Formulation of Small Test Sets Using Large Test Sets for Efficient Assessment of Quantum Chemistry Methods

Area of Science:

• Computational Chemistry
• Quantum Chemistry
• Method Development

Background:

• Accurate prediction of thermochemical quantities is crucial for computational chemistry.
• Existing large datasets for method assessment are computationally expensive to evaluate.
• Deviations in Density Functional Theory (DFT) methods require careful characterization.

Purpose of the Study:

• To develop a rapid and accurate model for evaluating quantum chemistry methods.
• To establish a reliable indicator for method robustness in thermochemical calculations.
• To create specialized models for assessing accuracy in specific chemical reaction types and material properties.

Main Methods:

• Literature data analysis of DFT method deviations for main-group thermochemical quantities (MGCDB82 dataset).
• Statistical techniques including lasso regularization and forward selection to devise the MG8 model.
• Development of additional small-data-set models (MOR13, SBG5, MB13) for specific properties.

Main Results:

• The devised MG8 model, using 64 data points, accurately estimates the mean absolute deviation (MAD) of the extensive MGCDB82 dataset.
• An estimated MAD (EMAD_MG8) of approximately 4 kJ/mol is proposed as an indicator of highly robust quantum chemistry methods.
• Specialized models were formulated for metal-organic reactions, semiconductor band gaps, and artificial species, expanding assessment capabilities.

Conclusions:

• The MG8 model offers a computationally efficient approach for evaluating new quantum chemistry procedures.
• Methods with lower EMAD_MG8 values are considered more robust, while higher values necessitate further investigation.
• The developed models provide versatile tools for comprehensive accuracy assessment across diverse chemical systems and properties.