Global structure optimization following imaginary phonon modes accelerated by machine learning potentials in Cu, Ag, and Au

抄録

Algorithms of crystal structure prediction produce many different structures, some of which are dynamically unstable. Following the imaginary phonon modes obtained by lattice dynamics calculations, dynamically stable structures can be rationally derived from unstable structures. Following the imaginary phonon modes, however, generally requires lengthy and often prohibitively expensive calculations. In this study, we employ polynomial machine learning potentials to predict globally stable and metastable structures following the imaginary phonon modes. As a result, we discover many dynamically stable and metastable structures efficiently, and the face-centered cubic structure is the globally stable structure consistent with experimental reports for the elemental Cu, Ag, and Au.