Measuring transferability issues in machine-learning force fields: The\n example of Gold-Iron interactions with linearized potentials

Abstract

Machine-learning force fields have been increasingly employed in order to\nextend the possibility of current first-principles calculations. However, the\ntransferability of the obtained potential can not always be guaranteed in\nsituations that are outside the original database. To study such limitation, we\nexamined the very difficult case of the interactions in gold-iron\nnanoparticles. For the machine-learning potential, we employed a linearized\nformulation that is parameterized using a penalizing regression scheme which\nallows us to control the complexity of the obtained potential. We showed that\nwhile having a more complex potential allows for a better agreement with the\ntraining database, it can also lead to overfitting issues and a lower accuracy\nin untrained systems.\n