Identifying Vulnerable Nodes to Cascading Failures: Optimization-Based Approach

A key challenge to ensuring robustness of complex systems is to correctly identify component systems, which we simply call nodes, that are more likely to trigger cascading failures. A recent approach takes advantage of the relationship between the cascading failure probability and the non-backtracking centrality of nodes when the Perron-Frobenius (P-F) eigenvalue of the associated non-backtracking matrix is close to one. However, this assumption is not guaranteed to hold in practice. Motivated by this observation, we propose a new approach that does not require the P-F eigenvalue to be close to one, and demonstrate that it offers good accuracy and outperforms the non-backtracking centrality-based approach for both synthetic and real networks.