Cellular Automaton Frameworks

• Cellular Automaton Frameworks are formal systems comprising lattices of locally interacting finite-state machines used to model emergent phenomena.
• They include classical, non-uniform, graph-based, continuous, and asynchronous models that simulate a wide range of physical, biological, and computational systems.
• These frameworks utilize rigorous mathematical foundations and category-theoretic abstractions to analyze dynamics and drive innovation in unconventional computing.

Cellular automaton (CA) frameworks are formal mathematical and computational systems characterized by arrays of locally interacting finite-state machines. These frameworks offer a foundation for modeling emergent phenomena, computation, and pattern formation across a wide range of scientific and engineering disciplines. CA frameworks exhibit diversity in topology, state spaces, update mechanisms, and expressivity, enabling rigorous study of both discrete and continuous, as well as deterministic and stochastic, complex systems.

1. Classical and Standard CA Frameworks

The canonical CA framework, as summarized by comprehensive surveys (Bhattacharjee et al., 2016), consists of a finite or infinite lattice $\mathscr{L} \subseteq \mathbb{Z}^D$, a finite state set $\mathcal{S}$, a neighborhood template $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$, and a local transition function $f:\mathcal{S}^{N} \to \mathcal{S}$. The global configuration is $x \in \mathcal{S}^{\mathscr{L}}$ and the system evolves synchronously via: $$G(x)_{\mathbf{v}} = f( x_{\mathbf{v}+\mathbf{v}_1},\,\ldots,\,x_{\mathbf{v}+\mathbf{v}_N} )$$ This structure underlies both 1D binary rules (e.g., Wolfram's ECAs), 2D totalistic automata such as Conway's Game of Life, and higher-dimensional variants. Classical frameworks assume spatial and temporal homogeneity—every cell applies the same rule in lock-step.

Key theoretical properties include intrinsic universality (existence of rules that can emulate any Turing machine), reversibility (bijective evolution), and conservation laws. The Curtis–Hedlund–Lyndon theorem formalizes that any CA is a continuous, shift-commuting map on $\mathcal{S}^{\mathscr{L}}$ with a finite neighborhood (Bhattacharjee et al., 2016Metkar et al., 3 Jan 2026).

2. Extensions and Generalizations

Non-uniform and Heterogeneous CA

Non-uniform CAs relax homogeneity, allowing cell-specific rules, varying update schedules (asynchronous CA), or variable neighborhood topologies. Hybrid CA frameworks assign each cell its own transition rule (rule vector), supporting hybrid-ECA rings, automata on graphs, or models of spatial heterogeneity in physical or biological systems (Bhattacharjee et al., 2016).

Extended and Graph-based CA

Graph-based cellular automata (GCA) replace regular lattices with arbitrary graphs, generalizing the neighborhood $N(i)$ to be determined by the (possibly time-dependent) edge structure of a graph $G=(V,E)$ (Grattarola et al., 2021). The state-update for node $i$ is: $\mathcal{S}$0 Graph CA admit both fixed and time-evolving topologies, supporting applications such as distributed control, flocking, and learnable emergent coordination.

Extended Cellular Automata (XCA), as formalized by dynamic-graph systems, treat the arcs of a graph as "cells," allowing topological rewiring at each time step. The XCA specification defines synchronous local state updates $\mathcal{S}$1 and local neighbor-exchange rules $\mathcal{S}$2, leading to models with emergent spatial structure and universality properties distinct from classical lattice-based CA (Eckel, 2015).

Recursive Estimation Extensions

The "Recursive Estimation of Neighbors" (REN) framework generalizes the local update radius by introducing a cell-wise perception area $\mathcal{S}$3 exceeding the base rule radius $\mathcal{S}$4 (Kayama, 2015). Local updates for perception radius $\mathcal{S}$5 are defined recursively, using base rule $\mathcal{S}$6 to estimate deeper neighborhood states, providing a parameterized spectrum of nonlocal interaction and information propagation.

Continuously-valued and Lenia-type CA

Frameworks like Lenia and Glaberish extend classical binary CA to continuously-valued cell states. In original Lenia, each cell $\mathcal{S}$7 is updated via convolutional neighborhood aggregation and a nonlinear growth function $\mathcal{S}$8. Glaberish, an extension, splits the update into conditional "genesis" ($\mathcal{S}$9) and "persistence" ($$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$0) functions, assigned according to the current cell's state, fully recovering the expressive power of all Life-like $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$1 rules in the continuous domain (Davis et al., 2022). This enables simulation of a wide class of complex, class IV automata, including those whose birth and survival intervals are disjoint.

Asynchronous and Stochastic CA

Asynchronous CA decouple cell updates from global synchronization, typically via random sequential single-site updates. This approach avoids correlation artifacts present in synchronous schemes and is better suited to modeling, e.g., diffusive processes and realistic physical dynamics (Ribera et al., 2019). Stochastic CA further generalize update rules to return probabilistic state transitions, crucial for modeling noisy, physically-motivated, or adaptive systems.

3. Categorical, Functional, and Advanced CA Abstractions

Category-theoretic frameworks have generalized the CA definition to arbitrary categories with products, $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$2, replacing the classical set-theoretic structure (Castillo-Ramirez et al., 3 Feb 2026). In this setting, a $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$3-CA is a morphism $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$4 (with $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$5, $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$6 a group) that is equivariant and uniform in the sense of the categorical Curtis–Hedlund–Lyndon theorem.

