Algorithmic Explorations of Unimolecular and Bimolecular Reaction Spaces

Abstract

Algorithmic reaction exploration based on transition state searches has already made inroads into many niche applications, but its potential as a general purpose tool is still largely unrealized. Ongoing progress in developing inexpensive model chemistries and expedited transition state searches is bringing this closer to fruition, but a remaining gap is the absence of benchmark problems involving larger molecules with diverse reactivity that would allow even-handed comparisons amongst various methods. To address this we have performed an algorithmic exploration of several reaction networks and bimolecular reactants, comprising a total of 581 reactions involving 51 distinct reactants. A mixture of previously studied and new systems were selected to facilitate comparisons across chemical space and between reaction exploration algorithms. Using a contemporary reaction exploration algorithm, we observe a much richer reactivity landscape than has been previously reported for several of these systems, including lower barrier reaction pathways and a strong dependence of reaction conformation in the apparent barrier of the reported reactions. Several of these benchmarks exhibit a large number of reacting sites that are handled by the algorithm with modest computational costs using widely available computational resources. Given the generality of these benchmarks, these results illustrate that reaction exploration algorithms are approaching general purpose capability.

Perspective on Automated Predictive Kinetics using Estimates derived from Large Datasets

“…[37,38,39,61,63,66,68,69,70,71,72,83] In recent years, it has become possible to use the computer to automatically discover transition state geometries [30,74,75,76] and so to compute reaction barriers for large numbers of reactions. [60,73,77]. With these large data sets, one can use machine-learning methods to build models for rapidly estimating the barriers for other reactions.…”