Non-normality and non-monotonic dynamics in complex reaction networks

TL;DR

This paper investigates how non-normality influences non-monotonic dynamics in complex reaction networks, providing a general condition for such behavior and demonstrating its role in entropy dynamics through a hydrogen combustion example.

Contribution

It establishes that non-normality promotes non-monotonicity but is not necessary for it, and develops a rigorous theory applicable to large-scale chemical reaction networks.

Findings

Non-normality promotes non-monotonic dynamics.

Non-normality is required for non-monotonic entropy behavior.

Large-scale simulations of reaction networks are feasible with the developed theory.

Abstract

Complex chemical reaction networks, which underlie many industrial and biological processes, often exhibit non-monotonic changes in chemical species concentrations, typically described using nonlinear models. Such non-monotonic dynamics are in principle possible even in linear models if the matrices defining the models are non-normal, as characterized by a necessarily non-orthogonal set of eigenvectors. However, the extent to which non-normality is responsible for non-monotonic behavior remains an open question. Here, using a master equation to model the reaction dynamics, we derive a general condition for observing non-monotonic dynamics of individual species, establishing that non-normality promotes non-monotonicity but is not a requirement for it. In contrast, we show that non-normality is a requirement for non-monotonic dynamics to be observed in the R\'enyi entropy. Using hydrogen combustion as an example application, we demonstrate that non-monotonic dynamics under experimental conditions are supported by a linear chain of connected components, in contrast with the dominance of a single giant component observed in typical random reaction networks. The exact linearity of the master equation enables development of rigorous theory and simulations for dynamical networks of unprecedented size (approaching $10^5$ dynamical variables, even for a network of only 20 reactions and involving less than 100 atoms). Our conclusions are expected to hold for other combustion processes, and the general theory we develop is applicable to all chemical reaction networks, including biological ones.