AI and the World's Hard Problems · Chemistry & materials · arXiv · 2026-10-21
MOFFieldBench: A Machine Learning Benchmark for Continuous Electronic Fields in Metal-Organic Frameworks
Machine-learning models for metal-organic frameworks (MOFs) mostly rely on atomic structures, while electron-density and electrostatic-potential fields remain underexplored because first-principles calculations are costly and systematically aligned datasets are scarce. MOFFieldBench is a multimodal benchmark of about 46,000 MOFs with one-to-one aligned crystallographic structures, ED fields and ESP fields, together with a unified workflow covering field generation, error-controlled compression and sparse point-cloud representation for tasks such as field reconstruction, prediction of 13 material properties, multimodal prediction and cross-modal retrieval. ED and ESP fields prove competitive for pore geometry, adsorption and separation properties, multimodal fusion improves several prediction tasks over the best unimodal reference, and initializing self-consistent-field calculations from predicted densities cuts computation time by 13.59% on average in paired tests.