the threshold, the entire interdependent system becomes completely fragmented, while above it a giant mutually connected cluster exists.Near pc, cascades can remain at a plateau for many stages before the giant component suddenly drops to zero.
- Erdős–Rényi networks: At pc, the average number of cascade stages scales with N^1/4.
- Scale-free networks: Coupled scale-free networks exhibit a nonzero percolation threshold even for 2 < γ ≤3, unlike single networks, and coupling increases vulnerability through larger pc.The cascade mechanism can be activated by removing a small fraction of nodes, including when hubs depend on low-degree nodes.
- Transition character: Interdependent networks undergo an abrupt, first-order transition at larger pc, whereas single networks show a smooth, second-order transition.The discontinuity in the order parameter reflects cascade-driven collapse in coupled systems.
4. – Partially Dependent Networks
Partially dependent networks model bidirectional dependency through fractions qA and qB, interpolating between autonomous and fully interdependent systems. Increasing coupling raises vulnerability and can change the percolation transition from continuous to discontinuous, while a critical point separates abrupt and smooth behavior.
- Model: The generalized model assigns fractions qA and qB of nodes in networks A and B dependencies on the other network, recovering full interdependence when qA = qB = 1.Autonomous nodes account for cases such as servers protected by emergency power supplies.
- Cascade process: Random node removal triggers cascading failures by removing dependent nodes and nodes connected to giant components only through removed nodes.The mutually connected giant component P∞ is evaluated as a function of p.
- Percolation transitions: Reducing coupling produces a second-order transition, whereas increasing coupling raises pc and changes the transition from second-order under weak coupling to first-order under strong coupling.For strong coupling q = 0.8, robustness resembles the fully interdependent limit q = 1; weak coupling corresponds to q = 0.1.
- Cascade dynamics: In strong coupling, cascade iterations show a plateau followed by an abrupt decrease of the giant component, consistent with a first-order transition.The comparison uses two coupled Erdős-Rényi networks with NA = NB = 8 × 10^5 and equal average degree.
- Phase diagram: As the fraction of independent nodes increases, the transition requires more node removals and the system becomes less vulnerable; above a critical point, the transition becomes smooth and second-order.With no independent nodes, only a small fraction of removals can fragment the system discontinuously.
5. – Final remarks
The review concludes that interdependence significantly alters percolation by increasing transition thresholds and changing transition order. Bidirectional node dependencies can trigger catastrophic cascades and abruptly decompose the mutually connected giant component.
- Final remarks: Bidirectional coupling significantly affects percolation properties, increasing the transition threshold and changing the transition order.These effects distinguish interdependent networks from single-network systems.
- Final remarks: Node interdependency means that failure of one node can disable its counterpart, igniting cascades that abruptly decompose the mutually connected giant component.For two interconnected ER graphs, random node removal produces a percolation transition.
- Final remarks: Stronger coupling lowers the fraction of nodes that must be removed to fragment the giant component and changes the transition from second to first order.Weak coupling yields a second-order transition, whereas strong coupling yields a first-order transition.
- Final remarks: The review identifies future questions concerning coupled network types, real networks, inter-network connections, and rewiring to improve failure resilience.These directions are presented as natural follow-up problems raised by the reported results.