Correlation-Based Diagnostics of Social Contagion Dynamics in Multiplex Networks
Joan Hernàndez Tey, Emanuele Cozzo · 2026
Read Correlation-Based Diagnostics of Social Contagion Dynamics in Multiplex Networks by Joan Hernàndez Tey, Emanuele Cozzo (2026) 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-2601-22459.txt.
Introduction
Joan Hernàndez Tey and Emanuele Cozzo read the time ties of spread on layered networks to tell its mode. On multiplex networks, a contagion may hug one layer or roam across all of them. Their 2026 preprint derives a closed-form mean-field account of those time ties.
The ties encode which mode the spread sits in. That gives watchers a dial they can read from data alone. The work fits disease, rumor, and fault spread on real layered systems. Early mode reads can guide where to act first. It is a neat link from math to watch-keeping.
Catalog record
| CREATOR | Joan Hernàndez Tey, Emanuele Cozzo |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-22459
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Correlation-Based Diagnostics of Social Contagion Dynamics in Multiplex Networks Joan Hernàndez Tey, Emanuele Cozzo Abstract Multiplex contagion dynamics display localization phenomena in which spreading activity concentrates on a subset of layers, as well as delocalized regimes where layers behave collectively. We investigate how these regimes are encoded in temporal correlations of node activity. By deriving a closed-form mean-field expression for node autocorrelations in a contact-based social contagion multiplex model and validating it through simulations, we show that lag-one autocorrelations act as sensitive indicators of both activation and localization transitions. Our results establish temporal correlations as lightweight, structure-agnostic probes of multiplex spreading dynamics, particularly valuable in partially observable systems. I Introduction In contemporary digital ecosystems, social activity, information diffusion, and collective behavior unfold simultaneously across multiple online platforms. Individuals typically maintain accounts on several services, interacting through distinct but interdependent networks. As a result, many dynamical processes of interest — ranging from opinion formation and attention cascades to coordination and collective mobilization — are intrinsically multiplex in nature. Understanding such processes therefore requires models and analytical tools that explicitly account for cross-platform coupling rather than treating each platform in isolation. A central challenge in this context is partial observability . In real-world settings, access to data is often asymmetric: some platforms provide extensive public data, while others impose strong restrictions or are entirely opaque. Consequently, researchers and practitioners are frequently faced with the problem of inferring the global state of a coupled, multi-platform system using observations from only a subset of its layers. This raises a fundamental question: to what extent can the macroscopic behavior of a multiplex dynamical process be diagnosed from limited, layer-specific observations? Addressing this question is crucial both for empirical analysis and for the design of timely interventions in networked systems where direct monitoring of all components is infeasible. In this work, we approach this inference problem through the lens of contact-based contagion dynamics on multiplex networks . While epidemic-like models have long been used to study spreading processes on single networks, their extension to multiplex settings reveals qualitatively new regimes of behavior. In particular, depending on the strength of interlayer coupling and the structural heterogeneity between layers, contagion may remain localized , with one layer effectively driving the dynamics, or become delocalized , with multiple layers activating simultaneously. Previous studies have shown that this localization–delocalization crossover can be predicted from structural properties of the multiplex network alone. Here, we go one step further by asking whether dynamical correlations , accessible from time series data on individual layers, can serve as reliable indicators of the global regime of the system, even when only a single layer is observed. To this end, we focus on node-level temporal autocorrelations as a minimal yet informative observable. Autocorrelations are directly accessible from time-resolved activity data, require no knowledge of the full network structure, and naturally encode the balance between endogenous spreading and externally driven fluctuations. We introduce a mean-field analytical approximation for the steady-state autocorrelation in a contact-based multiplex contagion model and analyze how its behavior reflects the main regions of the multiplex