Proper orthogonal descriptors for efficient and accurate interatomic potentials

We present the proper orthogonal descriptors for efficient and accuracy representation of potential energy surface. The surface is represented as a many-body expansion parametrized potentials in which are functions atom positions parameters. Karhunen-Lo\`eve (KL) employed to decompose into set (PODs). Because rapid convergence KL expansion, relevant snapshots can be sampled exhaustively represent atomic neighborhood environment accurately with small number descriptors. used develop interatomic by using linear determining coefficients from weighted least-squares regression against density functional theory (DFT) training set. comprehensive evaluation POD on previously published DFT data sets comprising Li, Mo, Cu, Ni, Si, Ge, Ta elements. diverse pool metals, transition semiconductors. comparable that state-of-the-art machine learning such spectral neighbor analysis (SNAP) cluster (ACE).