Molecular Automata: Mechanisms and Models

• Molecular automata are defined as molecular-scale information-processing systems whose discrete states and local interactions implement computation and memory.
• Research spans chemical reaction networks, RNA and protein-based state machines, and quantum-dot cellular automata, each employing unique transition rules and logic.
• These approaches emphasize tradeoffs among expressiveness, energy efficiency, and robustness, advancing nanoscale circuit design and biomolecular computation.

Searching arXiv for papers on "molecular automata" and adjacent formulations to ground the synthesis. Searching arXiv for papers on "molecular automata" and adjacent formulations to ground the synthesis. Molecular automata are molecular-scale information-processing systems in which discrete internal states, local interactions, and physically realized transition rules implement computation, memory, pattern recognition, or language recognition. In current research, the term spans several distinct lineages: chemically encoded language recognizers in one-pot reactors and input/output chemical reaction networks, RNA and polymer systems whose transformations realize automaton transitions, protein and protein-complex models with explicit state machines or asynchronous cellular-automaton dynamics, charge-based molecular quantum-dot cellular automata (QCA), and spatially extended substrates such as actin filaments or reconfigurable molecular layers (Duenas-Diez et al., 2019, Klinge et al., 2015, Zhang et al., 2024, Kocka et al., 21 Aug 2025, Blair, 2018).

1. Conceptual scope and formal vocabularies

The literature does not use a single universal definition of molecular automata. In “Rule-based Modeling and Simulation of Biochemical Systems with Molecular Finite Automata” (Yang et al., 2010), a molecular finite automaton is defined as

$M=(E_1,E_2,\ldots,E_n,\mathbf v),$

a collection of component extended finite automata sharing a variable structure $\mathbf v$. This formalism treats proteins as structured computing machines whose site- or domain-level transitions depend on biochemical inputs and on predicates over contextual variables. In “Automaton of molecular perceptions in biochemical reactions” (Maestri et al., 2021), the reaction automaton is extended to

$$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$

where perceiving states and $\epsilon$-transitions explicitly represent local sensing and aborted encounters. In “Single-molecule Automata: Harnessing Kinetic-Thermodynamic Discrepancy for Temporal Pattern Recognition” (Zhang et al., 2024), the automaton states are the $2^N$ conformations of a mechanically driven polymer. In “Nonequilibrium protein complexes as molecular automata” (Kocka et al., 21 Aug 2025), a ring of binary monomers under context-dependent enzymatic reactions maps, in the strongly driven limit, onto a stochastic, asynchronous cellular automaton whose local rules correspond to enzyme sets and Wolfram rule numbers.

Across these formalisms, several common structural elements recur. First, the state space is molecularly embodied: conformations, phosphorylation states, charge localizations, bound/unbound complexes, or occupancy patterns serve as automaton states. Second, transition semantics are local: neighboring monomers, nearby charges, cognate ligands, or adjacent polymer segments determine allowed updates. Third, computation is usually interpreted through coarse-grained observables rather than microscopic trajectories. In the single-polymer DFA model, for example, determinism is claimed for the dominant configuration rather than for every stochastic molecular path (Zhang et al., 2024). In the nonequilibrium protein-complex model, the automaton is a continuous-time Markov process whose attractors and transients are defined by the transition graph and generator $W$ rather than by synchronous symbolic rewrites (Kocka et al., 21 Aug 2025).

This diversity is substantive rather than terminological. Some molecular automata are designed as explicit finite-state devices; some emerge from thermodynamically consistent kinetics; some are chiefly models of biological information processing; and some are architectural proposals for molecular hardware. A plausible implication is that “molecular automata” now denotes a research program organized by state, locality, and physical embodiment, rather than a single canonical formalism.

