(614f) Data Science Shows That Entropy Correlates with Accelerated Zeolite Crystallization in Monte Carlo Simulation

Zeolites, with their intricate nanoporous structures, are vital materials for diverse applications spanning catalysis, separations, and storage. The increasing demand for tailor-made zeolites, particularly for sustainability goals like biofuel production and carbon capture, has intensified research into controlling zeolite formation. However, a fundamental understanding of pathways and rates of zeolite crystallization at the atomic level still need to be improved, hindering the development of efficient and targeted synthesis methods. Our group recently conducted a combined computational and experimental study to address this challenge, employing Monte Carlo (MC) simulations and zeolite synthesis experiments.[1] Our findings revealed that adding tetramethylammonium as a secondary organic structure-directing agent (OSDA) significantly accelerates the formation of all-silica LTA zeolite compared to using only one OSDA (1OSDA case). Because of the complexity of the simulation, understanding the physical cause behind the speedup remained challenging. The present study implemented data science methods to investigate the Monte Carlo structural datasets and to shed light on the origin of the speedup by including an additional OSDA (2OSDA case), concluding that entropy plays a critical role.

The Smooth Overlap of Atomic Positions (SOAP) [2] method was used to represent the evolving zeolite structures, encoding the local chemical environments within the system with 2-body and 3-body correlations. For segregating the two datasets, 1OSDA and 2OSDA, we employed various machine learning (ML) tools: unsupervised learning as Principal Component Analysis (PCA), supervised learning as Support Vector Machine (SVM) classification, and Principal Covariates Regression, which reduced dimensionality within the SVM regression exercise. In contrast to PCA, SVM successfully segregated the two datasets, even with only 2-body correlations, and showed that the 2OSDA system has a narrower SVM decision function distribution than that for 1OSDA. Building upon these observations, the pair entropy was computed using different conformations to test the idea of lower entropy for 2OSDA system zeolite formation. We found that the Si-Si conformation revealed the quantitative reproduction of speedup, 1.6-fold. The decease of silica network configurational entropy when adding a second OSDA facilitated crystallization by partially establishing the entropy loss that occurs as the system transitions from a disordered to an ordered state. We suggest that such an entropy effect can be produced in experiments by increasing pressure during synthesis. The SVM approach excelled at classifying structures and revealed kinetic differences between systems with and without the secondary OSDA. This difference prompted us to investigate physical entropy as a collective variable explaining these kinetic differences. From the analysis of the pair entropy differences, using transition state theory, exponentiating the most significant difference in computed entropy difference (ΔS = 0.46) had broad consistency with previously reported value in Monte Carlo simulation, exp(∆S/kB) ~ 1.6. We concluded by showing the reduced configurational entropy correlates with the acceleration of zeolite crystallization caused by additional OSDA.

References

[1] C. Bores, S. Luo, J. D. Lonergan et al. Phys. Chem. Chem. Phys. 24, 142–148 (2022).

[2] S. De, A. Bartók, G. Csányi et al. Phys. Chem. Chem. Phys. 18, 13754-13769 (2015).

[3] D. C. Wallace. J. Chem. Phys. 87 (4): 2282–2284 (1987).