Dataset for "Topological robustness of thermally disordered lattices: From average structures to ensemble electronic properties"

This dataset accompanies the manuscript "Topological robustness of thermally disordered lattices: From average structures to ensemble electronic properties". It contains the input files, selected output files and analysis scripts needed to trace how the finite-temperature band structure and band inversion of Bi2Se3 were obtained, from the molecular dynamics trajectories through to the final figures. All VASP calculations (version 6.5.1) use a 135-atom 3×3×1 supercell of Bi2Se3 at the PBE level with the D3 dispersion correction and PAW potentials. The ab initio molecular dynamics runs used an on-the-fly machine-learned force field and consist of 20,000 steps with a 2 fs time step. The dataset includes the NpT runs used to determine lattice parameters at 600, 750, 900 and 1050 K and the NVT production runs at 150, 300, 600, 750, 900 and 1050 K. For each run the input files, the trajectory (XDATCAR), the force-field training set and log (ML_ABN, ML_LOGFILE), OSZICAR, and CSV files with the temperature, lattice parameters, cell volume, stress tensor, pressure, band gap and mean-square displacements extracted from the runs are provided. At each temperature, the electronic structure was calculated for 21 snapshots drawn from the NVT trajectory, 10 structures from harmonic Monte Carlo sampling (labelled MTS in folder names), one Zacharias–Giustino special-displacement structure (ZG) and the time-averaged structure. Each structure was calculated at 15 values of the spin-orbit coupling scale λ between 0 and 1 using a modified VASP code that reads λ from a file named SOCSCALE. For every calculation the INCAR, KPOINTS, POSCAR, CONTCAR, IBZKPT, SOCSCALE and EIGENVAL files are included, together with the tables of absolute and reconstructed signed band gaps for each temperature. The static calculations used to relax the structure at 0 K and to generate the harmonic phonon-based structures are also included. The ABINIT calculations (version 10.8.3) test the effect of electronic temperature on the band gap using the KDT16 finite-temperature exchange-correlation functional in the primitive cell at 0, 300 and 1000 K. Their input, output and eigenvalue files are included. The Python scripts used for snapshot selection, mean-square displacement analysis, signed band-gap reconstruction and plotting are in vasp-calculations/plotting-scripts. The README there explains each script, its input parameters and the shell commands used to extract the CSV files. The README-MD-ML.md file in vasp-calculations describes the molecular dynamics files and the row layout of the extracted CSV files. Large output files (OUTCAR, vasprun.xml, WAVECAR) are not included because of their size, and POTCAR files are not included for licensing reasons.