A Machine-Learning-Based Investigation on the Formation and Evolution of Silicic Acid Oligomers Structurally Analogous to Zeolite Building Units.

A machine-learning (ML) based model that is capable of predicting the formation free energy of a silicic acid oligomer (OSA) with its SMILES string was developed and was used to investigate the formation and evolution of OSAs structurally analogous to zeolite secondary building units and composite building units (S/CBU-OSAs) to understand the mechanistic pathways for the integration of OSAs into zeolites. It was shown that most intermolecular condensation (IEC) and intramolecular condensation (IAC) forming the first cycle in OSAs are exergonic and are the major pathways for OSAs evolution. The exergonicity of an IAC depends strongly on the degree of condensation and distortion. The formation of an S/CBU-OSA is more exergonic than the competing formation of a non-S/CBU-OSA, and the exergonicity would be even greater if it were from a non-S/CBU-OSA. The work highlights the feasibility of using ML-based models to predict reaction thermodynamics and pinpoints the superior thermodynamic stability of S/CBU-OSAs.