(159f) Expert System for Automated Reaction Mechanism Generation

Complex reaction networks are present in numerous chemical and biochemical systems, such as combustion, pyrolysis, catalytic conversion of hydrocarbons, and cell metabolism. These systems contain a large number of species and reactions, easily generated by automated network generators¹ with initial reactants and reaction rules as inputs. Reaction rule specification plays a significant role in the reaction network generation affecting the size as well as the key species. These reaction rules are typically written heuristically based on a user’s knowledge about the reaction system resulting in either missing chemistry or including elementary steps that do not really happen. Furthermore, we have identified limitations in the representation of the adsorbate conformation in the published literature where the adsorption mode (e.g., σ vs. η) as well as the binding sites (top, bridge, and hollow sites) are not considered.^2,3 The reaction network generated, therefore, lacks information regarding conformations and catalytic site. Hence, the motivation of generating reaction rules from first-principles using Density Functional Theory (DFT) data.

Due to differences between ab initio- and heuristically written-published reaction mechanisms, we propose a first-principles based reaction mechanism generation framework using ab-initio data from DFT and published reaction mechanisms. The framework takes in as input a list of proposed reactions and species structure from DFT files. The code then generates reactant and product molecular graphs using the atomic coordinates of the species. The molecular graphs describe the connectivity of the atoms within the molecule and the catalyst. These molecular graphs are then matched to identify a common molecular graph; any edge that is present in the reactant graph and absent in the common graph is specified as a bond breaking step, while any edge that is present in the product graph and absent in the common graph is specified as a bond forming step. The reaction rules specifying the reactive center in the reactants and the transformation rules are then generated. These reaction rules are then imported into an automated network generator. The proposed framework can be used to generate reaction networks for systems that have not been studied and also for testing the accuracy of published mechanisms. Multiple reaction systems with varying carbon chain length are considered to show the efficacy of the proposed framework.

References

1. Rangarajan, S., Bhan, A., and Daoutidis, P. Language-oriented rule-based reaction network generation and analysis: Description of RING. Chem. Eng. 45, 114 (2012)
2. Rangarajan, S., Brydon, R., Bhan, A., and Daoutidis, P. Automated identification of energetically feasible mechanisms of complex reaction networks in heterogeneous catalysis: Application to glycerol conversion on transition metals. Green Chem. 16, 813 (2014)
3. Margraf, J. T., and Reuter, K. Systematic Enumeration of Elementary Reaction Steps in Surface Catalysis. ACS Omega. 4, 3370 (2019)