Cross-Diffusion SIR Models

• Cross-diffusion SIR models are advanced epidemiological frameworks that incorporate nonlinear spatial coupling to reflect density-driven movements.
• They simulate accelerating epidemic fronts, emergent buffer zones, and rich spatiotemporal patterns through directed cross-diffusive terms.
• Analytical and numerical studies reveal stability transitions and localized pattern formations that can inform targeted spatial interventions.

Cross-diffusion SIR models generalize classical Susceptible-Infected-Recovered (SIR) epidemiological dynamics to incorporate the spatial coupling of population movement and infection spread through nonlinear cross-diffusive terms. These models are derived as the continuum limits of discrete multi-patch or lattice-based SIR frameworks, where population groups interact locally as well as with neighboring sites. Unlike traditional reaction–diffusion models, cross-diffusion SIR equations feature fluxes that depend on the product of local densities and spatial gradients of opposite compartments, typically encoding directed, “commuting”-type movement. The result is a system of coupled nonlinear partial differential equations (PDEs) that produce richer spatiotemporal patterns, including accelerating traveling waves, the emergence of infection buffer zones, and complex wave phenomena due to vital population turnover (Ghosh, 1 Feb 2025, Ahmadpoortorkamani et al., 2024).

1. Mathematical Formulation and Derivation

The prototypical cross-diffusion SIR model arises by considering a chain (or lattice) of interacting patches, each following standard SIR dynamics but with local infection terms extended to include contributions from immediate spatial neighbors. Taking the continuum limit as the patch spacing $\Delta x \to 0$ leads to PDEs of the generic form:

$$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$

where $S(x, t)$, $I(x, t)$, $R(x, t)$ denote the densities of susceptibles, infecteds, and recovered, $A(x)$ and $B(x)$ are spatially varying contact and recovery rates, and $D(x)$, $D_R(x)$ are cross-diffusion coefficients (Ghosh, 1 Feb 2025). In higher dimensions, spatial derivatives generalize to gradients and Laplacians.

A related two-dimensional formulation incorporating vital dynamics (birth, death, logistic growth) is given by (Ahmadpoortorkamani et al., 2024):

$$\begin{aligned} &\partial_t S = -D \beta S \Delta I - \beta S I + r S - \mu S - (r/K) S^2 \ &\partial_t I = D \beta S \Delta I + \beta S I - m I \ &\partial_t R = \gamma I - \mu R \end{aligned}$$

where $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$0 is the birth rate, $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$1 is the natural death rate, $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$2 is carrying capacity, $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$3 is the recovery rate, $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$4 is the disease-induced death rate with $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$5, and $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$6 controls the cross-diffusive scale.

The mathematical hallmark of these models is the appearance of nonlinear fluxes (e.g., $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$7), distinguishing them from standard reaction–diffusion equations with ordinary Fickian diffusion ($$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$8 or $$\begin{aligned} &\partial_t S = -A(x) S I - \partial_x [ D(x) S \,\partial_x I ] \ &\partial_t I = A(x) S I - B(x) I + \partial_x [ D(x) I\, \partial_x S ] \ &\partial_t R = B(x) I + D_R(x) \Delta R \end{aligned}$$9).

2. Physical and Epidemiological Interpretation

Cross-diffusion in the SIR context represents nonrandom, directed individual movement driven by local infection gradients. For example, the flux $S(x, t)$0 in the susceptible equation quantifies the outflow of susceptibles from patches experiencing an influx of infectives, modeling heightened infection pressure due to spatial concentration gradients. The reciprocal term $S(x, t)$1 in the infected equation expresses how infective individuals are drawn toward regions rich in susceptibles, enhancing infection spread in areas of high susceptibility.

Physically, these mechanisms correspond to scenarios where individuals ‘commute’ between home regions and neighboring sites, interacting based on local density and returning home—a process formally analogous to discrete commuting models, and not captured by standard diffusion. Parameter $S(x, t)$2 quantifies how heterogeneity in infection density drives movement, with higher values leading to more rapid, spatially-driven infection propagation or mixing (Ghosh, 1 Feb 2025, Ahmadpoortorkamani et al., 2024).

3. Analytical Properties and Linear Stability

Analysis of cross-diffusion SIR models centers on spatially homogeneous fixed points and their stability. Two principal equilibria exist: the disease-free equilibrium (DFE), where $S(x, t)$3, and the endemic equilibrium (EE), where $S(x, t)$4.

Linearization about these equilibria reveals key features:

• At the DFE: The infection mode grows if the basic reproduction number $S(x, t)$5 exceeds unity, but the cross-diffusion term introduces a scale-dependent decay: the infection eigenvalue is

$S(x, t)$6

Thus, long-wavelength spatial instabilities dominate threshold behavior (Ahmadpoortorkamani et al., 2024).

• At the EE: The system can experience high-wavenumber instabilities (“dark-spike” phenomena) when birth exceeds death ($S(x, t)$7) in the presence of cross-diffusion, such that short-scale patterns can transiently grow before saturating nonlinearly. Stability diagrams depend sensitively on $S(x, t)$8, $S(x, t)$9, $I(x, t)$0, $I(x, t)$1, and $I(x, t)$2 (Ahmadpoortorkamani et al., 2024).
• Pattern formation: Nonlinear cross-diffusion is sufficient to trigger Turing-like pattern formation when spatial heterogeneity in $I(x, t)$3, $I(x, t)$4 is present or when vital population turnover generates oscillatory regimes (Ghosh, 1 Feb 2025).