2. Chemical language recognition and biomolecular compilation

One major tradition realizes automata directly in chemistry by mapping input symbols to sequential aliquots of reactants. “Native Chemical Automata and the Thermodynamic Interpretation of Their Experimental Accept/Reject Responses” (Duenas-Diez et al., 2019) gives non-biochemical realizations of a finite automaton, a push-down automaton, and a Turing-machine-level recognizer. The regular-language example uses

$$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$

to accept words containing both reactants via precipitate formation. The PDA example encodes the Dyck language of balanced parentheses by mapping “$($” to $\mathrm{NaOH}$, “$)$” to malonic acid, and stack height to pH relative to the midpoint of the titration curve. The context-sensitive example uses the Belousov–Zhabotinsky reaction to recognize

$\mathbf v$0

with $\mathbf v$1, $\mathbf v$2, and $\mathbf v$3 encoded as sodium bromate, malonic acid, and $\mathbf v$4, respectively. The paper also introduces thermodynamic metrics: enthalpy-based interpretation for the FA and PDA, and an area functional over the final oscillatory redox signal for the Turing-machine-level system, related to Gibbs free energy through the Nernst relation (Duenas-Diez et al., 2019).

A second line of work compiles automata into reaction networks. “Robust Biomolecular Finite Automata” (Klinge et al., 2015) gives a uniform translation from any NFA

$\mathbf v$5

to a deterministic mass-action input/output CRN. The input alphabet is encoded by species

$\mathbf v$6

and each symbol is presented as a three-phase event consisting of reset, symbol, and copy pulses. For each automaton state $\mathbf v$7, the construction uses state species $\mathbf v$8, portal species $\mathbf v$9, and dual species $$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$0. Transition logic is implemented by reset, transition, copy-back, and restoration reactions, so that the reachable NFA state set is represented by which $$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$1 are high. The main robustness theorem states that if

$$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$2

then the constructed I/O CRN $$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$3-satisfies the NFA-recognition requirement under perturbations of the input signal, output measurement, initial concentrations, and rate constants (Klinge et al., 2015).

A third development makes locality itself an automata invariant. “Bandwidth of Nondeterministic Finite Automata” (Cho et al., 30 May 2026) is motivated by co-transcriptional splicing, where an NFA is encoded along a circular RNA/DNA template and long-range deletions become chemically implausible. The paper defines a $$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$4-bandwidth NFA by requiring a cyclic ordering of states such that

$$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$5

This yields a strict hierarchy $$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$6. For finite languages, the main positive result is

$$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$7

while

$$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$8

The paper also gives a polynomial-time decision procedure for bandwidth $$\mathcal R'=(Q',\Sigma\cup\{\epsilon\},\delta',s_0,\{f\}),$$9 on finite languages presented as explicit word lists, with runtime $\epsilon$0, and proves that minimizing bandwidth-constrained NFAs is NP-hard for fixed $\epsilon$1 (Cho et al., 30 May 2026). Locality is therefore not just a fabrication nuisance; it changes expressiveness and synthesis complexity.

A more formal-language-theoretic reframing appears in “A Quantum Finite Automata Approach to Modeling the Chemical Reactions” (Bhatia et al., 2020), which models reaction patterns with two-way quantum finite automata. The paper uses 2QFA to recognize regular, context-free, and context-sensitive reaction languages, including balanced-parenthesis and $\epsilon$2 patterns, and emphasizes linear-time halting with bounded error. This is an automata-theoretic abstraction rather than a physical coherent-chemistry proposal, but it extends the language-recognition interpretation of chemical computation into the quantum-automata setting (Bhatia et al., 2020).

3. RNA and single-chain molecular automata

RNA-based proposals take the program-data duality of nucleic acids as central. “An Ansatz for computational undecidability in RNA automata” (Svahn et al., 2020) builds an automata hierarchy from three operations on RNA polymers: ligation $\epsilon$3, cleavage $\epsilon$4, and stasis $\epsilon$5. On that basis it defines an RNA finite automaton, an RNA pushdown automaton, an RNA two-stack PDA, and a universal RNA automaton. The RNA-FA uses state polymers, symbol polymers, and reactions such as

$\epsilon$6

The RNA-2PDA is identified with Turing-machine power through two stack polymers, and the RNA-UPDA simulates 2-tag systems by matching complementary symbols and copying right-hand sides of rules. The paper’s main conceptual move is then to connect universality to self-reference and undecidability: once an RNA automaton can encode its own description and manipulate that encoding, Liar-style and halting-style constructions become available (Svahn et al., 2020).

