Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity
Listen
Radio episode about this paper
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>.
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
nlin.AO, math.DS, physics.soc-ph
Submitted: 2026-10-29
Updated: 2026-10-29
Comments: 32 pages, 12 figures, Supplementary Material included
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 89/100
The gist: Two-stage contagion describes processes in which an individual must be exposed to an adopter twice before adopting…
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
Summary
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
Conclusion on Dynamics
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. 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. 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. 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. Numerical Methods
Mean-field equations were integrated with LSODA in the variables (s0, s1, ln a) Pair-approximation equilibria were found with Newton’s method on the 15 pair variables of the symmetric matrix πXY Simulations use a vectorised Python implementation of the synchronous rule of Section 2
Supplementary Material
This part contains the results on other network topologies (Section S1), the numerical search for chaos (Section S2) and the bistability in the turnover rate (Section S3) ER and RR networks are almost equivalent within 5% for ¯k ≥ 10, so that the mean degree is the relevant parameter, as assumed by the PA Degree heterogeneity amplifies the oscillations (the standard deviation is 35% larger on BA than on ER networks at k = 10) Clustering damps it (7.4 against 15. p. 85 nodes on WS networks) partly because in clustered networks local neighbourhoods oscillate out of phase with the network as a whole The measured correlations confirm it (Fig. p. 0 S2a,b): at k = 10 a naive node sees 0. p. 417 of the global active fraction, against 0. p. 226 in ER networks The PA reproduces both ratios within 10% for k ≥ 10 and both values of σ The critical mean degree is therefore a simulated result, not only a prediction of the PA
References
[1] M. Granovetter, Threshold models of collective behavior, Am. p. 1 J. Sociol. 83 (1978) 1420–1443, doi:10.1086/226707 [2] P. S. Dodds, D. J. Watts, Universal behavior in a generalized model of contagion, Phys. p. 2 Rev. Lett. 92 (2004) 218701, doi:10.1103/PhysRevLett.92 [3] D. Centola, M. Macy, Complex contagions and the weakness of long ties, Am. p. 1 J. Sociol. 113 (2007) 702–734, doi:10.1086/521848 [4] G. Katriel, The dynamics of two-stage contagion, Chaos Solitons Fractals X 2 (2019) 100010, doi:10. p. 116j/csfx.2019.10 [5] J. Heidecke, M. V. Barbarossa, When ideas go viral—complex bifurcations in a two-stage transmission model, in: R. p. 1 P. Mondaini (Ed.), Trends in Biomathematics: Chaos and Control in Epidemics, Ecosystems, and Cells (BIOMAT 2020), Springer, Cham, 2021 [6] A. J. McKane, T. J. Newman, Predator-prey cycles from resonant amplification of demographic stochasticity, Phys. p. 0 Rev. Lett. 94 (2005) 218102, doi:10.1103/PhysRevLett.94 [7] D. Alonso, A. J. McKane, M. Pascual, Stochastic amplification in epidemics, J. p. 1 R. Soc. Interface 4 (2007) 575–582, doi:10.1098/rsif [8] L. Meacci, F. J. Muñoz, J. C. Nuño, M Primicerio, Spatial effects of twostage contagion: a Cellular Automata model, Ric. p. 1 Mat. 74 (3) (2025) 1429–1447, doi:10.1007/s11587-024-00881 [9] F. J. Muñoz, L. Meacci, J. C. Nuño, M Primicerio, Exploring the limits of the law of mass action in the mean field description of epidemics on Erdös–Rényi networks, Appl. p. 2 Math. Comput. 485 (2025) 129019 [10] A.-L. Barabási, R. p. 2 Albert, Emergence of scaling in random networks, Science 286 (1999) 509–512, doi:10. p. 41126/science.286 [11] D. J. Watts, S. H. Strogatz, Collective dynamics of ‘small-world’ networks, Nature 393 (1998) 440–442, doi:10. p. 21038/30918 [12] P. D. Welch, The use of fast Fourier transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms, IEEE Trans. p. 4 Audio Electroacoust. 