correction of DFT energy via understanding of electronic environments
Generate data for training the correction scheme. A list of sample training systems are given in the ./preparation/sample_files folder. It's a list of C, H, O, N containing small molecules with single point calculation done at CCSD(T)/cc-PVTZ level of theory pull down from the NIST CCCBDB dataset. Two parts of information from the high level calculation are needed: energy and coordinates, as given in the folder of each system. With these information, input files for SPARC are prepared that could calculate the energies on DFT level, and the corresponding features for each point in space.
Please run the calculations and get the feature files. it should be a series of csv files corresponding to each features.
Note: only need to run the calcualtions for the functionals of interest instead of all of them, but make sure to change your code for the later part of the workflow
Once all the calculations are finished, please run ./preparation/save_to_hdf5.py script to collect the energy and feature information of all the systems with all functionals of interests into one giant .h5 file.
Note: Please change the path and list of functionals accordingly
Extract the information for each training system, calculate the core reference energies for reference and the functionals, and output the amount of corrections for each system
simply run
python extract_reference_energies.py [functional]
e.g.
python extract_reference_energies.py [GGA_PBE]
Subsample the prepared data files by going to ./training and run
python system_subsample.py [system_name] [data_file_path] [functional]
e.g.
python system_subsample.py C2H2 ../preparation/sample_files/C2H2/C2H2_MCSHLegendre_3_3_3.000000_Legendre_3.h5 GGA_PBE
Please do this for each system
Once all systems are subsampled and saved to ./training/system_subsampled_files/, run
python overall_subsample.py
to get the representitive envrionments across all systems. please make sure the resulting number of representitive environment is less than the total number of systems.