Data and Code for "Machine Learning for Bridging the Gap between Density Functional Theory and Coupled Cluster Energies"

Abstract

The datasets, models, and scripts were created to achieve an accurate prediction of the increment of single-point energies between density functional theory (DFT) and wavefunction-based methods, which led to our submitted article: 'A Machine Learning Approach for Bridging the Gap between Density Functional Theory and Coupled Cluster Energies'.

We used the ORCA quantum chemical package to compute the geometries of each species at the B3LYP-D3(BJ)/cc-pVTZ level of theory. The optimized structure was subsequently employed for single-point (SP) computations at the DLPNO-CCSD(T)/cc-pVTZ and CCSD(T)/cc-pVTZ levels of theory. All data were extracted from the calculations and compiled in the provided .csv files.

With the datasets and prediction scripts, it is possible to forecast the differences in single-point (SP) energies between the B3LYP-D3(BJ)/cc-pVTZ and DLPNO-CCSD(T)/cc-pVTZ (for monomers and dimers) levels of theory, as well as to the CCSD(T)/cc-pVTZ level of theory for monomers.

The datasets can be opened and read with any text editor. The Pytorch models can be loaded and manipulated as usual (https://pytorch.org/tutorials/beginner/saving_loading_models.html).

The prediction can be made by installing a suitable Python environment and setting the code line: test_database = f'TestDatabase_{mode}.csv' to the desired dataset for prediction. The format and column names of the file should match the uploaded dataset files. Once the line is modified, a prediction can be generated using the following command, for example, “python gen_predictions_CCSDt.py”.