Fig. 5: Schematic illustration and prediction performances of the equivGNN model.

a Overview of the equivGNN architecture, which integrates atom embedding, equivariant message-passing, and readout blocks. Parity plots of the DFT-calculated versus ML-predicted binding energies derived from the combined validation set under 5-fold cross-validation (CV) via the equivGNN model on (b) Simple Dataset33, c Complex Dataset33, and (d) HEA Dataset37. e The prediction performance of the equivGNN on B8-cluster Dataset10 with a fixed training set. The samples, mean absolute errors (MAE) and root mean square errors (RMSE) are provided in parity plots from the equivGNN; the violin plots in the inset show the absolute error distributions, and the inner dash line represents the median (unit: eV); the inset images are simplified diagrams for catalytic systems with various degrees of complexity. DFT density functional theory, ML machine learning.