Fig. 4: Final potential (TS-f) performance across materials properties.
From: Capturing short-range order in high-entropy alloys with machine learning potentials
The best performing potential for a fixed computational cost falls on the Pareto front. Blue rectangles mark ensembles of identically trained potentials for different model capacities (i.e., \({{\rm{lev}}}_{\max }\)). Potentials with the lowest root-mean-square energy error (εenergy) within each ensemble are marked in green. The four ensembles on the left have levmax = 6, 8, 10, and 12. A breakdown of the MLP performance across individual error metrics is provided in Supplementary Section 3.
