Behavioral-Epidemiological Model Dynamics

• Behavioral-Epidemiological Model is a framework that integrates adaptive individual behavior with SIS epidemic dynamics to study emergent phase transitions and epidemic thresholds.
• The model employs agent-based simulations combining random walk and preventive moves to capture changes in spatial clustering and contact network metrics.
• Parameter variations such as agent density and avoidance rate uncover hysteresis, bistability, and novel epidemic control phenomena affecting disease spread.

A Behavioral-Epidemiological Model is one in which epidemic dynamics are explicitly coupled to adaptive individual behavior—typically incorporating feedback mechanisms whereby agents respond to perceived infection risk by modifying their contacts, mobility, or prevention strategies. Such models integrate micro-scale behavioral rules and macro-scale epidemic states to capture emergent phenomena, phase transitions, and complex feedbacks not present in classical compartmental frameworks.

1. Agent-Based Model Framework and Spatial Dynamics

The archetypal behavioral-epidemiological model presented in "Epidemic spreading in populations of mobile agents with adaptive behavioral response" (Ventura et al., 2021) is an agent-based system embedded in two-dimensional space. $N$ agents are initialized uniformly in a square of side $L$ with periodic boundary conditions, where each agent $i$ has a position $(x_i(t), y_i(t))$ and discrete disease state $\sigma_i(t) \in \{\mathrm{S}, \mathrm{I}\}$.

Mobility is governed by two mechanisms:

• Random Walk (RM): At each time step $t \to t+1$, each agent moves distance $v$ in a uniformly random direction $\theta_i$.

$$x_i(t+1) = x_i(t) + v\cos\theta_i, \quad y_i(t+1) = y_i(t) + v\sin\theta_i$$

• Adaptive Preventive Move (PM): Each susceptible agent, with probability $a$, chooses to move directly away from one randomly chosen infected neighbor $j$ (within a fixed radius $r$), specifically:

$$\vec{r}_i(t+1) = \vec{r}_i(t) + v \frac{\vec{\Delta}}{||\vec{\Delta}||},\quad \vec{\Delta} = \vec{r}_i(t) - \vec{r}_j(t)$$

Agents not adopting PM revert to RM. Interaction and transmission occur whenever agents are within Euclidean distance $r$.

2. Coupling Mobility with Epidemic SIS Dynamics

At each timestep, an SIS (Susceptible-Infected-Susceptible) model is superimposed:

• An S–I pair in contact induces infection with probability $\beta$ in the following step.
• An infected agent recovers (I→S) with probability $\mu$.

The instantaneous contact network defines $L_{SI}(t)$, the count of S–I edges. The normalized contact density is

$l_{SI}(t) = \frac{L_{SI}(t)}{N}$

In the fast-mobility, slow-disease regime ($\mu, \beta \ll v$), prevalence evolves as

$$\dot{\rho}_I = \beta \, l_{SI}(\rho_I) - \mu\,\rho_I,\qquad \rho_I = \frac{N_I}{N}$$

where $l_{SI}(\rho_I)$ is treated as a prevalence-dependent mean-field function.

3. Semi-Analytic Fast-Mobility Approach: Fixed Points and Thresholds

To analyze stationary states, define the effective infection rate $\lambda = \beta/\mu$ and solve

$\rho_I^* = \lambda\,l_{SI}(\rho_I^*)$

Graphically, fixed points are intersections of $y = \lambda\,l_{SI}(\rho_I)$ with $y = \rho_I$; stability depends on the local slope $s(\rho) = d[\lambda l_{SI}]/d\rho$ with $s(\rho^*) < 1$ denoting stability.

The epidemic threshold is characterized as

$\lambda_c = \frac{1}{l_{SI}'(0)}$

For homogeneous mixing $l_{SI}(\rho) \approx k_H\,\rho$,

$k_H = \frac{N}{L^2}\pi r^2$

and the threshold simplifies to $R_H = \lambda\,k_H = 1$.

Preventive avoidance ($a > 0$) increases $\lambda_c$, thus raising the threshold for endemicity.

4. Phase Transitions: Bistability, Clustering, and Hysteresis

In the high-density, strong-avoidance regime ($k_H$ large, $a \to 1$), the system exhibits:

• Saddle-node bifurcation: $\lambda\,l_{SI}(\rho)$ develops a maximum above the diagonal, yielding three fixed points (disease-free, unstable intermediate, stable endemic).
• Hysteresis: Slowly varying $\lambda$ or $R_H$ leads to bistable epidemic outcomes depending on direction; the prevalence $\rho_I^*$ demonstrates a loop between transcritical and saddle-node regime boundaries.
• S-cluster formation: The bistable regime features transient spatial clusters ("flocks") of susceptible agents, reflected in sharply elevated $k_{SS}$ and depleted $k_{SI}$ compared to $k_H$. Stochastic cluster resilience or breakdown governs long-term disease outcomes.

5. Network Metrics and Contact Graph Structure

In the fast-mobility stationary state, the contact graph is analyzed by degree distributions:

• Degrees:
  • $k_S(\rho)$: Mean degree of S-nodes
  • $k_I(\rho)$: Mean degree of I-nodes
  • $k_{SS}, k_{SI}, k_{II}, k_{IS}$: Cross-type degree distributions
• Mixing and clustering:
  • Homogeneous mixing yields Poisson degree distributions.
  • Behavioral avoidance shifts $k_{SS}$ upwards, strongly suppresses $k_{SI}$, and leaves $k_{II}$ quasi-Poisson.
  • Cluster formation is flagged by peaks in normalized degrees $k_S/k_H$, clustering coefficient $C(\rho_I)$, and degree assortativity $r_{deg}(\rho_I)$ near the unstable fixed point $\rho_u$.

6. Model Implications and Key Conclusions

• For low agent density, behavioral avoidance continuously raises the epidemic threshold and lowers endemic prevalence, analogously to adaptive-network SIS models with contact rewiring.
• At higher densities, preventive mobility triggers a genuine discontinuous transition, bistability, and hysteresis between healthy and endemic equilibria.
• The semi-analytic measurement of $l_{SI}(\rho_I)$ from simulations succinctly reproduces both bifurcation types and forecasts stationary prevalence as a function of system parameters.
• Time-averaged networks reflect increased clustering and positive assortativity in the bistable regime, corresponding to S-cluster dominance.
• Even localized, minimal behavioral rules in mobile agent populations qualitatively alter epidemic phase behavior, merging spatial and adaptive-network phenomenology.

Table: Dynamical Regimes in the Behavioral-Epidemiological Model

Density (k_H)	Avoidance (a)	Epidemic Threshold	Bifurcation Type	Contact Structure
Low	$a>0$	Raised	Transcritical (continuous)	Poisson-like, weak clustering
High	$a\to1$	Sharply raised	Saddle-node (bistable)	S-clusters, high clustering

This behavioral-epidemiological framework provides a rigorous platform for empirical or simulation-based investigations into epidemic control by mobility, local risk avoidance, and the induced coarse-grained temporal network structures. It highlights that even basic feedback from individual behavioral adaptation can produce fundamentally novel epidemic phase phenomena including spatial clustering, discontinuous transitions, and strong hysteresis (Ventura et al., 2021).