Graph\nNeural Networks for Carbon Dioxide Adsorption\nPrediction in Aluminum-Substituted Zeolites

The ability to efficiently\npredict adsorption properties of zeolites\ncan be of large benefit in accelerating the design process of novel\nmaterials. The existing configuration space for these materials is\nwide, while existing molecular simulation methods are computationally\nexpensive. In this work, we propose a model which is 4 to 5 orders\nof magnitude faster at adsorption properties compared to molecular\nsimulations. To validate the model, we generated data sets containing\nvarious aluminum configurations for the MOR, MFI, RHO and ITW zeolites\nalong with their heat of adsorptions and Henry coefficients for CO<sub>2</sub>, obtained from Monte Carlo simulations. The predictions obtained\nfrom the Machine Learning model are in agreement with the values obtained\nfrom the Monte Carlo simulations, confirming that the model can be\nused for property prediction. Furthermore, we show that the model\ncan be used for identifying adsorption sites. Finally, we evaluate\nthe capability of our model for generating novel zeolite configurations\nby using it in combination with a genetic algorithm.