Prepare Dataset

This section describes how to prepare the dataset for training the SALTED model with different ab initio software packages.

What do we need?

  1. Product basis overlap matrices
  2. Density fitting coefficients

Generate Dataset

To date, support for generating these overlap matrices and coefficients is included in three electronic structure packages - PySCF, FHI-aims and CP2K. If you develop another package and would like to develop SALTED integration, please contact one of the developers.

Whichever code is used, the result should be the generation of new directories named overlaps and coefficients in the saltedpath directory. These will be used to train a SALTED model as described in the next section.

PySCF

  1. The following input arguments must be added to the inp.qm section:
    • qmcode: define the quantum-mechanical code as pyscf
    • path2qm: set the path where the PySCF data are going to be saved
    • qmbasis: define the wave function basis set for the Kohn-Sham calculation (example: cc-pvqz)
    • functional: define the functional for the Kohn-Sham calculation (example: b3lyp)
  2. Define the auxiliary basis set using the input variable dfbasis, as provided in the inp.qm section. This must be chosen consistently with the wave function basis set (example: RI-cc-pvqz). Then, add this basis set information to SALTED by running: 
    python3 -m salted.get_basis_info
  3. Run PySCF to compute the Kohn-Sham density matrices: 
    python3 -m salted.pyscf.run_pyscf
  4. From the computed density matrices, perform the density fitting on the selected auxiliary basis set by running: 
    python3 -m salted.pyscf.dm2df

FHI-aims

A detailed description of how to generate the training data for SALTED using FHI-aims can be found at the dedicated SALTED/FHI-aims tutorial.

CP2K

  1. The following input arguments must be included in the inp.qm section:
    • qmcode: define quantum-mechanical code as cp2k
    • path2qm: set the path where the CP2K data are going to be saved
    • periodic: set the periodicity of the system (0D,2D,3D)
    • coeffile: filename of RI density coefficients as printed by CP2K
    • ovlpfile: filename of 2-center RI integrals as printed by CP2K
    • dfbasis: RI (density-fitting) basis filename appended for each species, extracted from CP2K
    • pseudocharge: list of pseudocharges associated with the adopted GTH pseudopotential. NB: the list ordering must be consistent with the ordering of species provided in inp.system.species.
  2. Initialize the systems used for the CP2K calculation by running: 
    python3 -m salted.cp2k.xyz2sys
    System cells and coordinates are extracted from the configuration dataset in XYZ format and saved in folders named conf_1, conf_2, ... located in the path inp.qm.path2qm. NB: cell information (Lattice) must be included in second line of each XYZ configuration, even if it does not change.
  3. Run SCF calculations and save the optimized wavefunction for each configuration in the corresponding folders previously generated. An example CP2K input is provided in cp2k-inputs/SCF.inp.
  4. Print the RI density-fitting coefficients and 2-center RI integrals by restarting the CP2K calculation from the optimized wavefunction. This restart operation derives from the large memory required by the RI fitting procedure, which might require using larger computational resources than the plain SCF cycle. An example CP2K input is provided in cp2k-inputs/rho-RI-print.inp. NB: The RI basis is automatically generated by CP2K from the selected wavefunction basis set, as described in https://doi.org/10.1021/acs.jctc.6b01041, following SMALL, MEDIUM, or LARGE tiers.
  5. Print the RI basis set information required for SALTED postprocessing of the CP2K density. An example CP2K input is provided in cp2k-inputs/RI-basis.inp. This operation can be performed only once for any arbitrary configuration included in the dataset adopting the given choice of RI basis. The output is a single file including wavefunction and RI basis set information of all the species included in the selected test configuration. To extract the RI basis information for each species, run 
    python3 -m salted.cp2k.extract_basis cp2k_basis_filename
    with cp2k_basis_filename the output basis set filename. This will create a separate file for each species in the format, e.g., H-dfbasis, O-dfbasis.
  6. Add the RI basis set information to SALTED by running: 
    python3 -m salted.get_basis_info
  7. Set the inp.qm.coeffile and inp.qm.ovlpfile input arguments according to the filenames of the RI density-fitting coefficients and 2-center RI integrals generated at step 4. Then, convert the full training dataset in SALTED format by running: 
    python3 -m salted.cp2k.cp2k2salted