Structure-Mechanical Stability Relations of Metal-Organic Frameworks via Machine Learning

Abstract

Development of new materials via experiments alone is costly and can take years, if not decades, to complete. Advancements in the predictive power of computer simulations have enhanced our ability to design and develop materials in a fraction of the time required for experiments. Here, we demonstrate how the power of machine learning, trained by a combination of multi-level simulations, can predict the performance of metal-organic frameworks (MOFs), one of the most exciting advances of porous materials science. The machine-learning algorithm introduced here predicts the mechanical properties of existing and future MOFs in the order of seconds, allowing the design of robust structures. The principles of our computational approach can be translated to other problems so that MOF researchers can discover new materials for application in, e.g., catalysis, energy storage, and chemicals separation. We anticipate that our work will guide future efforts to make stable MOFs suitable for industry.