The single-molecule polymer DFA of (Zhang et al., 2024) is structurally different. Its substrate is a mechanically driven linear polymer of $\epsilon$7 foldable binary-state units, so the configuration space has exactly $\epsilon$8 states. The automaton alphabet is the set of extension controls $\epsilon$9, and the transition function is defined by dominant-state evolution under a designed nonequilibrium protocol. The paper introduces the “energy seascape,” meaning an energy/free-energy landscape extended by control-parameter degrees of freedom, and uses a deliberate kinetic-thermodynamic discrepancy: thermodynamically stable units are the slowest to fold, while the least stable are the fastest. In this regime, the dominant configuration obeys three deterministic rules—pop-unfold, fold, and exchange—and can recognize temporal input patterns. The paper further claims complete controllability: for $2^N$0, all

$2^N$1

configurations are made dominant by suitable non-equilibrium protocols (Zhang et al., 2024).

These RNA and single-polymer proposals embody two distinct notions of molecular automata. RNA automata are symbolic and rule-centered, with polymers serving as explicit encodings of states, stacks, and programs. The mechanically driven polymer is conformational and dynamical, with computation arising from metastable transient orderings in a controlled energy seascape. Both, however, exploit the same general principle: a single molecular species can carry both state and update-enabling structure.

4. Protein-centered automata and biochemical agency

Protein-oriented models often treat biomolecules as agents with modular internal state. In (Yang et al., 2010), proteins are represented as molecular finite automata built from component extended finite automata. State transitions are triggered by inputs such as ligand binding, phosphorylation, or dissociation, while internal variables store context such as binding partners or occupancy. Reaction rules synchronize transitions across multiple machines, so protein-protein interactions become coordinated machine reconfigurations. The formalism supports both deterministic ODE-based approximations and exact stochastic simulation, and the paper illustrates it with a scaffolded MAP kinase cascade involving SCF, M3K, M2K, and MPK (Yang et al., 2010). The significance of this work lies in its treatment of biochemical signaling as executable, context-sensitive state-machine dynamics rather than as a flat species-reaction list.

“Automaton of molecular perceptions in biochemical reactions” (Maestri et al., 2021) extends this agent view by separating perception from binding. Its perception-based reaction automaton introduces stable states $2^N$2, perceiving states $2^N$3, and a nondeterministic transition function over $2^N$4. The perception map

$2^N$5

formalizes how an enzyme in a stable state enters a perceiving state when it encounters a cognate enabling molecule. Binding may then succeed or fail, with $2^N$6-transitions returning the automaton to the prior stable state if commitment does not occur. The paper further embeds this in graph-based reaction systems, where the environment is a subgraph $2^N$7 and reaction enabling depends on the presence of perceptible species and the absence of inhibitor selectors (Maestri et al., 2021). The resulting framework is explicitly localist: molecules do not merely transform substrates; they perceive, conditionally commit, and act within a structured environment.

A thermodynamically grounded alternative appears in (Kocka et al., 21 Aug 2025). There the substrate is a circular complex of $2^N$8 binary monomers, each updated by enzymes specific to triplets $2^N$9. Under strong driving $W$0, each enzyme is either present with rate $W$1 or absent, giving $W$2 possible local rule sets, reduced to 88 dynamically distinct rules by cellular-automaton symmetries. The resulting dynamics form a continuous-time Markov process whose attractors include absorbing states, equilibrium multistate attractors, and nonequilibrium attractors with currents. The paper identifies rule 232 as a robust error-correcting memory rule, rule 170 as a two-attractor memory without spurious attractors, and rule 166 as a long-transient “molecular stopwatch” with $W$3 for $W$4 and exponential growth with $W$5. By externally switching enzyme sets, the same complexes can implement finite-state-machine behavior, including counters and order recorders (Kocka et al., 21 Aug 2025).

