Validation datasets for MBD-ML: Many-body dispersion from machine learning for molecules and materials

These are the datasets (.xyz format) that were used to validate the performance of the MBD-ML model in predicting a0 and C6 ratios of a wide variety of molecular and crystal structures. The files contain both the 3D coordinates of the input geometries and reference and MBD-ML predicted MBD quantities (a0, C6, energies, forces and stresses). The following keys were used: hirshfeld_ratios: DFT a0 ratios hirshfeld_ratios_so3lr: a0 ratios predicted by MBD-ML c6_ratios: DFT C6 ratios c6_ratios_so3lr: C6 ratios predicted by MBD-ML For MBD energies, forces and stresses for all datasets except for Omol25, the following keys were used: E_MBD_ref: MBD-NL energy E_MBD_so3lr: MBD-ML energy F_MBD_ref: MBD-NL atomic forces F_MBD_so3lr: MBD-ML atomic forces For OMol25, additionally D3 and D4 force corrections are provided, and the following keys are used instead: E_MBD_ref: MBD-NL energy E_MBD_pbe0_so3lr: MBD-ML energy F_MBD_ref: MBD-NL atomic forces F_MBD_pbe0_so3lr: MBD-ML atomic forces d3_vdw_forces: D3 atomic forces d4_vdw_forces: D4 atomic forces