Density Sensitivity of Empirical Functionals.

Area of Science:

• Computational chemistry
• Quantum chemistry
• Density Functional Theory (DFT)

Background:

• Empirical fitting of approximate density functionals is standard practice.
• This fitting conflates errors in electron density with energy functional errors.
• Density-corrected DFT (DC-DFT) offers a method to disentangle these error sources.

Purpose of the Study:

• To demonstrate the benefits of DC-DFT in correcting for density-driven errors.
• To illustrate cases where standard fitting procedures fail.
• To show improved accuracy by using Hartree-Fock (HF) densities instead of self-consistent densities.

Main Methods:

• Application of a toy functional to H2+ at varying bond lengths.
• Analysis of Grimme's D3 functional fit for noncovalent interactions using WATER27 and B30 datasets.
• Evaluation of double-hybrid functionals trained on self-consistent densities.
• Comparison of results using self-consistent densities versus HF densities.

Main Results:

• Standard fitting procedures can fail catastrophically when density errors are significant.
• DC-DFT using HF densities yields more accurate results for H2+, noncovalent interactions, and double-hybrid functionals.
• Significant error reduction was observed for binding energies of small water clusters.
• Range-separated hybrids with 100% HF at large distances show reduced sensitivity to these density errors.

Conclusions:

• Separating density and energy errors via DC-DFT is crucial for reliable functional fitting.
• Utilizing HF densities provides a cost-free improvement for DFT calculations prone to density-driven errors.
• The DC-DFT approach enhances the predictive power of approximate density functionals across various chemical applications.