Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity
summary
The gist
Two-stage contagion describes processes in which an individual must be exposed to an adopter twice before adopting…
In short
The study investigates oscillations in contagion models where individuals need two exposures to adopt. It finds that sustained cycles require two-stage contagion, slow demographic turnover, and dense networks. Finite populations exhibit noise-driven quasi-cycles above a critical threshold, while network structure (connectivity) dampens these oscillations below a certain mean degree.
Key concepts
- Two-stage Contagion
- This describes a process where an individual must be exposed to an adopter twice before they adopt the behavior. This mechanism is crucial because it allows for the emergence of oscillations and bistability, which are not possible in simpler one-stage contagion models.
- Mean-Field Dynamics
- This is a simplified mathematical description of the system where individuals are treated as a well-mixed population. The study shows that mean-field theory predicts oscillations only when two-stage contagion and slow turnover conditions are met, but it fails to capture the full effect of network structure.
- Finite Population Noise
- In real populations, random demographic changes (noise) can either destroy a deterministic cycle or sustain oscillations. The linear noise approximation predicts that above a certain threshold, fluctuations become quasi-cycles whose size depends on the population size N.
Terminology used across episodes
This episode discusses
- Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity · Paper Radio
The paper
Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity · Read on arXiv
Francisco J. Muñoza, Juan Carlos Nuñoa, Luca Meaccib, Mario Primicerio
Departamento de Matematica Aplicada, Universidad Politécnica de Madrid · Dipartimento di Matematica e Informatica “Ulisse Dini”, UniversitaÓ degli Studi di Firenze · Dipartimento di Ingegneria dell’Informazione (DINFO), UniversitaÓ degli Studi di Firenze
Transcript
Introduction to the show: ident: Nonlinear Sciences. The day in nonlinear science: chaos, patterns, networks and adaptation.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We're looking at the paper: "Oscillations in two-stage contagion on networks".
Mira: Here's the gist of the paper. Two-stage contagion describes processes in which an individual must be exposed to an adopter twice before adopting, and this study investigates oscillations in such a model with demographic turnover on networks Model Definition The model describes a network of nodes in five states: naive S0, informed S1, active (adopter) A, refractory (former adopter) R, and inactive I The rules governing contagion involve transitions based on the fraction of active neighbors φi and a small rate of spontaneous activation σ Turnover is modeled by choosing nodes to become inactive or naive at a rate ν Mean-Field Dynamics In the mean-field description, oscillations are born at a supercritical Hopf bifurcation requiring predominantly two-stage contagion (Q ≲ 0.17) and slow turnover Spontaneous activation acts as a perturbation that destroys these oscillations The oscillatory region is bounded by conditions such as requiring two-stage contagion, slow turnover, and intermediate recovery rate Finite Population Noise In finite populations demographic noise can either destroy a deterministic cycle by extinction or sustain oscillations where the deterministic dynamics is damped Linear noise approximation predicts quasi-cycles above the Hopf bifurcation with a spectrum and variance predicted by a linear noise approximation away from the Hopf bifurcation The variance of A grows as N (the relative amplitude decreases as N −1/2) Network Structure Effects Reducing the connectivity of the network damps the oscillation because naive nodes see fewer active neighbours than in a well-mixed population A pair approximation predicts that the sustained cycle disappears below a critical mean degree ¯kc, which decreases as Q and the turnover rate decrease Sparsity erases the signature of the two-stage mechanism at low connectivity Sustained, large oscillations need the combination of a predominantly two-stage contagion, a slow turnover and a dense network Reducing any of these three ingredients turns them into small noise-driven fluctuations Recurrent waves of adoption and abandonment are features of novelties that need repeated exposure (small Q), in slowly renewed populations (small ν), with little exogenous influence and on densely connected contact networks The irregular time series of the simulations are quasi-cycles generated by demographic noise acting on a regular deterministic skeleton, not deterministic chaos Bistability and Chaos Absence For σ = 0 the extinction state is absorbing, and the mean-field equations have a bistable window νT < ν < νF bounded by a transcritical bifurcation and a fold of the endemic branch A numerical search of the leading Lyapunov exponent found no exponent above the finite-time error of the estimate, so no chaotic regime was found The irregular time series are quasi-cycles generated by noise acting on a regular deterministic skeleton, not deterministic chaos Discussion and Implications Recurrent waves of adoption and abandonment are a feature of novelties that need repeated exposure (small Q), in slowly renewed populations (small ν), with little exogenous influence and on densely connected contact networks A modest exogenous influence turns sustained waves into irregular, noise-driven fluctuations of a similar period Quasi-cycles sustained by demographic noise are well known in epidemic and predator–prey models
6, 7: The pair approximations and edge-based descriptions of network contagion have a long history The role of the topology is captured by the pair approximation which captures the mean level, the damping of the oscillation and the neighbour correlations over two decades of ¯k Limitations The pair approximation treats the network as regular and does not include noise; a linear noise approximation of the pair equations, and a heterogeneous pair approximation for BA networks, would allow a quantitative theory of the quasi-cycles on sparse and heterogeneous networks The explanation offered for the effect of hubs is supported only indirectly by the degree dependence of the correlations The synchronous discrete-time update is slightly less damped than its continuous-time limit Acknowledgements This is Mario Primicerio’s final article.