Comonadic CA frameworks, as implemented in functional programming environments, encode CA arrays and neighborhoods as comonad instances, naturally supporting efficient, mathematically transparent, and stochastic (via monad stacks) CA simulations. These models allow arbitrary network topologies, multiple layers of parameters and random number streams, and close correspondence to abstract categorical semantics (Sas et al., 26 Dec 2025).

LOGOS-CA introduces a natural-language-based extension, mapping cell states and rules to natural language strings, with updates delegated to an LLM. This approach enables CA state spaces and transition functions to be defined in linguistic terms, extending expressivity beyond strictly numerical or finite automata (Utimula, 18 Jan 2026).

4. Structural, Dynamical, and Statistical Physics Perspectives

Modern CA frameworks span deterministic, probabilistic, and totalistic models, supporting classification into Wolfram classes (I–IV), reversibility, and various symmetry and conservation laws (Metkar et al., 3 Jan 2026). Macroscopic observables relevant to statistical physics—such as entropy rates, correlations, and transport coefficients—are computed via spatial and temporal statistics, discrete continuity equations, and Green–Kubo formulas. Universality classes, including directed percolation (DP) and KPZ/Burgers, inform the emergent behavior and physical analogues of CA models.

Coarse-graining, block renormalization, and computational mechanics (e.g., $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$7-machines) have been developed to analyze memory, complexity, and hydrodynamic limits of CA, bridging fine-grained rules with emergent PDEs and effective field theories (Metkar et al., 3 Jan 2026).

5. Computational, Algorithmic, and Implementation Aspects

CA evaluation can be encoded via category-theoretic models (comonads, monads), facilitating formal compositional reasoning, code generation, and simulation (Sas et al., 26 Dec 2025). Design choices—such as synchronous vs. asynchronous scheduling, fixed vs. adaptive topologies, deterministic vs. stochastic update rules, and continuous vs. discrete cell states—directly influence computational complexity, scalability, and the types of physical phenomena accessible.

Architectures for implementing CA include hardware (memristor arrays, photonic cavities), microfluidic and chemical substrates, modular robotics, and hybrid digital-analog co-processors. Each advances the practical deployment of CA-based computation and modeling, in some cases yielding near-thermodynamic-limit energy efficiency and robustness to faults (Martinez et al., 8 Aug 2025).

Diagrammatic methods formalize CA construction via alphabet factorization and layer coupling, providing a hierarchical, visual, and algebraic approach to building complex CA with nested structures, graded and $$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$8-decomposable rules (García-Morales, 2016).

6. Applications and Emerging Directions

CA frameworks have broad applications as summarized by (Bhattacharjee et al., 2016Martinez et al., 8 Aug 2025Davis et al., 2022León, 20 Mar 2026):

• Modeling of physical systems: lattice gas fluids, reaction–diffusion, crystal growth, and quantum field dynamics.
• Complex systems and artificial life: collective pattern formation, morphogenesis, excitable media, ecosystem modeling, information processing, artificial worlds with emergent geometry (XCA).
• Computation theory and unconventional computing: demonstration of Turing-universality, encoding of algorithms as CA trajectories, design of post-von Neumann architectures.
• Statistical physics: transport phenomena, disorder-to-order transitions, self-organized criticality.
• Device simulation and materials design: real-time simulation of thermal transport in nanostructures, scalable exploration of defect-laden or irregular geometries (León, 20 Mar 2026).

Emergent areas include hybrid natural language/numeric CA (LOGOS-CA (Utimula, 18 Jan 2026)), information-theoretic classification and glider discovery in continuous-state domains (Lenia/Glaberish (Davis et al., 2022)), and exploration of category-theoretic unification across CA model classes (Castillo-Ramirez et al., 3 Feb 2026).

7. Comparative Table of Representative CA Frameworks

Framework	Topology	State Space	Update Rule	Special Features/Expressivity
Standard CA	Lattice	Finite	$$\mathcal{N} = (\mathbf{v}_1,\ldots,\mathbf{v}_N) \subset \mathscr{L}-\mathscr{L}$$9	Synchronous, uniform, rule-table
Non-uniform/Hybrid CA	Lattice	Finite	Cell-wise $f:\mathcal{S}^{N} \to \mathcal{S}$0	Heterogeneity, asynchronous
Graph CA	Arbitrary Graph	Finite/Continuous	$f:\mathcal{S}^{N} \to \mathcal{S}$1	Irregular topology, message-passing
XCA	Dynamic Graph	Finite	$f:\mathcal{S}^{N} \to \mathcal{S}$2 with rewiring $f:\mathcal{S}^{N} \to \mathcal{S}$3	Evolving connectivity, emergent space
Lenia/Glaberish	Lattice	[0,1]	Convolution $f:\mathcal{S}^{N} \to \mathcal{S}$4 $f:\mathcal{S}^{N} \to \mathcal{S}$5	Continuous-state, full $f:\mathcal{S}^{N} \to \mathcal{S}$6 Life-like rules
Category-theoretic CA	Grothendieck site	Obj($f:\mathcal{S}^{N} \to \mathcal{S}$7)	Morphism in $f:\mathcal{S}^{N} \to \mathcal{S}$8	Generalized, product-closed, equivariant
Comonadic CA	Array/Network	Any	CoKleisli arrow	Functional, N-D, stochastic monad stack
LOGOS-CA	Lattice	Textual	LLM-interpreted NL rule	Natural language state/rule, LLM-in-the-loop

These CA frameworks constitute a core infrastructure for studying, implementing, and understanding both computation and emergence in dynamical systems, with ongoing innovation of new frameworks that increasingly unify discrete, continuous, linguistic, and category-theoretic paradigms.