Zeolites are widely used as adsorbents and membranes due to their rich topological diversity, particularly for ion adsorption and exchange, such as removing heavy metals and extracting potassium salts. However, the discovery of zeolite-like structures with high selectivity for ${\text{K}}^{+}$ is high cost and time consuming due to the complexity of the ion-exchange process. Here, we propose a data-driven framework that combines database construction, unsupervised learning, generative models, and Density Functional Theory (DFT) calculations to identify porous structures with high ${\text{K}}^{+}$ selectivity. DFT calculations provide insights into the physical origins of the ${\text{K}}^{+}$ selectivity. Leveraging the established zeolite-like structure database, unsupervised learning enables us to rapidly identify candidates with high ${\text{K}}^{+}$ selectivity from thousands of structures with aluminum sites generated by diffusion models. We further discover a porous structure with the highest capacity for ${\text{K}}^{+}$ to date via DFT calculations. Our approach introduces a novel computational paradigm for accelerating the discovery of advanced materials, paving the way for innovative design of function-led structures.

Close

PrevNext

View figure in article

• Adsorption
• Atomic & molecular collisions
• Structural properties
• 3-dimensional systems
• Complex materials
• Crystal structures
• Porous materials
• Atomic Properties
• Crystal phenomena
• Crystallography
• Density functional theory
• First-principles calculations
• Neuroscience, neural computation & artificial intelligence
• Schroedinger equation