The role of nuclear charges in unifying the descriptions of neural networks (NN)-based force fields

Abstract NN-based force fields development has shown its merit of low cost, accuracy and versatility; where the choice of descriptors plays a central role. In this contribution, descriptors representing atomic environments with/without nuclear charges are employed respectively to construct force fields for three elemental materials Fe, Cr and Al. The descriptions of the force fields can be unified when the nuclear charges are included as one descriptor. This strategy greatly enhance the efficiency of force fields building for different elemental materials, which is treated as feasible and can be extendable to other types of elemental and multi-elemental materials.