Welcome to VSpecNN page.

VSpecNN is a highly efficient multi-task ML surrogate model to accurately calculate infrared (IR) and Raman spectra based on dipole moments and polarizabilities obtained on-the-fly via ML-enhanced molecular dynamics (MD) simulations. For more details about VSpecNN, please visit https://arxiv.org/abs/2402.06911.

environment preparation.

The environment for training is controlled by anaconda. After installation of anaconda, simply run:

conda env create -f environ.yaml

Data preparation.

In order to train VSpecNN from scratch, the users need to prepare their own dataset (coordinates, energies, gradients, dipoles and polarizablities) for training. We have prepared demo dataset in ./data/train, in which contains 5 files:

1. xyz.dat: coordinates in extxyz format in (n_moles * n_atoms, 3) (Ångström)
2. grad.dat: gradients (n_moles * n_atoms, 3) (Hartree/Bohr)
3. energy.dat: energeis (n_moles, 1) (kcal/mol)
4. dipole.dat: dipoles (n_moles, 3) (a.u.)
5. polarizability.dat: polarizabilities (n_moles, 9) (a.u.)

Once the above 5 files exist, run gen_data.py train in ./data folder to generate a data.pt file.

Model training.

After preparation of data.pt, run run_total.py in VSpecNN folder, the best model will be saved in log folder.