Graph Neural Networks for Carbon Dioxide Adsorption Prediction in Aluminum-Substituted Zeolites

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Predicting zeolite adsorption properties is crucial for designing new materials but is hindered by vast configuration spaces and computationally expensive molecular simulations.
• Existing methods for evaluating zeolite performance, such as Monte Carlo simulations, are time-consuming, limiting rapid material discovery.
• The development of faster, accurate methods for predicting adsorption properties is essential for accelerating the design of novel zeolites.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for the rapid and efficient prediction of zeolite adsorption properties.
• To significantly reduce the computational cost associated with predicting adsorption characteristics compared to traditional molecular simulation techniques.
• To demonstrate the model's utility in identifying adsorption sites and generating novel zeolite configurations.

Main Methods:

• Generation of datasets for MOR, MFI, RHO, and ITW zeolites, including various aluminum configurations.
• Calculation of heats of adsorption and Henry coefficients for CO2 using Monte Carlo simulations for dataset creation.
• Development and application of a machine learning model trained on simulated data for property prediction and site identification.

Main Results:

• The proposed machine learning model achieves prediction speeds 4 to 5 orders of magnitude faster than molecular simulations.
• Model predictions for adsorption properties show strong agreement with values obtained from Monte Carlo simulations.
• The model successfully identified adsorption sites within the zeolite structures and demonstrated capability in generating novel zeolite configurations when combined with a genetic algorithm.

Conclusions:

• The developed machine learning model offers a highly efficient and accurate approach for predicting zeolite adsorption properties.
• This computational tool can accelerate the discovery and design of new materials with desired adsorption characteristics.
• The model's ability to identify adsorption sites and generate novel configurations highlights its potential in materials informatics.