Enhancing resilience of interdependent networks by healing

I. INTRODUCTION

• Interdependent networks combine connectivity links within networks with dependency links between networks, so failure of one supplier can disable dependent nodes elsewhere.The paper uses infrastructure, economic, and financial interdependencies as motivating examples.Source: §I · p. 1
• Existing interdependent-network studies show that coupling can reduce robustness, while proposed strengthening strategies include autonomous nodes and controlled interconnection.Broader degree distributions can increase vulnerability, whereas upgrading nodes to operate autonomously is costly.Source: §I · p. 1
• Healing models adaptive reorganization after failures by substituting failed elements or reallocating their roles among functioning participants.Such mechanisms had previously been studied for simple networks, motivating their extension here.Source: §I · p. 1
• The paper extends cascading-failure models by attempting, after each removal, to bypass failed nodes with new connectivity links formed with probability w.This dynamic procedure applies healing throughout the evolving interdependent system.Source: §I · p. 2, Figure 1 · p. 2
• The study introduces failures one by one, applies healing after each event, and analyzes scaling, cascade properties, and numerical results across subsequent sections.The dynamic timing enables healing at every failure event rather than only after an initial batch attack.Source: §I · p. 2

II. THE MODEL

• Two topologically identical networks contain within-network connectivity links and bidirectional one-to-one dependency links, with nodes functioning only inside both networks’ largest connected components.The simulations use square-lattice systems of size N = L × L as an example topology.Source: §II · p. 2
• Failures are introduced one node at a time in random order, with time normalized by the system size N.Each external attack defines one time step before relaxation through cascades.Source: §II · p. 2
• A failed node can fragment its network’s largest connected component; only the largest part survives, and dependency links then propagate failures to the other network.The cascade repeats until no additional failures occur.Source: §II · p. 3
• After each failure propagation, every pair of functioning neighbors of a failed node becomes a candidate healing link, independently accepted with probability w.Healing links are established simultaneously and can bridge increasingly distant parts of the topology.Source: Figure 2 · p. 2, §II · p. 3
• At w = 0, the dynamic process reduces to the step-by-step version of the established Li et al. model, with an implicit relation connecting attack conventions.The two thresholds remain close because random attacks rarely select nodes already disconnected from the giant component.Source: §II · p. 3
• The no-healing threshold is pc = 0.690 ± 0.001, consistent with the previously reported result for the corresponding model.This value is obtained by extrapolating to infinite system size.Source: §II · p. 3

• The critical attack fraction increases monotonically with healing probability w.Source: §III · p. 3
• The scaling analysis uses square lattices with sizes L = 20, 40, 80, 160, and 320 under periodic and open boundary conditions.Source: §III · p. 3
• The rescaled P∞(p,w) curves collapse asymptotically in the w →0 limit.Source: §III · p. 3, Figure 3 · p. 4
• The transition changes from discontinuous to smooth above wc = 0.351 ± 0.002, where the giant component size immediately before breakdown becomes zero.Source: Figure 4 · p. 4
• Finite-size scaling gives νp = 1.10 ± 0.06 for periodic boundaries and νo = 1.20 ± 0.06 for open boundaries, implying ν ≈ 1.15.Source: §III · p. 4, Figure 5 · p. 4
• For the infinite-size network, systems with periodic and open boundary conditions collapse well, yielding h = 0.703 ± 0.005 and γ = 1.034 ± 0.009.These parameters are obtained by fitting each system size and extrapolating with 1/L.Source: §III · p. 4

IV. CASCADES CHANGE TOPOLOGY

• For w > wc, no macroscopic cascade occurs and P∞(p) decreases smoothly to zero through a second-order transition.The simulation gives wc = 0.351 ± 0.002.Source: §IV · p. 4
• Below wc, the network undergoes a sharp breakdown and the average degree decreases monotonically as the fraction of surviving nodes falls.The healing threshold is defined as the lowest w for which P∞(p) lacks an infinite slope.Source: §IV · pp. 4–5
• At wc = 0.348 ± 0.003, the average degree remains approximately constant during the process for the N = 320 × 320 system.Figure 6 reports the average degree as a function of the fraction of dead nodes and identifies this near-conservation point.Source: Figure 6 · p. 5, §IV · p. 5
• Healing adds new links on average, with a rough mean-field balance condition 2w = k and w_mean-field = 1/3 for square lattices with k = 4.The measured near-constant-degree threshold agrees with this estimate.Source: §IV · p. 5
• The critical healing probability preserves the average degree near its initial value while changing the network topology.Below the threshold, connectivity links remain local and the average degree declines as nodes fail.Source: §IV · p. 5

V. CONCLUSIONS

• Healing by new random links delays network collapse by hindering cascades and can completely suppress macroscopic cascading failures above a critical probability.At the critical probability, the average degree remains close to its initial value even though the topology changes.Source: §V · p. 5
• The model uses a starting lattice, unbounded dependency range, high dependency-link density, and non-local healing links, limiting direct realism.The authors identify these features as unrealistic and call for further study of their effects.Source: §V · p. 5