Taken together, these works show three levels of protein-centered molecular automata: explicit engineered state machines, perception-sensitive agents, and emergent asynchronous cellular automata derived from nonequilibrium enzymatic kinetics.

5. Charge-configurational molecular automata and molecular QCA

Quantum-dot cellular automata constitute the most hardware-oriented branch of molecular automata. In QCA, a cell is a small set of coupled quantum dots with a few mobile charges; binary information is stored in charge localization rather than current flow, and neighboring cells interact through Coulomb coupling. Arrays of such cells implement binary wires, inverters, and majority logic, while naturally mixing logic, memory, and interconnect (Blair, 2018). Molecular QCA is attractive because molecules can provide extremely small cells, with the literature emphasizing ultra-high density, potentially very high switching speeds, and room-temperature operation (Blair, 2018).

A central engineering problem is input transduction. “Electric-field Inputs for Molecular Quantum-dot Cellular Automata Circuits” (Blair, 2018) proposes writing bits into molecular QCA not by single-molecule electrodes or fixed-polarization molecules, but by lithographic electrodes that create an in-plane field over a field-sensitive input section. In the two-dot model, the field enters as a dipole detuning

$W$6

and the relevant interaction scale is the kink energy $W$7. For the geometry studied,

$W$8

A longitudinal two-cell array can be switched with

$W$9

whereas transverse arrays exhibit parity-dependent failures if directly exposed to the same field. The proposed architecture therefore uses a longitudinal field-sensitive selector followed by a transverse binary wire, and the preferred solution confines the write field to the input section with planar electrodes separated by

$$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$0

so that the uniform-field failure mode disappears (Blair, 2018).

“Robust Electric-field Input Circuits for Clocked Molecular Quantum-dot Cellular Automata” (Cong et al., 2021) extends this idea to synchronous, clocked molecular QCA based on three-dot cells. The cell polarization is

$$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$1

and the clocking field acts through a null-state bias $$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$2, while the input field again acts through $$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$3. The paper solves the full $$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$4-dimensional Hamiltonian for $$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$5 cells and shows that rotating the shift-register molecules by $$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$6 makes them immune to unwanted $$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$7 fringing because $$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$8. With the parameter choice

$$\mathrm{KIO_3 + AgNO_3 \rightarrow AgIO_3(s) + KNO_3}$$9

the kink energy is

$($0

The rotated design tolerates substantial $($1 contamination, with an approximate safe regime

$($2

for the functional output (Cong et al., 2021). This work shifts molecular QCA from abstract logic to robust interfacing under realistic fringing fields.

The internal switching physics of a single molecular QCA cell is analyzed in “Molecular reorganization energy in quantum-dot cellular automata switching” (Pidaparthi et al., 2021). There the mixed-valence molecule is modeled as a two-state system coupled semiclassically to nuclear relaxation. The total Hamiltonian includes the nonlinear term

$($3

and open-system dynamics are treated with a Lindblad equation. The central result is that reorganization energy $($4 produces hysteresis under bias reversal, enhances localization, and yields single-molecule memory, but at the cost of increased dissipation. The paper explicitly studies $($5, bias sweeps from $($6 to $($7, and shows that at zero bias the same molecule can hold $($8 or $($9 depending on sweep direction (Pidaparthi et al., 2021). This is molecular automata behavior in its minimal form: a two-state element with intrinsic history dependence.

Modeling these circuits is itself a many-body problem. “Variational Quantum Eigensolver Models of Molecular Quantum Dot Cellular Automata” (Gautam et al., 14 Oct 2025) notes that in the two-state approximation each QCA cell corresponds directly to a single qubit, so the circuit Hamiltonian is of Ising type with $\mathrm{NaOH}$0, $\mathrm{NaOH}$1, and $\mathrm{NaOH}$2 terms. The paper uses VQE on logic primitives including wires, inverters, and majority gates, on ideal simulators, noisy simulators, and IBM hardware. It reports strong agreement with exact diagonalization in noiseless cases, successful hardware modeling of a 15-cell wire with a three-parameter ansatz, and failure on a 30-cell wire due to noise; it also finds that RMSE improvements saturate beyond about 16k shots (Gautam et al., 14 Oct 2025). Molecular QCA is therefore both a candidate computational substrate and a benchmark for quantum simulation of molecular automata.

