Abstract

Machine-learned interatomic potentials open up the possibility of high-throughput pressure-temperature phase diagram prediction [1]. Phonons are a key component of these calculations. The use of interatomic potentials in lattice dynamics simulations has a long history, as have their inaccuracies resulting from the lack of electronic structure information. While there is a rich and growing literature on the calculation of phonons with machine-learned potentials [2], it is therefore not obvious a priori that a general-purpose potential will have the accuracy across structures and pressures necessary to predict high-quality phase diagrams. In this talk, I show how to use the ephemeral data-derived potentials (EDDPs) [3] in high-throughput phonon calculations. I discuss their strengths as well as weaknesses, and demonstrate how to balance them in order to successfully predict the phase diagram of elemental lead up to 20 GPa and 2500 K.

[1] Pascal T. Salzbrenner, Se Hun Joo, Lewis J. Conway, Peter I. C. Cooke, Bonan Zhu, Milosz P. Matraszek, William C. Witt, Chris J. Pickard; Developments and further applications of ephemeral data derived potentials. J. Chem. Phys. October 2023; 159 (14): 144801. [2] Janine George, Geoffroy Hautier, Albert P. Bartók, Gábor Csányi, Volker L. Deringer; Combining phonon accuracy with high transferability in Gaussian approximation potential models. J. Chem. Phys. July 2020; 153 (4): 044104. [3] Chris J. Pickard; Ephemeral data derived potentials for random structure search. Phys. Rev. B. July 2022; 106: 014102