38 Glass Models Generated Using the MACE Potential in VASP Format

A machine-learning interatomic potential, MACE, was employed to construct structural models of 38 high-refractive glasses. Pretrained foundation models, MACE-MP-L0 and L2, were utilized. The glass models were constructed using a melt-quench simulation, involving melting at 2000 K for 500 ps, followed by a cooling simulation at a rate of 2 K/ps down to 300 K. The models were further equilibrated at 300 K for 500 ps. All simulations were performed using a canonical (NVT) ensemble. The model volume was set to be consistent with the experimental density or the density estimated by GlassNet when experimental data was unavailable. All MD simulations were performed using the LAMMPS package. The atomic motions were integrated with a 1-fs time step, and the temperature was controlled by a Nosé-Hoover thermostat. DFT calculation results based on these glass models will be published in Computational Materials Science under the title "Leveraging LLMs to Identify Patent-Free Glass Materials".