6. Spatially extended and reconfigurable molecular media

A different branch of molecular automata treats extended molecular assemblies as excitable or dynamically rewritable media. “Actin automata: Phenomenology and localizations” (Adamatzky et al., 2014) models filamentous actin as two coupled one-dimensional binary-state semi-totalistic automaton chains, with neighborhoods

$\mathrm{NaOH}$3

The full rule space has

$\mathrm{NaOH}$4

rules. The paper finds that 705 of 1024 rules support no localizations, about 31% support some localization, and only about 3% support rich dynamics of both traveling and stationary localizations. It also extracts characteristic transition templates: traveling-localization rules tend to excite resting nodes with 2 or 4 excited neighbors and keep excited nodes active with 0 excited neighbors, whereas stationary-localization rules tend to excite only at neighbor count 3 and keep excited nodes active when they have fewer than 4 excited neighbors (Adamatzky et al., 2014). The computational significance is collision-based: glider-like packets, breathers, still lifes, and glider-gun-like structures appear on a biomolecularly motivated topology.

“Massively parallel computing on an organic molecular layer” (Bandyopadhyay et al., 2011) moves from abstract automata to a reconfigurable physical substrate. The system is a bilayer of DDQ molecules on Au(111), where each molecule can occupy four conducting states $\mathrm{NaOH}$5, written by STM pulses and read by STM imaging. The top DDQ monolayer supports eight distinct molecular circuits, with effective local neighborhoods varying from 2 to 6 neighbors. The paper identifies seven categories of logic-state transport rules, including convergent attraction toward pseudo-positive charge, repulsive collisions, state-2 trails, group motion, and rule priority determined by local charge density and circuit composition (Bandyopadhyay et al., 2011). On this basis it demonstrates logic gates, density classification, Voronoi decomposition, diffusion-like behavior, and a cancer-mutation analogue. Unlike standard cellular automata, both neighborhood graph and dominant update rule are problem-dependent and can be reshaped by the encoded pattern itself.

“Turning Machines: a simple algorithmic model for molecular robotics” (Kostitsyna et al., 2020) gives a minimal chain-folding automaton on the triangular grid. Each monomer carries an integer turning number, and a local turn drags the entire suffix by a unit translation. Under the continuous-time Markov-chain execution model, almost-full line rotations of

$\mathrm{NaOH}$6

are possible, but a full

$\mathrm{NaOH}$7

rotation is impossible for $\mathrm{NaOH}$8. For $\mathrm{NaOH}$9, the line-rotation machine $)$0 reaches its target in expected time $)$1, and the same primitive yields exact folding of zig-zag paths and rastered squares, perimeter-bounded approximation of $)$2-monotone shapes, exact folding of some scaled separator-bearing shapes, and impossibility results for thin spirals (Kostitsyna et al., 2020). This is a molecular-automata model in which state is local, transitions are local and stochastic, but correctness is geometric and global.

A recurring misconception is that molecular automata are necessarily deterministic, digital, and already experimentally mature. The literature does not support that view. Determinism is often coarse-grained, as in dominant-state single-molecule DFA behavior (Zhang et al., 2024); some of the strongest proposals are still proof-of-principle theoretical constructions (Blair, 2018, Svahn et al., 2020); and physical feasibility is repeatedly limited by locality, fringing fields, dissipation, rate separation, noise, or synthesis complexity (Cho et al., 30 May 2026, Pidaparthi et al., 2021, Gautam et al., 14 Oct 2025). What the field does establish is narrower and more precise: molecular systems can realize automaton-like state spaces and transition rules in several physically distinct ways, and those realizations expose nontrivial tradeoffs among expressiveness, robustness, energetics, and fabrication. This suggests that the lasting significance of molecular automata lies not in any single substrate, but in the unification of automata theory with molecular-scale mechanisms of state retention, local update, and collective dynamics.