Finely tuned inverse design of metal-organic frameworks with user-desired Xe/Kr selectivity

Abstract

Inverse materials design entails providing desired properties as inputs and obtaining fine-tuned materials that fit the given criteria as outputs. Although this workflow would in principle lead to significant efficiency in materials design, it is difficult in practice to successfully implement a robust, accurate inverse design platform. In this work, we used a validated platform which integrates a genetic algorithm with machine learning to design user-desired metal-organic frameworks (MOFs) with the xenon/krypton separation being presented as a case study. Using our platform, we obtained two record-breaking MOFs that show significant improvement over the current record. Moreover, with facile modification in the cost function, we demonstrate that our platform can generate MOFs that are finely tuned to the specific desires of users across multiple properties and a range of property values.