phase diagram. By combining analytical results with dynamical simulations, we show that autocorrelations provide clear and robust signatures distinguishing inactive, active–localized, and active–delocalized regimes, and we assess the conditions under which the state of an unobserved layer can be inferred from partial observations. The paper is organized as follows: in Sec. II we introduce the model and summarize its phase diagram; in Sec. III we derive the analytical approximation for node autocorrelations; in Sec. IV we validate and interpret these predictions through numerical simulations; and we conclude with a discussion of implications and limitations for inference in partially observable multiplex systems. II The model In this work we use the contact-based social contagion model on multiplex networks introduced in Ref. 1 and further analyzed in Ref. 8 as a minimal framework for modeling propagation of information across multiple, interdependent social platforms. The model incorporates both intra-layer spreading within each platform and interlayer transmission between replicas of the same user across platforms. Since our focus here is on dynamical autocorrelations, we only summarize the main ingredients of the model and its qualitative regimes, referring the reader to Refs. 1 ; 8 for full details and derivations. We consider a multiplex network composed of M layers, each containing the same set of N nodes. Following the formalism introduced in 2 , latin indices u,v\in\{1,\dots,N\} label nodes within a layer, while Greek indices \alpha,\kappa\in\{1,\dots,M\} denote layers. Global indices i,j\in\{1,\dots,MN\} refer to nodes in the full multiplex and are defined through the bijection i=u\alpha . The model interpolates between the contact process (CP), where an infected node makes at most one effective contact per time step, and the reactive process (RP), where it attempts to infect all its neighbors in the layer at each time step. For each layer \alpha , the contact probability matrix R^{\alpha} is defined as in the single layer case introduced in 6 : (R^{\alpha})_{uv}=1-\left(1-\frac{(A^{\alpha})_{uv}}{k_{u}^{\alpha}}\right)^{\gamma_{u}^{\alpha}}, (1) where A^{\alpha} is the adjacency matrix of layer \alpha , k_{u}^{\alpha} is the degree of node u , and \gamma_{u}^{\alpha} controls the interpolation between CP ( \gamma=1 ) and RP ( \gamma\to\infty ) dynamics. The full multiplex dynamics is encoded in the supra-contact probability matrix \bar{R}=\bigoplus_{\alpha=1}^{M}R^{\alpha}+\epsilon\,C, (2) where C is the interlayer coupling matrix that connects each node to its replicas across layers, and \epsilon=\eta/\beta is the ratio between the interlayer infection probability \eta and the intralayer infection probability \beta . Let p_{i}(t) denote the probability that node i is infected at time t . Under a mean-field approximation that neglects joint infection probabilities, the discrete-time evolution equation reads 1 ; 6 p_{i}(t+1)=[1-p_{i}(t)][1-q_{i}(t)]+(1-\mu)p_{i}(t)+\mu[1-q_{i}(t)]p_{i}(t), (3) where \mu is the recovery probability and q_{i}(t)=\prod_{j}\left(1-\beta\bar{R}_{ij}p_{j}(t)\right) (4) is the probability that node i does not receive infection from any neighbor at time t . At stationarity, the absorbing solution p_{i}=0 becomes unstable when \frac{\beta}{\mu}>\frac{1}{\bar{\Lambda}_{\max}}, (5) where \bar{\Lambda}_{\max} is the largest eigenvalue of the supra-contact matrix \bar{R} . The critical point separating the inactive and active phases is therefore given by \left(\frac{\beta}{\mu}\right)_{c}=\frac{1}{\bar{\Lambda}_{\max}}. (6) The interplay between intra- and interlayer couplings gives rise to two qualitatively distinct spreading regimes 8 ; 4 ; 3 : • Localized regime (AL). As illustrated in Fig. 1 , the localized regime corresponds to the region labeled AL in the phase diagram. In this regime the multiplex is endemic because the dominant layer (the one with the largest leading eigenvalue) is supercritical, while the non-dominant layer remains below its activation threshold and is only weakly driven by interlayer transmission. As a result, activity is effectively confined to the dominant layer, and the two layers display markedly different activity levels. • Delocalized regimes (AD 1 and AD 2 ). Delocalized behavior occurs in the regions labeled AD 1 and AD 2 in Fig. 1 , where both layers sustain endemic activity and their dynamical behavior becomes comparable. In region AD 1 , delocalization is induced by increasing the spreading rate \beta/\mu above the activation threshold of the non-dominant layer at weak interlayer coupling. In contrast, region AD 2 corresponds to strong interlayer coupling, where increasing \eta/\beta synchronizes the activation of the layers and the system behaves as a single effective network, independently of