Instead, state-of-the-art reaction networks are generally built by hand, based on chemical intuition and (sparse) experimental evidence.
#MICRO FOCUS OO FULL#
Indeed, not even the topology of the full network is usually taken into account. This information is typically not available. One would need at least an approximate notion of the reaction thermochemistry (and ideally the kinetics) of the full network, to be able to do this on a rational basis.
Notwithstanding, the difficulty lies in knowing which parts of the full network to keep. This offers a non-empirical route to understanding complex reaction mechanisms. Meanwhile, the big advantage of focusing on sub-graphs is that the kinetics and thermochemistry of each elementary step may be explicitly computed from first principles. It is therefore entirely possible that a microkinetic model based on a reduced reaction network correctly describes the overall kinetics of a complex process 1, 6, 20. This is not automatically a problem, as large parts of the latter may not be thermodynamically accessible. The reaction networks typically used in microkinetic studies of natural and industrial processes are therefore necessarily merely sub-graphs of the full network of possible reactions (see Fig. For example, we recently reported a database of over 1 million elementary reactions for molecules no larger than four non-hydrogen atoms containing carbon, oxygen and hydrogen 22. In many cases, however, the understanding of complex chemical processes is hampered by the sheer size of the networks in question 1, 13, 14, 15, 16, 17, 18, 19, 20, 21. Indeed, any study of chemical kinetics or selectivity is essentially a study of a reaction network. Reaction networks are essential tools for the description, illustration, and fundamental understanding of chemical processes in such diverse fields as catalysis 1, 2, 3, 4, combustion 5, 6, 7, polymerization 8, atmospheric chemistry 9, systems chemistry 10, 11, and the origin of life 12. Showcased by the application to methane combustion, we demonstrate that the learned reaction energies offer a non-empirical route to rationally extract reduced reaction networks for detailed microkinetic analyses.
We show that the special topology of reaction spaces, with central hub molecules involved in multiple reactions, requires a modification of existing compound space ML-concepts. As an important basis for ‘reactive’ ML, we establish a first-principles database (Rad-6) containing closed and open-shell organic molecules, along with an associated database of chemical reaction energies (Rad-6-RE). Central to chemistry as a science of transformations, this space contains all possible chemical reactions. Here, we therefore engage in the ML-driven study of even larger reaction space. While intractable as a whole, modern machine learning (ML) is increasingly capable of accurately predicting molecular properties in important subsets. Chemical compound space refers to the vast set of all possible chemical compounds, estimated to contain 10 60 molecules.