Polyfunctional methodology for improved DFT thermochemical predictions.

Area of Science:

• Computational chemistry
• Quantum chemistry
• Materials science

Background:

• Density functional theory (DFT) is widely used for predicting molecular properties.
• Accurate prediction of enthalpies of formation is crucial for computational thermodynamics.
• Existing DFT functionals exhibit systematic errors in predicting thermodynamic properties.

Purpose of the Study:

• To analyze statistical error distributions for enthalpies of formation predicted by various DFT functionals.
• To identify systematic errors and develop strategies for improving prediction accuracy.
• To explore linear combinations of DFT predictions for enhanced thermodynamic accuracy.

Main Methods:

• Analysis of error distributions for 18 DFT functionals using a test set of 675 molecules.
• Application of a simple empirical correction to single functionals.
• Development and evaluation of linear combinations of enthalpy estimates from different functionals.
• Identification of the best linear unbiased estimator (BLUE) combination, termed polyfunctional 3 (PF3).

Main Results:

• Systematic errors, dependent on valence electrons, were identified for some DFT functionals.
• Empirical corrections significantly improved prediction errors for individual functionals.
• The PF3 method, a combination of B3LYP, BLYP, and VSXC, achieved a mean absolute deviation (MAD) of 2.4 kcal/mol on the G3 set.
• PF3 demonstrated a MAD of 3.0 kcal/mol on a larger set of 675 molecules, outperforming B3LYP (MAD 4.9 kcal/mol) and approaching the accuracy of the more computationally expensive G3 method (MAD 1.2 kcal/mol).

Conclusions:

• Linear combinations of DFT predictions can effectively exploit error correlations to improve accuracy.
• The PF3 method offers a favorable balance between accuracy and computational cost for thermodynamic predictions.
• Further improvements in accuracy are possible through continued method development.