Universal non-covalent interaction correction to DFT from machine learning
Install the QML code: http://www.qmlcode.org/installation.html
Uncompress the dft-nci.tar.gz file: "tar -xvf dft-nci.tar.gz"
Modify the dft-nci.py file:
Required modifications:
Select a baseline method by defining the variable "method = N", where N can be 0: Direct ML 1: BLYP-NCI 2: PBE-NCI 3: TPSS-NCI 4: SCAN-NCI 5: B3LYP-NCI 6: PBE0-NCI
Specify the folders containing the xyz files of the test systems as "X_test = rep_mol("test/folder", X_test)" for molecular systems "X_test = rep_sol("test/folder", X_test)" for periodic solids
Optional modifications:
Include more training folders with the energy (txt) and geometries (xyz) files as "X_training = rep_mol("training/folder", X_training)" for molecular systems "X_training = rep_sol("training/folder", X_training)" for periodic solids
Alter the list of kernel widths (sigmas), or the regularization strength (lmda) as "sigmas = [2.0**8]" "lmda = 1e-8"
Include more atoms in the dictionary, or increase the size of the representation as "choices = {'H': 1, 'C': 6, 'N': 7, 'O': 8, 'F': 9, 'Cl': 17, 'Br': 35, 'I': 53}" "max_size=155", "neighbors=429"
Run the code: "python dft-nci.py"