Molecular dynamics and machine learning insights into the mechanical behavior of zeolites under large deformation

The ongoing research on zeolite mechanical properties is often focused solely on the elastic response, which prevents a complete understanding of their behavior in various pore configurations under large deformation. In this study, reactive-force-field-based molecular dynamics is employed in combination with machine learning to investigate the relationship between zeolite structures and their mechanical properties across both elastic and plastic regimes. We performed uniaxial compression tests on zeolites with diverse structures, thus generating 2710 data points. A subsequent statistical analysis confirmed the correlations between the mechanical properties-such as energy absorption, compressive stress, and stress-strain curve linearity-and structural features, such as density and bond angle. The analysis showed that both global features (e.g., density and volume) and local features (e.g., the Si-O-Si angle) are key factors governing energy absorption. Furthermore, we screened 39,589 potentially isotropic zeolites from a total of 590,448 hypothetical structures and developed a machine learning model to predict their energy absorption capabilities. The proposed model identified 50 zeolites each with the highest and lowest energy predictions. These identified zeolites were rigorously validated through molecular dynamics simulations, thereby confirming the validity of the proposed model. The identified structure exhibited a 20 % greater energy absorption compared to the maximum value in the pre-existing database of 2710 zeolites. Current findings revealed the complex relationship between the zeolite structure and mechanical properties in the plastic regime, thus introducing new avenues for advancing zeolite design and possible applications through integrated atomistic simulation and machine learning.