Fig. 7: Effective configuration identification during active learning finetuning in large scale simulations for friction target space.

a Friction evolution with time of self-mated a-C:H with 40 at.%H. b Snapshots of sliding interfacial atomic structure of self-mated a-C:H with 40 at.%H at different timepoints during the friction and the corresponding distribution of atomic force error. The purple and gray atoms represented the carbon atoms and the yellow and light blue atoms represented the hydrogen atoms in the snapshots. c Local high error region extraction and labeling during active learning finetuning. Take the snapshots of sliding interfacial atomic structure at 400 ps as an example. The high error local structures were extracted from large friction system with the high error center fixed and the surrounding atoms were fully relaxed in DFT calculation.