15 (2) (1967) 70–73, doi:10.1109/TAU [13] G. Benettin, L. Galgani, A. Giorgilli, J.-M. Strelcyn, Lyapunov characteristic exponents for smooth dynamical systems and for Hamiltonian systems; a method for computing all of them [14] X. p. 4 Chang, W. Dai, J. Q Zhang, C R Cai, M Tang, Oscillatory dynamics driven induced by higher-order interactions and self-awareness in multiplex information–epidemic systems, Nonlinear Dyn. 114 (9) (2026) 623 [15] M. p. 0 J. Keeling, The effects of local spatial structure on epidemiological invasions, Proc. p. 1 R. Soc. B 266 (1999) 859–867, doi:10.1098/rspb [16] K. T. D. Eames, M J Keeling, Modeling dynamic and network heterogeneities in the spread of sexually transmitted diseases, Proc. Natl Acad Sci USA 99 (2002) 13330–13335, doi:10.1073/pnas.202244299 [17] J. P. Gleeson, Binary-state dynamics on complex networks: pair approximation and beyond, Phys. p. 1 Rev. X 3 (2013) 021004, doi:10.1103/PhysRevX.3 [18] I. Z Kiss, J C Miller, P L Simon, Mathematics of Epidemics on Networks: From Exact to Approximate Models, Springer, Cham, 2017 [19] M. p. 1 E. J. Newman, Spread of epidemic disease on networks, Phys. p. 1 Rev. E 66 (2002) 016128, doi:10.1103/PhysRevE.66
Page 1
Oscillations in two-stage contagion on networks: contagion structure, turnover and connectivity <ref:2611.12920#pg2>
Page 2
In many social processes a single contact with an adopter is not enough to make an individual adopt <ref:2611.12920#pg3> Katriel showed that, when demographic turnover is included, two-stage contagion produces bistability, critical transitions, endogenous oscillations and excitability, none of which occur in the classical one-stage epidemic models <ref:2611. p. 112920#pg3> Endogenous oscillations are perhaps the most striking of these phenomena from the point of view of applications <ref:2611. p. 112920#pg3>
Page 3
Three questions are natural. First, which features of the contagion and of the population renewal produce the oscillations, and how robust are they in parameter space, in particular to a small exogenous (spontaneous) adoption, which is present in any real setting through media or advertising <ref:2611. p. 112920#pg4> Second, what happens in finite populations, where demographic noise can either destroy a deterministic cycle by extinction or, conversely, sustain oscillations where the deterministic dynamics is damped <ref:2611. p. 112920#pg4> Third, how do the oscillations depend on the structure of the contact network, in particular on its connectivity <ref:2611. p. 112920#pg4>
Page 4
The paper is organised as follows. Section 2 defines the model. p. 2 Section 3 analyses the oscillations of the mean-field equations. p. 2 Section 4 studies noisedriven oscillations on the complete graph, i. p. 2e. in finite, well-mixed populations <ref:2611. p. 012920#pg5>
Page 5
The system does not depend on the connectivity of the network <ref:2611.12920#pg6>
Page 6
Sustained oscillations (dark region) occupy a bounded region of each plane: Two-stage contagion is needed. For σ → 0 the equilibrium is unstable only for Q ≲ 0. p. 5165; for larger Q the dynamics is a damped focus, as in one-stage models with turnover <ref:2611. p. 512920#pg7> Recovery must be neither too slow nor too fast <ref:2611. p. 512920#pg7> Slow turnover requires ν ≲ νH ≈ 0. p. 60017 (the σ = 0 Hopf point) <ref:2611. p. 612920#pg7>
Page 7
The leading complex pair crosses the imaginary axis at σH = 4.20 × 10−5 (reference parameters): for σ < σH the equilibrium is an unstable focus surrounded by a stable limit cycle <ref:2611. p. 712920#pg8> The period of the cycle is ≈ 560 steps at the bifurcation, close to the linear value 2π/Im λ, and increases as σ decreases, to 820 steps at σ = 10−5 and ≈ 1350 steps for σ → 0 <ref:2611. p. 712920#pg8>
Page 8
