Evaluating Universal Machine Learning Force Fields Against Experimental Measurements

The MinX dataset has been meticulously curated and handcrafted from different database to serve as a reliable benchmark data for validating UMLFFs against experimental mineral structures. The main goal of releasing this data is to reduce the complexity and inconsistencies often encountered with real-world minerals. By providing a experimentally validated structures, we aim to facilitate consistent and reproducible benchmarking across the community.