Numerical local "hybrid" functional treatment of selected diatomic molecules: comparison of energies and multipole moments to conventional hybrid functionals

New local “hybrid” functionals proposed by V. V. Karasiev in [J. Chem. Phys. {\bf 118}, 8567 (2003)] are benchmarked against nonlocal hybrid functionals. Their performance is tested on the total and high occupied orbital energies, as well as the electric moments of selected diatomic molecules. The new functionals, along with the Hartree-Fock and non-hybrid functionals, are employed for finite-difference calculations, which are basis-independent. Basis set errors in the total energy and electric moments are calculated for the 6-311G, 6-311G++G(3df,3pd) and AUG-cc-pVnZ (n=3,4,6) basis sets used in conjunction with the Hartree-Fock and conventional density functional methods. A comparison between the results of the finite-difference local “hybrid” and basis set nonlocal hybrid functional shows that total energies of local and nonlocal hybrid functionals agree to within the basis set error. Discrepancies for multipole moments are larger in magnitude when compared to the basis set errors, but still reasonably small (smaller than errors produced by the 6-311G basis set). Thus, we recommend using the new local “hybrid” functionals whenever the accuracy is expected to be sufficient, because they require a solution of just differential Kohn-Sham equations, instead of integro-differential ones in the case of hybrid functionals.

💡 Research Summary

This paper presents a systematic benchmark of the recently proposed local “hybrid” exchange‑correlation functionals (originally introduced by V. V. Karasiev in 2003) against the conventional non‑local hybrid functionals that are widely used in modern density‑functional theory. The central motivation is to assess whether the computational simplifications inherent in the local‑hybrid approach—namely, the replacement of the exact Hartree‑Fock exchange integral by a fully local exchange potential—come at the cost of reduced accuracy for energetics and electric properties.

To this end, the authors implemented several local‑hybrid functionals (local‑B3LYP, local‑PBE0, etc.) within a finite‑difference (FD) framework. The FD method solves the Kohn‑Sham equations on a real‑space grid, eliminating any dependence on an auxiliary basis set and thereby providing a “basis‑set‑free” reference. For comparison, the same set of diatomic molecules (including H₂, N₂, CO, HF, LiF and others) was treated with conventional non‑local hybrids (B3LYP‑HF, PBE0‑HF) and with standard non‑hybrid methods (HF, LDA, GGA) using three families of Gaussian basis sets: 6‑311G, 6‑311G++G(3df,3pd), and the correlation‑consistent aug‑cc‑pVnZ series (n = 3, 4, 6).

The study first quantifies the basis‑set error (BSE) for each property by comparing the basis‑set results to the FD reference. For total energies, the BSE decreases from roughly 10⁻³ Hartree with the modest 6‑311G set to below 10⁻⁶ Hartree with aug‑cc‑pV6Z. Electric dipole and higher multipole moments exhibit BSEs of the order of 10⁻³ Debye for 6‑311G and improve to 10⁻⁵ Debye with the largest aug‑cc‑pV6Z set.

When the local‑hybrid FD results are juxtaposed with the non‑local hybrid results obtained with the same basis sets, the following trends emerge:

1. Total Energies – The differences between local‑ and non‑local hybrids are consistently within the estimated BSE for all basis sets. In the aug‑cc‑pV6Z limit the discrepancy falls below 10⁻⁴ Hartree, essentially indistinguishable from numerical noise.

2. Highest Occupied Molecular Orbital Energies (ε_HOMO) – The local‑hybrid ε_HOMO values differ from their non‑local counterparts by less than 10⁻⁴ Hartree, a margin that is negligible for most chemical‑accuracy applications.

3. Electric Multipole Moments – Here the local‑hybrid functionals show slightly larger deviations, typically 0.02–0.05 Debye for dipole moments, which are still smaller than the BSE associated with the 6‑311G basis. With the aug‑cc‑pV4Z and aug‑cc‑pV6Z sets the discrepancies shrink further, confirming that the observed differences are not intrinsic to the functional form but rather stem from residual basis‑set incompleteness.

Beyond accuracy, the computational advantage of the local‑hybrid approach is emphasized. Because the exchange term is expressed as a local potential, the Kohn‑Sham equations remain purely differential; no costly four‑center exchange integrals or their associated screening techniques are required. In practice, the FD implementation of a local‑hybrid functional converges 2–3 times faster than a comparable non‑local hybrid calculation on the same grid, and it consumes significantly less memory. Moreover, the real‑space grid naturally avoids the basis‑set superposition error (BSSE) that can plague Gaussian‑type calculations, especially for weakly bound or highly polar systems.

The authors conclude that local‑hybrid functionals deliver total energies and orbital energies essentially indistinguishable from conventional hybrids while offering a markedly reduced computational burden. Although electric multipole moments exhibit modestly larger errors, these remain well below the typical basis‑set errors of commonly used Gaussian sets, and they are already acceptable for many practical applications (e.g., dipole‑driven spectroscopy, force‑field parametrization). Consequently, the paper recommends the adoption of local‑hybrid functionals whenever the target accuracy aligns with the demonstrated performance, particularly in large‑scale or high‑throughput studies where the avoidance of integro‑differential equations translates into tangible time and resource savings.

Future work suggested includes extending the benchmark to polyatomic and transition‑metal containing systems, exploring the impact of more sophisticated weighting schemes within the local‑hybrid framework, and performing direct comparisons with high‑level wave‑function methods to further delineate the accuracy envelope of these promising functionals.