Limits of Predictability of Cascading Overload Failures in Spatially-Embedded Networks with Distributed Flows.

Area of Science:

• Network science
• Complex systems analysis
• Infrastructure resilience

Background:

• Cascading failures pose significant risks to complex information and infrastructure networks.
• Understanding load-based failures is crucial for network robustness.

Purpose of the Study:

• Investigate load-based cascading failures in real and synthetic networks.
• Propose and evaluate mitigation strategies for network failures.
• Analyze the impact of resource distribution and network stability on failure propagation.

Main Methods:

• Developed a stochastic method for optimal heterogeneous resource distribution (node capacities).
• Designed and compared N-stable and (N-1)-stable network-capacity allocations.
• Simulated cascading failures using real-world node-attack and node-failure scenarios.

Main Results:

• Node protection effectively mitigates single-node failures but is less effective against multiple simultaneous failures.
• A critical tolerance level for node protection was identified, leading to a phase transition where networks largely survive attacks.
• Cascade-size distributions near this critical tolerance exhibit power-law decay.
• Strong correlation found between cascade sizes from single and multiple initial failures.
• Network topology alone is a poor predictor of cascading failure progression.

Conclusions:

• Optimal resource distribution and targeted node protection are key to enhancing network resilience.
• The identified critical tolerance and phase transition offer insights into robust network design.
• Network topology is less critical than capacity allocation and protection strategies in managing cascading failures.