4. Spatiotemporal Phenomena: Waves and Buffer Zones

Simulations and analytical studies reveal several emergent spatiotemporal behaviors unique to cross-diffusion SIR models:

• Traveling epidemic fronts: Localized infection seeds split into accelerating or decelerating traveling wavefronts, with wave speed and acceleration set by $I(x, t)$5, $I(x, t)$6, $I(x, t)$7. The front speed can exceed that of classical reaction–diffusion models and may follow non-constant temporal scaling, $I(x, t)$8 with an exponent $I(x, t)$9 dependent on $R(x, t)$0 (Ahmadpoortorkamani et al., 2024).
• Buffer zones and fire-breaks: In regions where the local reproductive ratio $R(x, t)$1, infection can decay exponentially, causing propagating fronts to stall and resulting in stationary buffer zones or “fire-breaks.” This leads to long-lived uninfected areas (“islands” in higher dimensions), contingent on spatial heterogeneity in $R(x, t)$2, $R(x, t)$3, or $R(x, t)$4 (Ahmadpoortorkamani et al., 2024).
• Quasi-periodic and recurrent waves: With vital dynamics ($R(x, t)$5), recurring epidemic waves emerge as susceptibles regrow in the wake of a previous outbreak. The delay and amplitude of secondary waves are acutely sensitive to demographic parameters and initial conditions, producing phenomena such as concentric infectious rings in two dimensions (Ahmadpoortorkamani et al., 2024).
• Transient heterogeneity and dark spikes: In parameter regimes susceptible to high-wavenumber instability, narrow “dark spikes,” i.e., transient troughs of low susceptible density, can grow and subsequently heal, representing self-organized protected corridors (Ahmadpoortorkamani et al., 2024).

The key distinguishing aspect is the prevalence of accelerating, pattern-forming, and spatially-resonant phenomena that do not arise in well-mixed or classical diffusive SIR frameworks (Ghosh, 1 Feb 2025, Ahmadpoortorkamani et al., 2024).

5. Numerical Methods and Boundary Conditions

Typical numerical approaches employ explicit or implicit finite-difference (second-order central in space) or finite-element discretizations in space, and (possibly operator-split) time-stepping schemes such as forward Euler or IMEX to accommodate stiff reaction terms and nonlinear cross-diffusive couplings (Ghosh, 1 Feb 2025, Ahmadpoortorkamani et al., 2024). Neumann (no-flux) or periodic boundary conditions are standard, ensuring conservation of total population within the computational domain.

Representative simulations use initial conditions corresponding to either uniform susceptibles with localized infection seeds or spatially heterogeneous $R(x, t)$6 to probe buffer zone formation, traveling front dynamics, or wave recurrence. Parameters are chosen to probe varying $R(x, t)$7, $R(x, t)$8, $R(x, t)$9, and vital rates ($A(x)$0, $A(x)$1, $A(x)$2).

Pattern Class	PDE Form	Initial/Boundary Condition
Traveling Waves	No vital dynamics	$A(x)$3, $A(x)$4, $A(x)$5
1D Buffer Zones	No vital dynamics	$A(x)$6, $A(x)$7, $A(x)$8
Quasi-periodic 2D Waves	With vital dynamics	$A(x)$9, $B(x)$0, $B(x)$1
Dark Spikes (“corridors”)	With vital dynamics	$B(x)$2, $B(x)$3, $B(x)$4

Key observations include robust front acceleration/deceleration, long-lived stationary buffer zones, and intricate spatial recurrence patterns depending on parameter regime (Ahmadpoortorkamani et al., 2024).

6. Implications, Applications, and Distinctions from Classical SIR

Cross-diffusion SIR models highlight how directed, density-dependent spatial movement fundamentally alters epidemic spread relative to both well-mixed and classical reaction–diffusion settings. The cross-diffusive structure generically predicts:

• Front speeds that are nonconstant and can exceed those of standard models.
• Self-organizing fire-breaks and buffer zones driven by local susceptibility and migration structures.
• Persistent, quasi-periodic spatial outbreaks driven solely by demographic renewal, even in the absence of seasonal forcing.
• High-wavenumber instabilities that generate transient structured heterogeneity (“dark spikes”), corresponding to protected demographic regions that spontaneously emerge and vanish.
• Scenarios where targeted spatial interventions (e.g., commuting restrictions, lockdowns in regions of high susceptibility) can exploit these mechanisms to impede epidemic propagation more effectively than uniform mitigation strategies (Ghosh, 1 Feb 2025, Ahmadpoortorkamani et al., 2024).

A plausible implication is that spatially structured demographic and behavioral interventions, informed by cross-diffusive SIR frameworks, can improve outbreak control—by anticipating buffer formation, leveraging endogenous fire-breaks, or managing secondary wave recurrence.

Key differences from well-mixed ODE SIR models include the existence of traveling fronts, rich spatiotemporal patterning, spatially pinned infection, and the persistence of endemic “hot spots” via migration. Reaction–diffusion SIR with only Fickian diffusion ($B(x)$5, $B(x)$6) cannot capture these effects: it always predicts monotonic front propagation and homogenization, lacking acceleration and buffer zone phenomena characteristic of cross-diffusion (Ghosh, 1 Feb 2025, Ahmadpoortorkamani et al., 2024).