dftd3

This keyword is used to add the DFT-D3 dispersion correction to NEP. It has no effect when any other potential is used. For theoretical background on DFT-D3, see [Grimme2010] and [Grimme2011].

Syntax

dftd3 <functional> <potential_cutoff> <coordination_number_cutoff>

where functional is the exchange-correlation functional used in generating the reference data for training the NEP model, potential_cutoff is the cutoff radius (in units of Angstrom) for the D3 potential, and coordination_number_cutoff is the cutoff radius (in units of Angstrom) for calculating the coordination numbers.

This keyword can be put anywhere in the run.in input file.

We have only implemented the Becke-Johnson (BJ) damping, and a full list of supported functionals can be found from the following link: https://www.chemiebn.uni-bonn.de/pctc/mulliken-center/software/dft-d3/functionalsbj

Example

dftd3 pbe 12 6

This will add the DFT-D3 dispersion correction to NEP with the PBE functional and a cutoff radius of 12 Angstrom for the D3 potential and a cutoff radius of 6 Angstrom for the coordination number.

Tips

  • It usually requires to test the convergence with respect to the cutoff radii, and a balance between accuracy and efficiency must be achieved.

  • The user is responsible for not double counting the dispersion correction, i.e., it is not a good idea to add the DFT-D3 correction to a NEP model that has been trained against a dataset containing dispersion correction (no matter what flavor it is).

  • Currently cannot use DFT-D3 for the multi-GPU version of NEP.