Adaptive Percolation Reveals Topology-Independent Phase Transition in Network Dismantling

The study introduces an adaptive biased percolation algorithm that iteratively removes nodes to optimize dismantling while tracking connectivity metrics in the thermodynamic limit. Across synthetic models with heterogeneous and homogeneous degree sequences, the process produces a first-order transition marked by simultaneous collapse of the giant component and 2-core. Real-network simulations on infrastructure and biological graphs reproduce the same critical behavior, suggesting the transition is insensitive to broad topological variation.

This universality challenges prior algorithm-focused work that treated dismantling as a finite-size optimization problem without examining critical scaling. The finding implies that topology-agnostic theories of network resilience may suffice for predicting failure thresholds, with direct relevance to power-grid vulnerability and neural-network robustness under targeted attack.

Next steps include analytic derivation of the critical exponents and extension to temporal and multiplex networks. Validation on larger empirical datasets will test whether the reported universality holds when community structure or spatial embedding is dominant.