(389h) Data Science Shows That Entropy Accelerates Zeolite Crystallization

Zeolites, nanoporous aluminosilicate materials, offer high surface area, uniform pore sizes, acidity, and ion exchange capacity, making them invaluable in catalysis, separations, and storage. Meeting sustainability goals requires tailor-made zeolites, yet their precise synthesis remains challenging due to the complex molecular dynamics governing crystallization. Towards this end, we recently reported both Monte Caro(MC) simulations and zeolite synthesis experiments showed using tetramethylammonium (TMA) as a secondary organic structure-directing agent (OSDA) can speed up the formation of all-silica LTA zeolite. However, understanding the underlying mechanisms has been hindered by computational complexity. However, the fundamental cause of the speed-up in the MC simulations remained poorly understood because of the complexity of the simulations. Our study applied data science methods to investigate the cause of the speed-up by including the secondary OSDA, finding that entropy plays a critical role.

The Smooth Overlap of Atomic Positions (SOAP) was used to represent structures from the MC simulation snapshots, thereby encoding local environments with translational-, rotational-, and exchange-invariance. Machine learning (ML) tools, including Principal Component Analysis (PCA) and linear Support Vector Machine (SVM), were deployed to discern systems with and without TMA. While PCA failed to distinguish nuances, supervised SVM classification proved highly effective. Principal Covariates Regression (PCovR) further streamlined analysis. SVM decision functions reveal that the TMA-inclusive systems presented narrow histograms, suggesting a lower entropy during zeolite formation. Inspired by this finding, the pair Si-O and Si-Si entropies were computed during the process of zeolite formation, finding that the Si-Si pair entropy quantitatively reproduced the apparent crystallization speed-up of 1.6. Adding a secondary agent within a fixed volume decreased configurational entropy in simulations, reducing the crystallization barrier associated with entropy loss upon ordering. Exploring entropy's role in crystallization rates under fixed volume conditions could be facilitated by utilizing pressure as a control variable in synthesis.