Revealing the proton slingshot mechanism in solid acid electrolytes through machine learning molecular dynamics

Training data and molecular dynamics trajectories for the superprotonic phases of cesium dihydrogen phosphate (CsH2PO4, CDP) and cesium hydrogen sulfate (CsHSO4, CHS), as used in the accompanying paper Revealing the proton slingshot mechanism in solid acid electrolytes through machine learning molecular dynamics. The record has two parts. 1. DFT dataset (DFT_data.zip). 1196 DFT-labeled configurations for CsH2PO4 (300–1000 K, 64/128/144 atom cells, Pm-3m and P21/m phases) and 1636 for CsHSO4 (200–2000 K, 112 atom cells, I41/amd and both P21/c phases), collected with the FLARE Bayesian active learning workflow and used to train the Allegro equivariant neural network force field. Energies and forces computed in VASP with the PBE functional. 2. ML MD trajectories (20 files). Production NVT trajectories driven by the trained Allegro force field, run in LAMMPS at fixed experimental lattice parameters with a 0.5 fs timestep and snapshots every 50 fs: CsH2PO4: 1000 atoms: 4 ns at 525, 540, 550, 560 and 575 K; 3 ns at 600, 625 and 650 K; plus 1.5 ns at 525 K. CsHSO4: 1344 atoms: 4 ns at 415, 425, 450, 460, 475, 500, 525, 550, 575 and 600 K; plus 1.5 ns at 500 K. See README.md for the file format, the atom type mapping, and code snippets for reading the trajectories with OVITO or ASE.