Figure 2. Performance comparison of uMLIPs on the GB-56 grain boundary dataset. (a, b) Energy and force RMSE (root-mean-square error) predictions using CHGNet; (c, d) corresponding predictions using eqV2-31M-omat-mp-salex. Numbers in the periodic table show element-specific average RMSE, with color gradients (dark purple to yellow) indicating low to high error magnitudes.

Figure 2. Performance comparison of uMLIPs on the GB-56 grain boundary dataset. (a, b) Energy and force RMSE (root-mean-square error) predictions using CHGNet; (c, d) corresponding predictions using eqV2-31M-omat-mp-salex. Numbers in the periodic table show element-specific average RMSE, with color gradients (dark purple to yellow) indicating low to high error magnitudes.