Global minimization of gold clusters by combining neural network potentials and the basin-hopping method

Neural network potentials trained by first-principles density functional theory total energies were applied to search for global minima of gold nanoclusters within the basin-hopping method. Using Au58 as an example, we found a new putative global minimum which has a core-shell structure of Au10@Au48 and C4 symmetry. This new structure of Au58 is 0.24 eV per formula more stable than the best previous model that has C1 symmetry. This work demonstrates that neural network potentials combined with the basin-hopping method could be very useful in global minimization for medium-sized metal clusters which might be computationally prohibitive for first principles density functional theory.