Construction of an isotropic cellular automaton for a reaction-diffusion equation by means of a random walk

We propose a new method to construct an isotropic cellular automaton corresponding to a reaction-diffusion equation. The method consists of replacing the diffusion term and the reaction term of the reaction-diffusion equation with a random walk of microscopic particles and a discrete vector field which defines the time evolution of the particles. The cellular automaton thus obtained can retain isotropy and therefore reproduces the patterns found in the numerical solutions of the reaction-diffusion equation. As a specific example, we apply the method to the Belousov-Zhabotinsky reaction in excitable media.

💡 Research Summary

The paper introduces a novel framework for translating continuous reaction‑diffusion equations (RDEs) into discrete cellular automata (CA) while preserving isotropy, a property often compromised in traditional lattice‑based models. The authors address two primary shortcomings of existing CA approaches: directional bias introduced by the underlying grid and the computational overhead associated with complex, anisotropic diffusion kernels. Their solution replaces the diffusion term with a stochastic random‑walk process and the nonlinear reaction term with a discrete vector field that governs state transitions of quantized chemical species.

In the diffusion component, each lattice site contains a certain number of “particles” representing concentration. At each time step, a particle moves to one of its neighboring sites (typically the four or eight nearest neighbors) with a probability p. This probability is derived from the continuous diffusion coefficient D via the relation p = D Δt / Δx², ensuring that the macroscopic diffusion behavior of the CA matches that of the original PDE when Δt and Δx are appropriately chosen. Because the movement is random and unbiased, the average flux is direction‑independent, guaranteeing isotropy regardless of the underlying square lattice. Moreover, the random‑walk formulation is trivially parallelizable, making it well‑suited for high‑performance computing.

For the reaction part, the authors discretize the Oregonator model of the Belousov‑Zhabotinsky (BZ) reaction. Continuous concentrations of key intermediates (e.g., HBrO₂, Br⁻, catalyst) are quantized into integer levels. A set of deterministic update rules—expressed as a vector field in the state‑space of these integers—specifies how each site’s state changes depending on local concentration thresholds. The rules embody the classic excitatory‑inhibitory dynamics: when an activation threshold is crossed, the site undergoes a rapid increase (the “excitation” phase), followed by a slower decay (the “recovery” phase). Crucially, these reaction rules are independent of the diffusion step, allowing the authors to tune excitation thresholds and recovery rates to match the temporal scales of the continuous BZ system.

Simulation results demonstrate two hallmark patterns of excitable media. First, a circular wavefront propagates outward from an initial stimulus without noticeable lattice‑induced distortion, confirming that the random‑walk diffusion faithfully reproduces isotropic spreading. Second, spiral waves self‑organize and rotate with a core and rotation period that closely match those obtained from direct numerical integration of the Oregonator PDEs. The authors also perform a sensitivity analysis: decreasing p below a critical value suppresses wave propagation (diffusion‑limited regime), while increasing p excessively blurs the wavefront (over‑diffusion). Adjusting the reaction thresholds shifts the excitability window, allowing precise control over wave initiation and refractory periods. These findings illustrate that the CA parameters can be directly mapped to physical quantities in experimental chemical reactors or biological tissues.

Beyond the BZ example, the authors argue that the methodology is generic. Any reaction‑diffusion system describable by a set of coupled PDEs can be recast into a CA by (i) deriving appropriate random‑walk probabilities from the diffusion coefficients and (ii) constructing a discrete vector field that approximates the nonlinear reaction kinetics. Because the random‑walk step is inherently isotropic and the reaction step is local, the resulting CA scales efficiently on parallel architectures, making it attractive for large‑scale simulations of pattern formation, morphogenesis, or neural activity where real‑time performance is essential.

In summary, the paper presents a clear, mathematically grounded procedure for building isotropic cellular automata that accurately emulate the dynamics of reaction‑diffusion equations. By decoupling diffusion and reaction into stochastic and deterministic discrete processes, respectively, the authors achieve isotropy, numerical stability, and a transparent correspondence between CA rules and physical parameters. This contribution advances the state of the art in discrete modeling of excitable media and opens avenues for efficient, large‑scale computational studies of complex spatiotemporal phenomena.