Training sets for: Graph Neural Network Potentials for Temperature-Dependent Ferroelectric Instability in Pb1-xSnxTe

Given are the training sets and the description of how they were generated for the Bachelor's End Project of Laura Morales Navarrete titled "Graph Neural Network Potentials for Temperature-Dependent Ferroelectric Instability in Pb1-xSnxTe". The training sets are provided in .extxyz format, with the stress tensors and total energy written in the header, and forces written in the last three columns for each structure. In the project, these training sets are used to train Allegro models. Procedure for generating the training sets: Training structures are sampled from three 100 ps-long molecular dynamics (MD) runs with MD timesteps of 1 fs using the on-the-fly learning procedure as implemented in VASP. These runs are performed using 2x2x2 supercells of the conventional cubic cell (64 atoms). The first of the three runs is performed at 300 K, followed by a run at 100 K, and finally a uniform cooling run from 300 K to 50 K in NPT ensembles at 10^5 Pa. To maintain constant temperature and pressure, Parinello-Rahman dynamics is used with friction coefficients set to 3 ps^-1 for all atomic species and lattice degrees of freedom. After sampling, density functional theory (DFT) single-point calculations are performed on all structures. For the DFT calculations, the projector-augmented wave (PAW) technique is used to model the electron-ion interaction, with the outermost electrons of Pb (6s^26p^2), Sn (5s^25p^2) and Te (5s^25p^4) treated as valence electrons. The electronic interactions are modeled using the PBEsol exchange-correlation functional within the generalized gradient approximation (GGA). Energy convergence criterion of 10^-6 eV is used for the Self-consistent field cycles. The calculations are performed using a 2x2x2 Gamma-centred Monkhorst-Pack k-grid and a kinetic energy cutoff of 250 eV.