Dynamics of critical cascades in interdependent networks

Dolev Dilmoney, Bnaya Gross, Shlomo Havlin, Nadav M. Shnerb · 2025

Read Dynamics of critical cascades in interdependent networks by Dolev Dilmoney, Bnaya Gross, Shlomo Havlin, Nadav M. Shnerb (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-06862.txt.

Introduction

Dolev Dilmoney and co-authors study cascading failures in linked networks in a 2025 paper. When networks depend on each other, a failure can trigger others in an avalanche. Near a critical point each failure sparks about one more on average, but as losses mount the network weakens and each failure sparks more, which can end in sudden total collapse.

The authors map this process onto birth-death models from population math, which yields new formulas and much faster calculations. They use the frame to work out the chance of collapse, how long avalanches last, and how long the flat phase before collapse can run. They also show how system size and the size of the first shock shape each of those outcomes.

Catalog record

CREATORDolev Dilmoney, Bnaya Gross, Shlomo Havlin, Nadav M. Shnerb
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2504-06862

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Dynamics of critical cascades in interdependent networks Dolev Dilmoney, Bnaya Gross, Shlomo Havlin, Nadav M. Shnerb Abstract The collapse of interdependent networks, as well as similar avalanche phenomena, is driven by cascading failures. At the critical point, the cascade begins as a critical branching process, where each failing node (element) triggers, on average, the failure of one other node. As nodes continue to fail, the network becomes increasingly fragile and the branching factor grows. If the failure process does not reach extinction during its critical phase, the network undergoes an abrupt collapse. Here, we implement the analogy between this dynamic and birth-death processes to derive new analytical results and significantly optimize numerical calculations. Using this approach, we analyze three key aspects of the dynamics: the probability of collapse, the duration of avalanches, and the length of the cascading plateau phase preceding a collapse. This analysis quantifies how system size and the intensity of the initial triggering event influence these characteristics. Network science has experienced remarkable development in recent decades, finding extensive applications across various fields, including the study of neural, social, metabolic, and communication networks Barabási 2009 ; Mata 2020 . Within this domain, the topic of interdependent networks has gained significant attention in recent years Buldyrev et al. 2010 . This concept has also found a wide range of applications, from critical infrastructures and cybersecurity to the dynamics of ecological communities Pocock et al. 2012 ; Ouyang 2014 ; Rocha et al. 2018 . Recent work has effectively demonstrated the key features of interdependent network collapse using a clear and physically straightforward experimental model, which involves thermally connected layers of amorphous superconductors Bonamassa et al. 2023 . Interdependent networks are characterized by two types of links: connectivity links within each network and dependency links between nodes of different networks. Each link type serves a distinct role. Connectivity links are understood as in standard percolation problems, so nodes that disconnect from the giant connected component (GCC) of their network are considered non-functional. Dependency links, on the other hand, propagate damage between the networks: if a node in one network fails, its dependent node in the other network will fail as well, even if it is still connected to the GCC of its own network (see Figure 1 (a) ). When nodes are removed from one network, the interplay between these two types of links can trigger a cascading failure. Therefore, while a single network experiences a continuous second-order transition in which the GCC collapses when 1-p_{c} of the nodes are removed, interdependent networks undergo an abrupt, first order transition, at {\tilde{p}}>p_{c} (Figure 1( b )) Buldyrev et al. 2010 . These cascading failures are a general characteristic of avalanche phenomena Lee et al. 2016 , which have also been extensively studied in recent decades in various contexts such as magnetic materials, plastic distortions, crack propagation, earthquakes, neural networks and so on Perković et al. 1995 ; Sethna et al. 2001 ; Haldeman and Beggs 2005 ; Friedman et al. 2012 ; Gross et al. 2024 . In all these systems, a failure or change in the state of a single component, such as a spin, imposes ’pressure’ on adjacent components, and a critical point in parameter space separates localized avalanches from widespread collapses. Such cascades are mathematically analogous to stochastic birth-death processes, where individuals can produce a random number of offspring Harris et al. 1963 ; Lomnitz-Adler et al. 1992 ; Haldeman and Beggs 2005 ; Shekhawat et al. 2013 ; Zhou et al. 2014 . In this Letter, we aim to examine the analogy between demographic processes and the dynamics of cascading failures. In particular, we will focus on the critical point, where the corresponding dynamics begins as a neutral birth-death process Kimura 1985 ; Hubbell 2011 ; Azaele et al. 2015 ; Azaele et al. 2016 but, over time, diverges from neutrality. We will implement this analogy to derive novel qualitative and quantitative insights. Specifically, our study addresses three crucial questions: First, how does the likelihood of a collapse change in relation to the scale of the initial attack? Second, the impact of this initial attack size on the ’plateau time’ — the period of slow dynamics preceding the collapse (Figure 1 (c) ). Third, we examine how the initial attack size impacts the mean ’avalanche time,’ the duration until an avalanche ends, whether or not it causes collapse. Figure 1: Cascading failures in interdependent networks conceptualized as a birth-death process. Panel (a) depicts two networks, A and B, interconnected by dependency links (green) alongside internal connectivity links (black). When a node fails, it topples down (in the same network) all the nodes it disconnects from GCC. Additionally, each failed node also causes the dependent node in the parallel network to fail, leading to a cascading effect. For instance, the failure of node 1A (marked with red X) results in the detachment of 2A from the GCC, causing the failure of nodes 1B and 2B due to dependency. Consequently, 3B becomes detached from the GCC of network B, which cascades back to topple 3A. Thus the failure of 1A leads to the failure of 3A in the subsequent iteration. Panel (b) : while single networks exhibit a continuous phase transition (red) in which the likelihood of a node being part of the GCC ( P_{\infty} ) continuously diminishes to zero, interdependent networks undergo an abrupt transition (blue) due to cascading failures. At p={\tilde{p}} (Panel (c) ), the cascade of failures involves a long-lasting plateau, where damage stays relatively constant over several iterations. Panel (d) : The cascading failure dynamics mirrors a birth-death process, wherein each failed node [like 1A in panel (a)] has a given number of offspring (like 3A) in the subsequent iteration. When p>{\tilde{p}} the mean offspring count, {\overline{m}} , is less than one, resulting in an exponential decay of the demographic process and network survival. Conversely, for p<{\tilde{p}} , {\overline{m}}>1 leading to exponential damage growth and a swift collapse. At p={\tilde{p}} , {\overline{m}}=1 during the plateau, and the dynamics is neutral From Nodes to Generations:– As depicted in Figure 1( a ), damage cascades reciprocally between dependent networks, with each failed node having a chance P_{1\to m} of triggering the failure of m nodes within the same network in the subsequent iteration. With a sufficiently large number of nodes, N , and a small number of failed nodes, failures are uncorrelated. Consequently, at the t -th step, if the count of failed nodes is n_{t} , then n_{t+1} —the tally of failed nodes in the subsequent step in the same network—is the sum of n_{t} independent variables, following a probability distribution function of P_{1\to m} (see Appendix A ). Hence, this branching process mirrors a demographic dynamics, where nodes represent individuals and iterations are equivalent to generations (Figure 1( d )). At p={\tilde{p}} , the process is neutral , meaning that during the plateau (Figure 1( c )), the mean number of offspring per node is one. The “population dynamics” of n_{t} is influenced by demographic stochasticity, namely the binomial noise associated with the randomness of the birth-death process, and to deterministic forces that yield exponential growth (if \overline{m}>1 ) or exponential decay (if \overline{m}<1

Read the complete text