Integrating Solution Physical Properties into Zeolite Synthesis Prediction via Causal Machine Learning.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Zeolites are critical porous crystalline materials widely used in catalysis and molecular separations.
• Targeted synthesis of specific zeolite frameworks is complex, influenced by multiple interconnected factors including solution chemistry, structure-directing agents, solvents, and crystallization conditions.

Purpose of the Study:

• To develop a predictive model for zeolite crystallization outcomes by integrating synthesis parameters with solution physical properties.
• To enhance the accuracy and efficiency of targeted zeolite synthesis through a data-driven approach.

Main Methods:

• A data-driven framework was developed, combining state-of-the-art synthesis parameters with physical properties of the solution obtained from molecular simulations.
• The model was trained and validated on 366 literature-reported zeolite syntheses.

Main Results:

• The model achieved 96.4% accuracy in classifying 20 different zeolite frameworks.
• It demonstrated 87.7% accuracy in distinguishing between four structural aperture classes, surpassing composition-only prediction methods.
• Analysis revealed that solution density and dielectric constant independently influence zeolite formation, with opposing effects on small- and large-aperture structures.

Conclusions:

• Zeolite framework selection is determined by both chemical composition and emergent solution properties.
• The developed physically informed strategy offers a novel approach for predictive zeolite synthesis, advancing materials design for catalysis and separations.