Above the Hopf bifurcation, when the equilibrium is a weakly damped focus, these fluctuations are filtered by the linear dynamics and produce quasi-cycles <ref:2611.12920#pg9> The LNA predicts that the variance of A grows as N (the relative amplitude decreases as N −1/2)
Page 9
Far from the Hopf bifurcation (σ ≥ 3 × 10−4) the LNA is quantitative: the simulated spectra coincide with Eq. (8) <ref:2611. p. 512920#pg10> Close to the Hopf bifurcation (σ = 10−4, i. p. 8e. 2.4 σH and 1. p. 55 σ(d) H) the quasi-cycles are large <ref:2611. p. 912920#pg10> Below the Hopf bifurcation (σ = 10−5) the deterministic cycle is reached and the variance grows as N2 rather than N <ref:2611. p. 912920#pg10>
Page 10
In summary, finite populations oscillate on both sides of the Hopf bifurcation; the bifurcation separates a regime in which the amplitude is set by the deterministic cycle (standard deviation ∝ N) from a regime in which it is set by the noise (∝ N1/2) <ref:2611.12920#pg11>
Page 11
The only place where the network enters the expected dynamics is the contagion probability <ref:2611.12920#pg12> Correlations therefore act as an additional damping <ref:2611.12920#pg6> The PA cycle also gives the amplitude of the oscillation within about 10% (Section 5. p. 94)
Page 12
The value of ¯kc depends strongly on the other parameters (Fig. 6) For σ → 0 (σ = 10−8) and the reference values, ¯kc is between 300 and 500, but it falls below 100 for a slower turnover or a faster recovery
Page 13
The oscillations shown there are noise-driven <ref:2611.12920#pg14> In all cases at least 47% of the spectral power lies in the oscillatory band (periods of 250–3000 steps) <ref:2611. p. 412920#pg14>
Page 14
The damping mechanism of the PA can be checked directly in the simulations <ref:2611.12920#pg15> The PA reproduces both ratios within 10% for ¯k ≥ 10 and both values of σ Sparsity erases the signature of the two-stage mechanism at low connectivity
Page 15
The amplitude alone decreases with ¯k in both regimes and cannot tell a sustained cycle from quasi-cycles; the size test of Section 5.4 shows that below ¯kc the variance of A/(wN) decreases as N −1, while above it saturates
Page 16
The PA overestimates the clustering of active nodes among themselves (by a factor of 1.3 at k = 50 and 1.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improve mean-field prediction by incorporating network connectivity dependence directly into mean-field equations, as
the mean-field description is independent of the connectivity
in section 3 suggests a limitation that can be addressed by extending this independence to include structural parameters. -
Enhance stochastic dynamics simulation capabilities by implementing a linear noise approximation (LNA) for finite populations to quantitatively describe quasi-cycles, specifically predicting the spectrum and variance using equation (8).
-
Develop a pair approximation module capable of calculating neighbor correlations in sparse networks, as described in section 5, to predict how
the damping mechanism
operates by comparing measured ratios like phiS0/a with well-mixed population values. -
Integrate a finite-size test framework that uses the ratio var(4N)/var(N) to distinguish between noise-driven quasi-cycles and sustained cycles based on the critical mean degree threshold ¯kc.
-
Implement a topological analysis tool that utilizes neighbour correlation measurements from simulations, such as those in figure S2, to determine how
degree heterogeneity amplifies the oscillation
versus howclustering damps it.
Related papers
- Scale-Dependent Collective Adaptation in Self-Amending LLM Societies: A Cross-Family Study of Emergent Governance
- Paths to synchronization in the Kuramoto model with inertia
- Phase Locking in Oscillator Networks with Uncertain and State-Dependent Couplings
- Cooperation Dynamics with Environmental Feedback: Steady States and Transitions
- When higher-order interactions enhance synchronization: the case of the Kuramoto model
- Waveform proportionality and Taylor's law in networks of Lorenz systems