Global optimization of Ta_n clusters by deep neural network

Global optimization is performed on Ta_n (n = 9–13) clusters by deep neural network (DNN) combined with density functional theory (DFT) method. All of the previously known cluster isomers within relative energy of 1.5 eV (except 2.0 eV for Ta₁₀) are confirmed by our calculations. Moreover, new cluster isomers within relative energy of 1.5 eV (except 2.0 eV for Ta₁₀) are reported. More complicated high-dimensional PESs that correspond to larger-sized clusters can be better explored by the DNN method because more new low-lying energy isomer configurations are found with increasing cluster size.