Performance of two complementary machine-learned potentials in modelling chemically complex systems

This paper explores the performance of two complementary machine-learned potentials—the moment tensor potential (MTP) and the Gaussian moment neural network (GM-NN)—in modeling the Ta-V-Cr-W alloy family. Both models accurately describe configurational and vibrational degrees of freedom, achieving low root-mean-square-errors (RMSEs) in energies and forces, even for compositions not included in the training data. MTPs demonstrate faster convergence during training, while GM-NNs offer faster execution speeds. Active learning proves partially beneficial, requiring integration with traditional data generation.

Konstantin Gubaev, Viktor Zaverkin, Prashanth Srinivasan, Andrew Ian Duff, Johannes Kästner, Blazej Grabowski