Validation of moment tensor potentials for fcc and bcc metals using EXAFS spectra

Machine-learning potentials for materials, namely the moment tensor (MTPs), were validated using experimental EXAFS spectra first time. The MTPs four metals (bcc W and Mo, fcc Cu Ni) obtained by active learning algorithm of fitting to results calculations density functional theory (DFT). MTP accuracy was assessed comparing metal K-edge experimentally computed from molecular dynamics (MD) simulations. sensitivity method various aspects MD DFT models demonstrated Ni as an example. Good agreement found W, Mo recommended PAW pseudopotentials, whereas a more accurate pseudopotential with 18 valence electrons required achieve similar agreement. use allows one estimate ability in reproducing both average dynamic atomic structures.