Open chemical reaction networks that exchange matter with reservoirs can display irregular concentration dynamics on macroscopic scales. The standard description in terms of deterministic rate equations is capable of generating chaos, as observed in experiments on the Belousov–Zhabotinsky reaction. At the molecular level, however, reactions proceed as discrete random events, so a reaction network of finite volume is fundamentally described by the chemical master equation, equivalent to a Markov jump process for particle numbers. This stochastic description remains applicable even when the macroscopic rate equation is chaotic, and it has been used to study intrinsic fluctuations in such systems. The standard deterministic notion of chaos does not apply directly to such a process, so the connection between macroscopic chemical chaos and the stochastic description is analogous in spirit to deriving thermodynamic behavior from molecular statistical mechanics. The work stands somewhat apart: nearby in the archive there are studies of quantum mixing, quantum circuits, disorder, and chaos in the quantum world, but there are almost no direct neighbors on the topic of chemical chaos and information theory.
A chemical reaction network of three species and seven reaction channels is considered, a modified version of the CS14 model proposed by Plesa and Sprott. For a finite volume V, the dynamics is described by the chemical master equation, and in the limit V→∞ it turns into deterministic rate equations that, for the chosen parameters, exhibit chaos with a positive leading Lyapunov exponent of 0.065. The key quantity is the mutual information between the concentration at time t and the initial concentration, where the initial state is drawn from the stationary distribution. The authors use a decomposition of the mutual information in terms of entropies and apply a system-size expansion: up to a crossover time that grows with volume, the conditional distribution of the concentration at a fixed initial condition is well approximated by a Gaussian whose covariance evolves along the linearized deterministic flow. Numerically, the mutual information is estimated by the k-nearest-neighbor method with k=10, using samples from 50 initial concentrations and $10^6$ trajectories for each.
It is shown that the rate of decay of mutual information, defined as −dI($X_t$; $X_0$)/dt, converges in the large-volume limit to the Kolmogorov–Sinai entropy of the deterministic model, equal to the sum of the positive Lyapunov exponents (here 0.065). The mechanism is transparent in the system-size expansion: stationarity makes it possible to express the loss of mutual information as the growth of the conditional Shannon entropy. Up to the crossover time $t_c$(V), which diverges as V→∞, the conditional fluctuations are locally Gaussian, their covariance evolves under the linearized flow, its width grows exponentially along unstable directions and remains bounded along stable ones. Therefore the growth rate of the conditional entropy equals the sum of the positive Lyapunov exponents, and Pesin's equality identifies this sum with the Kolmogorov–Sinai entropy. Numerical simulations for V=$10^2$,...,$10^6$ confirm this: at small volumes, mutual information drops rapidly because of strong reaction fluctuations, and as V grows an intermediate regime of approximately linear decay appears, with a slope that tends to 0.065. The relative information loss ΔI(t;$t_{ref}$) with $t_{ref}$=5 follows the Gaussian approximation for longer and longer as V increases, and the finite-time rate of Gaussian information loss approaches $h_{KS}$.
The work provides the first explicit information-theoretic bridge between the stochastic master equation and deterministic chaos. Unlike neighboring studies that consider quantum mixing, quantum circuits, or chaos in the quantum world, the object here is a classical Markov jump process with noise arising from finite volume. The method requires no external auxiliary bath or measurement apparatus, as in quantum constructions of the Kolmogorov–Sinai entropy, and instead relies on the two-time distribution of an autonomous process. This complements existing approaches and offers a new tool for analyzing chaos in stochastic systems with a deterministic limit.
The formulation can be extended to general stochastic systems with a deterministic limit, for example chaotic systems with weak noise. One can also consider probabilistic cellular automata and lattice gas models, where discreteness persists even in the deterministic limit. It would be interesting to measure the rate of mutual information loss in such systems in order to characterize chaotic behavior. Another direction is quantum chaos: the connection to the quantum Kolmogorov–Sinai entropy, defined via a weak auxiliary bath or weak measurements, remains unresolved.
The result may influence the analysis of stochastic models in chemical kinetics and related fields, where it is important to distinguish deterministic chaos from noise. It also offers a new way to quantify chaos in systems with intrinsic fluctuations.
The next step is to test the convergence for other chaotic chemical systems and for stochastic systems with non-Gaussian fluctuations. It is also worth investigating the behavior of the information loss rate near bifurcation points and for systems with multiple attractors.
In statistical mechanics, the question of how thermodynamic behavior emerges from molecular dynamics remains open. This work links an information measure of a stochastic process to deterministic chaotic dynamics, shedding light on an analogous problem for nonequilibrium chemical systems. What remains unsolved is the task of rigorously proving the convergence of finite-volume stationary measures to the invariant measure of the deterministic system in the general case.
This section is addressed to the author of the paper and their colleagues. Neighbouring works were found by meaning-based search across our archive, not by word overlap. We do not review: we show the field around the work.
What the work does
The paper shows how the rate of mutual information loss between the initial and current states in a chemical master equation converges to the Kolmogorov-Sinai entropy of the deterministic model in the large-volume limit. A system-size expansion is used to link stochastic fluctuations to the linearized flow, along with numerical simulation of a Markov jump process for a three-component system with seven reactions.
Where its strength is
The strength of the approach lies in the explicit connection between the finite-volume stochastic description and deterministic chaos through information loss, which opens a path to interpreting chaos in terms of statistical mechanics.
Where it could go next
Worth trying
Why this may matter
Such a connection could provide a unified view of chaos in stochastic and deterministic systems, which matters for understanding fluctuations in chemical and biological networks.
Nearby works in our archive
The archive is sparse around this work: the closest review is only 0.54, against a usual 0.64. It looks like few have drilled nearby.
The Kolmogorov–Sinai entropy for the modified CS14 model is 0.065, which corresponds to an uncertainty doubling time of about 10.7 time units. At volume V=$10^6$, the crossover time after which the Gaussian approximation breaks down is estimated as $t_c$ ≈ (1/0.065) ln($10^6$) ≈ 212 time units.