Analyzing Network Robustness under Adaptive Farthest-First Attack Strategies

Robustness in complex networks reflects the capacity of a system to maintain connectivity and essential functionality under random failures or deliberate attacks. Accurately assessing how structural properties influence resilience is critical for designing stable infrastructures and preventing cascading disruptions. Traditional robustness analyses often rely on random or degree-based node removals, but these approaches remain insufficient at capturing the spatial dispersion and topological diversity that shape real-world network vulnerability. This study investigates network robustness under non-adaptive, semi-adaptive, and fully adaptive node removal strategies using two recently proposed dispersion-oriented measures: Farthest-First (FF) traversal and Farthest-First intrapartition Degree Centrality (FFiDC). These methods prioritize nodes based on structural separation and intra-community significance, offering a complementary perspective to conventional centrality-driven dismantling. We evaluate their effectiveness in fragmenting network connectivity across both synthetic and real-world networks using the Largest Connected Component (LCC) as the primary robustness metric. Results show that the FFiDC approach surpasses or performs at-par with traditional degree-based strategies in reducing the LCC, particularly in adaptive scenarios. The semi-adaptive framework provides a favorable balance between computational efficiency and dismantling performance, preserving much of the adaptive strategy’s strength with significantly reduced overhead. The results show that the proposed FFiDC approach under semi-adaptive node removal strategy consistently reduces the execution time across all network models and configurations. The runtime reduction ranges from 5.1% to 27.5% for Barabási–Albert (BA) networks, 4.0% to 24.8% for Erdös–Rényi (ER) networks, and 10.5% to 23.9% for Watts–Strogatz (WS) networks.