CHASAM 2026: Datasets for Chalmers School on Atomic Modeling

Data used in the hands-on notebooks of the CHASAM 2026 school, chasam.materialsmodeling.org/. All MD was run with GPUMD using a NEP potential, and each trajectory archive holds the input structure (model.xyz), the NEP model (nep.txt), the GPUMD input (run.in) and the trajectory (dump.xyz). Zenodo record: 10.5281/zenodo.22790961 Files Archive System Contents BaTiO3_models.tar BaTiO3 structures, a pretrained qNEP model and pretrained tensorial NEP models for dipole, polarizability and atomic virial BaTiO3_spectroscopy.tar BaTiO3 GPUMD runs at 100, 240, 360 and 500 K with dipole-derivative (dpdt.out) and polarizability output CsPbI3_NPT_T650to1_size12_nframes2000.tar.gz CsPbI3 NPT cooling run from 650 K to 1 K, 12x12x12 supercell with 8640 atoms, 2000 frames CsPbI3_NVE_T600_size4_nframes100000.tar.gz CsPbI3 NVE run at 600 K, 4x4x4 supercell with 320 atoms, 100000 frames CsPbI3_NVT_T650_size18_nframes500.tar.gz CsPbI3 NVT run at 650 K, 18x18x18 supercell with 29160 atoms, 500 frames day1_basic_training.zip NiAl initial DFT training and validation databases, and one NEP model with five cross-validation splits day2_active_learning.zip NiAl R2SCAN reference databases, validation data and three active-learning generations, each with structure generation, model and tests day2_advanced_nep_hyperparameters.zip NiAl training and validation data, and NEP models scanning n_max, l_max and ZBL, with RMSE and MD-speed summaries day2_advanced_two_stage_training.zip NiAl training and validation data, and single-stage and two-stage NEP models day3_thermodynamic_integration.tar.gz SiO2 Frenkel-Ladd and reversible-scaling GPUMD runs for coesite and stishovite at 8 GPa, holding NPT equilibration at 300 K, five Frenkel-Ladd switching runs per phase and three reversible-scaling ramps per phase from 300 to 1800 K FAPI_md-run-active-learning.tar.gz FAPbI3 NPT heating run from 1 K to 600 K, 2x2x2 supercell with 96 atoms, 2000 frames FAPI_training.tar.gz FAPbI3 DFT databases for two active-learning generations, and NEP models trained on 100 to 900 structures, with and without ZBL mcmd-vcsgc-data.tar.gz FeCr 700 VCSGC Monte Carlo/MD runs of 8000 atoms, at seven temperatures from 300 to 900 K and 100 values of phi from -2.5 to 2.5 with kappa = 200, holding concentration traces (mcmd-*.out) and thermodynamic output (thermo-*.out)