Can behavioral feedback and seasonality together generate chaos in multi-pathogen epidemics?

Two pathogens, one population, three modeling stages

The paper asks whether two familiar forces — seasonal changes in transmission and behavioral feedback that reduces contacts as infections rise — can together push a two-pathogen epidemic into chaotic, non-repeating multi-annual patterns. It answers by building a compartmental model in which each individual can be susceptible, infectious, or recovered with respect to each of two diseases, giving nine compartments. Immunity wanes, so recovered people return to susceptibility, and the two pathogens interact through a single coupling parameter C: values above 1 mean one infection facilitates the other, values below 1 mean competition.

The analysis proceeds in three stages rather than testing everything at once. First the authors study the coupled system with no seasonality and no behavioral mitigation. Then they add each mechanism separately. Finally they combine them and map stability across seasonal forcing, mitigation strength, and cooperativity. This staged design is what lets the paper attribute different oscillatory and chaotic regimes to different mechanisms instead of treating the full model as one opaque output.

A backward bifurcation makes outbreaks jump

Without external forcing, the coupled SIRS system shows an abrupt transition between the disease-free equilibrium and the endemic equilibrium. The authors trace this to a backward bifurcation for C > 2, in a region where both the disease-free and endemic states are stable. In plain terms, the system can sit in either of two stable states under the same conditions, so a small change can tip it from near-zero prevalence to sustained endemic infection.

This matters because it means the transition is not a smooth threshold that a single reproduction number fully describes. The paper reports that this first-order-like transition persists and is amplified relative to the earlier SIR-SIR case it builds on: bistability appears even for reproduction numbers below one. The authors attribute the effect to a positive feedback loop in which existing infections raise the force of infection on the other pathogen, and they note that comparable transitions arise in models with strong self-boosting feedback that outweighs susceptible depletion.

Seasonality and feedback oscillate on different clocks

Adding either mechanism alone can generate limit cycles near the tipping point, but for different reasons. Seasonality forces transmission at a slow, fixed rate set by the seasonal period — the paper fixes the seasonal frequency at 2π/360 days⁻¹ — independent of the current state of the epidemic. Behavioral feedback instead responds to the history of infection prevalence: when cases rise, risk awareness rises with a delay, and mitigation reduces transmission. The paper fixes the risk-perception delay at 36 days and varies the maximum mitigation strength.

The distinction explains why the two mechanisms do not simply add up. Because mitigation reacts to case history, its oscillations can match the time scale of rising case numbers, producing low-incidence limit cycles or a stable high-prevalence fixed point depending on how strong the maximum mitigation is. Seasonality, by contrast, keeps a fixed rhythm regardless of prevalence. The paper also uses a rate-scaling parameter κ to slow or speed disease dynamics relative to that fixed seasonal period, which is why κ appears as an axis in the stability maps.

Combined forcing alternates high and low years

When seasonality and behavioral feedback have comparable strength, the full model produces multi-annual wave patterns in which high-incidence cycles are driven by seasonality and low-incidence intervals are driven by behavioral feedback. The authors describe limit cycles that can alternate between low and high prevalence every second year. In their stability maps, fixed points are absent across the explored parameter space: only limit cycles or more complex long-term behavior appear.

The paper's numerical protocol matters for reading these maps. Prevalence colors show the average over 100 years after a transient of up to 900 years, and the number of distinct peaks in the time series is used as an early indicator of period-doubling and chaos. With weak seasonality (amplitude s = 0.25), multidimensional period-doubling cascades emerge across the combined space of maximum mitigation, cooperativity, and κ. These are model-generated regimes, not observations of any real outbreak.

Cooperation opens a second route to chaos

Cooperativity is not just a background parameter. For fast disease dynamics (low κ) and intermediate mitigation strengths, the paper observes a period-doubling cascade in the cooperativity parameter itself. This cascade interacts with the period-doubling behavior that arises from the interplay of mitigation and seasonality at low C, which the authors note was previously reported in single-disease models. The lower maxima of the cooperation-induced bifurcations appear to approach the disease-free equilibrium asymptotically, producing years of near-zero prevalence inside recurring endemic dynamics.

For stronger cooperation, a third cascade emerges at high maximum mitigation and eventually merges with the cooperativity-induced cascade, forming a structure the authors describe as resembling an Arnold tongue. The paper is explicit that the influence of cooperation is not confined to large C: coexisting attractors that appear at moderate cooperation first interrupt the mitigation-and-seasonality cascades, and chaotic signatures return as C increases further. The large values used for mapping (C = 15 and 25) are there to expose the full range of dynamics, not to serve as empirical estimates.

Symmetry and constant cooperation bound the claims

The model assumes the two diseases are symmetric in initial conditions and parameters, sharing rates and coupling. The authors argue this is reasonable for drifting strains of one pathogen or endemic diseases sharing seasonal cycles, but they note that real settings are rarely symmetric. They speculate, without testing, that unequal cooperativities could split the single first-order transition into two staggered ones, that distinct disease time scales could favor quasiperiodic rather than period-doubling dynamics, and that unequal mitigation or seasonality could desynchronize the two waves.

Two further limits sit close to the results. Cooperativity is treated as a constant multiplicative factor, whereas physiological facilitation should saturate at high co-infection pressure; the authors counter that the backward bifurcation and cooperation-induced period-doublings already emerge at moderate C ≳ 2. And the model is a deterministic mean-field ODE system, so it does not capture stochasticity such as noise-induced tipping below the deterministic critical transmission rate. The stated goal is to characterize the landscape of possible dynamics, not to time actual transmission events.

Cooperation changes which regimes are reachable

The paper's central comparison is between two sources of complex dynamics that a single-pathogen model would not contain. One comes from the interaction of seasonality and behavioral feedback and is already known from single-disease models. The other comes from inter-pathogen cooperativity acting through the backward bifurcation, and it produces its own period-doubling cascade and coexisting attractors. Because these two routes interact across parameter ranges, cooperation fundamentally expands the spectrum of possible epidemic patterns rather than merely shifting a threshold.

For readers deciding how to model or interpret a two-pathogen system, the practical priority follows from that comparison: when pathogens facilitate each other, a model built on one disease can miss regimes that only appear once cooperation and the bistability it induces are included. The paper's own framing is that cooperation must be explicitly accounted for in epidemic modeling and control strategies, because mitigation rules derived from single-disease assumptions may not remain effective when cooperating pathogens and changing spreading conditions are present.

Sources used in this answer