Traversing Dense Networks of Elementary Chemical Reactions to Predict Minimum‐Energy Reaction Mechanisms

Abstract

Numerous different algorithms have been developed over the last few years which are capable of generating large, dense chemical reaction networks describing the inherent chemical reactivity of a collection of discrete molecules. For all elementary reactions in a given reaction network, reaction rate calculations, followed by direct micro-kinetic modelling, enables one to predict macroscopic outcomes (e. g. rate laws, product selectivity) based on atomistic input data. However, for chemical reaction networks containing thousands of reactant molecules, such simulations can be extremely time-consuming; in addition, the complex coupled time-dependence of molecular concentrations can present challenges when seeking essential mechanistic features. In this Article, we instead present an algorithm which seeks to predict the "most likely" reaction mechanism, or competing mechanisms, connecting any two user-selected reactant and product species, given a previouslygenerated reaction network as input. The approach is successfully tested for reaction networks (containing tens of thousands of possible reactions) describing the carbon monoxide oxidation on platinum nanoparticles.

A Chemically Consistent Graph Architecture for Massive Reaction Networks Applied to Solid-Electrolyte Interphase Formation

“…However, as described above, these algo-rithms will not necessarily produce chemically valid pathways in networks containing multi-reactant reactions. It has historically been necessary to use custom methods that have suboptimal performance and scale poorly with network size based on tree traversal algorithms 22 , breadth-first-search 25 , or depth-firstsearch 11,25 . Performant stochastic sampling approaches can instead be employed, 30 but they are not guaranteed to find the true best solution(s).…”

“…Instead, researchers employ custom approaches that can calculate path costs while respecting reaction stoichiometry. 25 These algorithms suffer from performance and scaling limitations 11,25 that have historically made it necessary to significantly restrict network size. As a result, reaction networks have not previously been applied to complex electrochemical or photochemical systems which may include thousands of species and millions of reactions.…”

A chemically consistent graph architecture for massive reaction networks applied to solid-electrolyte interphase formation

“…Instead, researchers employ custom approaches that can calculate path costs while respecting reaction stoichiometry. 25 These algorithms suffer from performance and scaling limitations 11,25 that have historically made it necessary to significantly restrict network size. As a result, reaction networks have not previously been applied to complex electrochemical or photochemical systems which may include thousands of species and millions of reactions.…”

Fast and Automated Identification of Reactions with Low Barriers: The Decomposition of 3-Hydroperoxypropanal

“…However, several very promising methodologies have been developed to automate the discovery of reaction mechanisms using QM. [1,2,3,4,5,6,7,8,9,10,11] Following Dewyer et al [11] these methods can be categorized into four categories: 1) encoding mechanistic steps, 2) generate transition state (TS) guesses, 3) generate intermediates using graph theory, and 4) generate driving coordinates. Subsequently, Grimme introduced a meta-dynamics approach [9] that arguably represents a fifth category and Lavigne et al [10] have recently combined this approach with a driving coordinate approach.…”