(169cb) Development of Chimes Machine Learned Interatomic Potentials for All Silica MFI Zeolites

The Haber-Bosch (HB) process has revolutionized agricultural output to match the ever-growing global food supply demand by enabling mass production of ammonia, a key precursor in production of synthetic fertilizers. The HB process works by combining nitrogen and hydrogen at high temperatures and pressures in the presence of catalysts, causing the entire process to be highly energy intensive. The process used to separate the ammonia from the gaseous reactive mixture is particularly energy intensive, requiring the gas to be cooled to 240 K to liquefy the ammonia. The current strategy for reducing energy requirements is two pronged: developing better catalysts to drive the reaction equilibrium and materials for enhanced separation efficiency.

Zeolites membranes are naturally occurring nanoporous structures of aluminosilicate compounds which show great promise as separation medium, exhibiting a near continuum of pore sizes that allow for separation of target gas molecules. Zeolite membranes have high thermal and chemical stabilities and show good selectivity and permeability for the separation of various mixtures. [1–4] Key factors controlling performance of the zeolite membrane are spatial variation in adsorption, diffusion pathways and pore channel length, all of which are affected by the atomic arrangement of the zeolite framework. [5] Experimental techniques have been used to probe these features with atomic level resolution to image zeolite atomic arrangements. [6,7] The understanding of the atomic arrangement can provide some key insights into the possible channel and diffusion dynamics. An extensive understanding of the molecular mechanisms affecting adsorption, diffusion and selectivity and dynamic behavior of the framework are hard to approach using current experimental techniques. [5–7] Here, we present a new machine learned interatomic model for bulk MFI zeolite using the ChIMES physics-informed machine learned interatomic model and development framework. [8–14] We will show that bulk structural properties predicted with our model such as radial distribution functions and vibrational power spectra compare well with similar properties derived from DFT simulations. Our newly developed ChIMES machine learned interatomic potential for all-silica MFI zeolite can be used to provide useful insight into the role of zeolite as a separation media for Haber-Bosch process. Our model can be used to simulate the zeolite pore channels and compare the predicted structural configurations with observed experimental structures derived from imaging techniques such as ptychography to better inform and predict atomic level framework dynamics and performance characteristics. [5] We will discuss a possible strategies to extend our MFI zeolite model to other all-silica zeolite structures. Ultimately, our newly developed ChIMES machine learned IAP for all-silica MFI zeolite can be used to provide useful insight into the role of zeolite as a separation media for Haber-Bosch process and to study the underlying interactions in zeolite-mediated catalytic processes.

References

[1] Hisao Inami et al. Membranes 11.5 (2021)

[2] Miguel Palomino et al. Langmuir 26.3 (2010)

[3] Jin Shang et al. Chem. Commun. 50 (35 2014)

[4] Jin Shang et al. J. Am. Chem. Soc. 134.46 (2012)

[5] Hui Zhang et al. Science 380.6645 (2023), pp. 633–638.

[6] Zhuoya Dong et al. J. Am. Chem. Soc. 145.12 (2023).

[7] Haozhi Sha et al. Science Advances 9.11 (2023)

[8] Rebecca K. Lindsey et al. J. Chem. Theory Comput. 13.12 (Dec. 2017).

[9] Rebecca K. Lindsey et al. The Journal of Chemical Physics 153.13 (2020)

[10] Rebecca K. Lindsey et al. J. Chem. Theory Comput. 15.1 (2019)

[12] ChIMES Calculator. URL: https://github.com/rk-lindsey/chimes_calculator.

[13] ChIMES Generator. URL: https://github.com/rk-lindsey/chimes_lsq.

[14] ChIMES Active Learning Driver. URL: https://github.com/rk-lindsey/al_driver.