Fig. 5: Performance evaluation of UNEP-v1 for MoTaVW alloys.
From: General-purpose machine-learned potential for 16 elemental metals and their alloys
a Mono-vacancy formation energies from the first version of unified neuroevolution potential (UNEP-v1) and embedded-atom method (EAM)35 compared to density functional theory (DFT) data for an equimolar MoTaVW alloy with 128 atoms sampled from hybrid Monte Carlo and molecular dynamics simulations. Mean absolute error (MAE) and R2 (coefficient of determination) values are provided for comparison. b Peierls barrier for 1/2〈111〉 screw dislocation migration in elemental W (see Fig. S21 for the other three species). c Comparisons of UNEP-v1, EAM35, and DFT results for equimolar MoTaVW alloys sampled from molecular dynamics simulations using 256-atom supercells for a melting process from 10 to 5000 K during 10 ns. UNEP-v1 shows much better predictions than EAM, with a much smaller force MAE as indicated in the legend (see Fig. S22 for similar comparisons for deformation processes). Source data are provided as a Source Data file88.
