Fig. 2: Performance of uMLIPs for cross-domain scenarios.
From: Optimizing cross-domain transfer for universal machine learning interatomic potentials
a MAE in predicting torsional energy barriers. The y-axis lists the uMLIP models. For multi-task uMLIPs, the inference channel is indicated to the right of each bar; for single-task uMLIPs, the corresponding training set is shown. White bullets mark the accuracy of the hybrid-functional channel (parentheses) in reference to the ωB97M-D3 results. b MAE of reaction energy predictions for organometallic complexes. c MAE of cohesive energy predictions for organic crystals. d MAE of formation energy predictions for hybrid organic-inorganic perovskites. e MAE of adsorption energy predictions for molecular inhibitors, computed by energy changes from physisorbed to chemisorbed states. f Error distributions of four benchmark tasks for metal--organic frameworks, represented as box plots. The box and red horizontal line in the box plot show the quartiles and median of the error distribution, respectively, while whiskers represent 1.5 times the inter-quartile range. Individual data points are randomly jittered along the horizontal axis for visual clarity. In (a–f), all reference DFT data are obtained with PBE-D3. Solid bars in (a–e) represent the best-performing channel or training database. Individual parity plots are provided in Supplementary Figs. 9–11,14,15. Source data are provided as a Source Data file.
