TY - JOUR

T1 - Machine learning-based predicting of the equilibrium cation distribution in faujasite-type zeolites

AU - Гренев, Иван Васильевич

AU - Бобков, Матвей Евгеньевич

AU - Иванов, Антон Дмитриевич

AU - Uliankina, Anastasiia I.

AU - Шубин, Александр Аркадьевич

N1 - The reported study was supported by the Russian Science Foundation (projects number 24-71-10096, https://rscf.ru/en/project/24-71-10096/). The Siberian Branch of the Russian Academy of Sciences (SB RAS) Siberian Supercomputer Center, as well as the Information and Computing Center of Novosibirsk State University are gratefully acknowledged for providing supercomputer facilities.

PY - 2025/2/1

Y1 - 2025/2/1

N2 - A surrogate predictive model for estimating the cationic site occupancy in faujasite-type (FAU) zeolites based on their chemical composition is presented. First, a reference experimental database containing structural data for 108 dehydrated FAU zeolites was compiled from literature sources. A quantitative criterion was introduced to assess the applicability of models predicting cationic site occupancy in reference zeolite frameworks. The canonical replica-exchange Monte Carlo method was used to generate a training data set. The best agreement between the simulated and experimental cationic site occupancies was achieved using a custom zeolite framework model with randomly distributed Al atoms and an extended DFT/CC-derived force field model. A set of descriptors was developed to translate the chemical composition of zeolite adsorbents into numerical parameters for modeling. The machine-learning surrogate model was trained on a simulation-derived database containing cation site occupancies for 250 monocationic and 3962 bicationic FAU zeolites generated using replica-exchange Monte Carlo simulations. Since surrogate predictive models are not limited by a set of force field parameters, they can be used to predict the cationic site occupancies in zeolites for a wide range of cations. The predictive ability of the surrogate model was demonstrated for a set of FAU zeolite frameworks from the reference database.

AB - A surrogate predictive model for estimating the cationic site occupancy in faujasite-type (FAU) zeolites based on their chemical composition is presented. First, a reference experimental database containing structural data for 108 dehydrated FAU zeolites was compiled from literature sources. A quantitative criterion was introduced to assess the applicability of models predicting cationic site occupancy in reference zeolite frameworks. The canonical replica-exchange Monte Carlo method was used to generate a training data set. The best agreement between the simulated and experimental cationic site occupancies was achieved using a custom zeolite framework model with randomly distributed Al atoms and an extended DFT/CC-derived force field model. A set of descriptors was developed to translate the chemical composition of zeolite adsorbents into numerical parameters for modeling. The machine-learning surrogate model was trained on a simulation-derived database containing cation site occupancies for 250 monocationic and 3962 bicationic FAU zeolites generated using replica-exchange Monte Carlo simulations. Since surrogate predictive models are not limited by a set of force field parameters, they can be used to predict the cationic site occupancies in zeolites for a wide range of cations. The predictive ability of the surrogate model was demonstrated for a set of FAU zeolite frameworks from the reference database.

KW - Zeolites

KW - FAU

KW - Cation distribution

KW - Database

KW - Replica-exchange Monte Carlo

KW - Force field

UR - https://www.scopus.com/pages/publications/105023592032

U2 - 10.1016/j.matchemphys.2025.131853

DO - 10.1016/j.matchemphys.2025.131853

M3 - Article

VL - 349

JO - Materials Chemistry and Physics

JF - Materials Chemistry and Physics

SN - 0254-0584

IS - Part 2

M1 - 131853

ER -