Kai: It completes the series on twostage contagion that he initiated with the Madrid group Luca Meacci acknowledges the support of the Gruppo Nazionale per la Fisica Matematica (GNFM) of the Istituto Nazionale di Alta Matematica (INdAM), Italy, and a fellowship funded by the Italian Ministry and awarded through DINFO, Università degli Studi di Firenze The authors used Claude (Anthropic) to assist in writing, preparing figures and editing the text Data Availability The code used to produce the results is provided as supplementary files The simulation data are available from the corresponding author on reasonable request The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work Declaration of Generative AI and AI-assisted Technologies in the Manuscript Preparation Process During the preparation of this work the authors used Claude (Anthropic) to assist in writing, preparing figures and editing the text After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article Appendix A.
Mira: Noise Covariance of the Linear Noise Approximation For the complete graph, per-step transitions contribute re vev T e to D The turnover removes exactly m = νwN nodes per step from the non-inactive population, chosen without replacement, and returns m naive nodes Appendix B.
Kai: Pair-Approximation Equations The expected dynamics is described by equations (B.1) and (B.2) A transition caused by contact is more likely for nodes with more active neighbours; conditioning on it, the node has on average 1 + (¯k − 1)φX active neighbours and (¯k − 1)qY Xe neighbours in any other state Y Appendix C.
Title and authors: Kai: Now let's get into the specific title and authors of this paper, "Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity." The title is very descriptive—it immediately tells you the three main parameters they are investigating.
Mira: It’s a very clear summary. It sets up the expectation that we're going to look at how the contagion structure, like requiring two exposures, interacts with turnover, like demographic renewal, and network connectivity all influence those oscillations.
Lev: From an engineering standpoint, I want to understand what this means for hardware—like how we build a system where these parameters can be really tuned and measure the fidelity of those components.
Kai: Right, so when we talk about tuning the turnover rate or Q less than or equal to zero point seventeen we're talking about setting specific constraints on the underlying process.
Mira: And this means that if you want to achieve sustained waves in a two-stage contagion model, you can't just pick any initial probability; it has to be within that specific range for the mean-field description to show those oscillations.
Lev: If we think about real hardware, setting constraints like Q less than or equal to zero point seventeen means you are designing a system where the first contact probability is constrained by the need for sustained dynamics.
Kai: Exactly. It’s not just about getting an arbitrary number; it's about hitting that specific bifurcation point where those oscillations are born in the first place.
Mira: And they confirm this threshold of Q less than or equal to zero point seventeen is confirmed by simulations on three point two times ten five nodes.
Lev: Three point two times ten five nodes is a sample size, so it gives us a concrete scale for how large the system needs to be before those mean-field results become reliable.
Kai: So the paper says that threshold of Q less than or equal to zero point seventeen is confirmed by simulations on three point two times ten five nodes.
Mira: And they also mention that at low connectivity, the amplitude no longer depends on Q, which is a result that suggests sparsity erases the signature of the two-stage mechanism because if you only look at Q less than or equal to zero point seventeen.
Lev: So sparsity changes the dynamics so much that it can make those initial conditions irrelevant for sustained waves.
Kai: That’s a strong statement about how network structure can override the intrinsic rules of contagion when you reduce connectivity.
Mira: It means if you are designing a system, topology becomes just as important as the interaction rules in determining whether you get those sustained oscillations or not.
Lev: So, it's not just about tuning Q; it’s about designing a system where the structure itself is conducive to the desired outcome.
Kai: The next thing we need to look at is how these findings are summarized in a way that makes sense for someone who isn't deep into complex contagion theory.
Mira: They do that by focusing on the summary, which boils down to needing two-stage contagion, slow turnover, and a dense network structure for sustained oscillations.
Lev: So if you want to keep it simple for the listener, that’s the core takeaway of this paper: density and slow renewal are key ingredients.
Kai: And they also mention that modest exogenous influence turns those sustained waves into irregular fluctuations of a similar period, driven by demographic noise acting on a regular deterministic skeleton.
Mira: So the paper is telling us that external influences don't necessarily create new phenomena, they just modulate what's already there.
