Refactored DL_FFLUX Benchmarking

Description

The models and datasets used to benchmark DL_FFLUX and make a comparison with other MLPs (such as DeePMD-SE, SchNet, NequIP, etc.) are presented here. The datasets have 10,000 geometries with their Cartesian coordinates, atomic IQA energy, charge, multipole moments and forces. The models are trained on these datasets. Models are system-specific.

Steps to reproduce

Metadynamics simulations were performed to sample the initial geometries. Gaussian16 with the level of theory of B3LYP/aug-cc-pVTZ was performed on the geometries, and their wavefunction files were extracted. AIMAll (version 19.10.12) was used to get IQA atomic energies and QTAIM atomic charges and multipole moments.

Institutions

• University of Manchester

  England, Manchester

Categories

Licence