hide
Free keywords:
-
Abstract:
Kohn –Sham density functional theory (KS-DFT) has become a cornerstone of computational chemistry in the last few decades. It is used for the computation of chemical properties and the simulation of chemical reactions throughout the various fields of chemical research. Over time, increasingly more complex density functionals (DFs) have been developed to perform calculations faster and more accurate, necessitating numerical procedures in KS-DFT. In the past,
grid-based numerical procedures have been successfully used in KS-DFT. But these procedures are computationally expensive and introduce “grid noise”.
In this work, a new approach to KS-DFT, called analytical DFT (ADFT), developed by F. Neese in 2021, is presented that tries to overcome the hardships of grid -based numerical procedures. ADFT is an analytical variant of the Xα method, which uses an auxiliary exchange fitting basis set (XFit/1–4) to replace the grid. It allows to scale the exchange contribution to the total energy on a per-atom-type basis by scaling atomic α exchange parameters. By default ADFT has the same α parameters of 2/3 as Xα for all atom types (ADFT(Xα)). This work aims to extend ADFT(Xα) by determining a new set of Slater parameters and to compare the performance of ADFT(Xα) to Xα. In the first part the atomic α parameters in ADFT are fitted to atomic unrestricted Hartree–Fock (UHF) energies (called ADFT(HF)). It is shown that the α parameters only have a weak dependence on the orbital basis set. Several benchmarks are performed to compare ADFT(HF) to Hartree– Fock (HF) and ADFT(Xα) to Xα. The obtained results show that the UHF-parametrization does not project the characteristics of HF into ADFT(HF). Furthermore, they show that ADFT(Xα) has a good agreement with Xα with an absolute error of approximately 0.4 kcal mol−1 for the largest exchange fitting basis used in ADFT.
In the second part the performance of ADFT(Xα) and Xαare compared for non-covalently bound systems. It is shown that for π-interaction complexes the agreement of ADFT(Xα) with Xα can be improved by adding two additional f-functions on the atoms C and N to the XFit/3 basis. Additionally, the DFT-D4 dispersion correction is fitted for ADFT(Xα) and its impact is tested in benchmark studies. The studies show that the D4 correction does not improve results of Xα and ADFT(Xα) for small non-covalently bound systems and systems with strong intramolecular dispersion effects, as both methods show slight overbinding for these kinds of systems. For large non-covalently bound systems both methods significantly benefit from the dispersion correction, which decreases their error by approximately −3.0 kcal mol−1 with respect to reference interaction energies from the S30L and L7 benchmark sets.