Lev: That’s a pragmatic view—the system is fundamentally predictable, and external stuff just adds noise to that predictability.
The paper's summary: Kai: So we’ve established the basic mechanics of two-stage contagion and the role of turnover, and now we need to look at what the paper actually summarizes about these mechanisms.
Mira: They summarize it by saying that sustained oscillations happen when you combine two-stage contagion with slow turnover and a dense network structure.
Lev: That's the core recipe for getting those waves of adoption and abandonment in this model.
Kai: And they also point out that the irregular time series observed in simulations are not actually chaos, but quasi-cycles generated by demographic noise acting on a regular deterministic skeleton.
Mira: So, so when we talk about those irregular time series, it’s important to stress that they are quasi-cycles and not chaotic dynamics.
Lev: That distinction is important for anyone trying to model this on real hardware because if you think it's chaos, you need entirely different error correction strategies than if it's just noise acting on a regular skeleton.
Kai: Because the deterministic part is still there, which means there’s some underlying rhythm that isn't completely lost.
Mira: So they are essentially saying that those recurrent waves of adoption and abandonment are features of novelties that need repeated exposure, in slowly renewed populations, with little exogenous influence and on densely connected contact networks.
Lev: That’s the big picture for anyone listening—it's about finding the right conditions to get those waves to happen naturally.
Kai: And they also found that the pair approximation gives us a good prediction of the oscillation amplitude within about ten percent, which is a result we can use practically.
Mira: So, so that means if you're interested in using this for practical applications, you get a practical measure of how big those cycles are.
Lev: That gives engineers something concrete to test against when designing their systems.
The paper's improvements: Kai: Now let's talk about what the authors suggest to improve this work, because they aren't just stopping at the mean-field results.
Mira: They suggest incorporating network connectivity dependence directly into the mean-field equations instead of leaving it as a separate parameter in section three.
Lev: That would mean extending that independence so it doesn't just be about some fixed average value, but really accounting for how the structure itself influences the dynamics at every step.
Kai: And that would require a more detailed look at how the pair approximation works when you move from well-mixed assumptions to structured ones.
Mira: They also suggest developing a pair approximation module capable of calculating neighbor correlations in sparse networks to predict how damping operates by comparing those measured ratios, like phi sub zero over alpha with values from well-mixed populations.
Lev: That’s a tangible goal for error correction research—getting that kind of quantitative tool for handling real network structure is what we need to run on actual hardware.
Kai: And they also mention that a finite-size test framework using the ratio of variance of four N over the variance of N over N, which can distinguish between noise-driven quasi-cycles and sustained cycles based on a critical mean degree kappa sub c.
Mira: That gives us a way to mathematically separate those two regimes based on how the system scales with population size.
Lev: If we could build something that uses that framework, it would be very valuable for figuring out if you're looking at noise-driven quasi-cycles or a sustained cycle.
Kai: And finally, they suggest a topological analysis tool that uses neighbour correlation measurements from simulations to see how degree heterogeneity amplifies the oscillation versus how clustering damps it.
Mira: That would provide a direct picture of the interplay between local structure and the global dynamics, which is something that’s hard to get just by looking at averages.
Lev: Having a tool that does that would help us understand where we need to go for better error correction overhead analysis in complex systems.
Conclusion: Kai: So wrapping up on "Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity." The main thing is that sustained oscillations require two-stage contagion, slow turnover, and a dense network structure.
Mira: And they confirm that modest exogenous influence turns those sustained waves into irregular fluctuations of a similar period, driven by demographic noise acting on a regular deterministic skeleton.
Lev: From an engineering view, the final thought is that while the pair approximation gives us good approximations for mean degree greater than or equal to ten we'd need more sophisticated tools if we want to predict behavior on truly sparse and heterogeneous networks accurately <ref:2611.12920#pg3>.
Lev: For me, the final thought is that while the pair approximation gives us good approximations for mean degree greater than or equal to ten we'd need more sophisticated tools if we want to predict behavior on truly sparse and heterogeneous networks accurately <ref:2611.12920#pg3>.
More episodes
- 2608.20276-Flip rate prediction in the double pendulum
- The day in conversation for 2026-11-18
- Daily Summary for 2026-11-18
- 2603.16051-Gradual emergence of temporal structures depending on the distance between neighboring callers in natural habitat of male treefrogs
- 2603.21801-Diffraction of deep-water solitons
- The day in conversation for 2026-11-17
- Daily Summary for 2026-11-17
- 2204.04335-Measuring and improving community resilience: a Fuzzy Logic approach
- 2604.11624-Prediction of chaotic dynamics from data: An introduction
- 2610.21575-Cooperation Dynamics with Environmental Feedback: Steady States and Transitions