Elucidating Thermodynamically Driven Structure-Property Relations for Zeolite Adsorption Using Neural Networks.

Area of Science:

• Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Designing zeolites for catalysis, gas storage, and separations requires understanding molecular confinement.
• Previous studies proposed linear correlations for adsorption thermodynamics, but their broad applicability was unproven.

Purpose of the Study:

• To investigate the generalizability of linear adsorption models across diverse molecules and zeolites.
• To develop predictive models for adsorption thermodynamics in confined zeolite systems.
• To identify key molecular and zeolite structural features influencing adsorption.

Main Methods:

• Conducted >3500 molecular simulations of adsorbate-zeolite combinations.
• Developed and validated nonlinear predictive models, specifically bootstrapped neural networks.
• Utilized SHAP analysis to determine feature importance for adsorption property prediction.

Main Results:

• Linear adsorption trends were found to collapse in highly confined zeolite systems.
• No universal linear models were identified to predict adsorption properties from molecular and zeolite structures.
• Nonlinear models accurately predicted entropy of adsorption, isosteric heat, and Henry's constant.
• Framework features, particularly pore diameter, were more critical for predicting entropy of adsorption than adsorbate features.

Conclusions:

• Linear models are insufficient for predicting adsorption in confined zeolites.
• Machine learning models using geometric and physical descriptors offer accurate predictions of adsorption thermodynamics.
• Zeolite pore structure significantly influences adsorption entropy, while molecular size is key for Henry's constant.