Fig. 3: Performance of different classical potentials and ML methods in elastic constants prediction. | npj Computational Materials

Fig. 3: Performance of different classical potentials and ML methods in elastic constants prediction.

From: Interpretable ensemble learning for materials property prediction with classical interatomic potentials

Fig. 3

a MAEs of the total elastic constants relative to DFT reference under different methods, Min represents the best performance for all structures by using these nine classical potentials. AB and XGB have better or similar performance than the Min, and all ensemble models are better than the Tersoff. b Predicted elastic constants under Tersoff, RF, and AB versus DFT reference, both RF and AB models have lower MAEs than before. The points in the green circles have large errors, and all these points come from the complicated structures which are out of training sets in both ensemble models.

Back to article page