[Paper Review] False Prediction of Fundamental Properties of Metals by Hybrid Functionals

Weiwei Gao, Tesfaye A. Abtew+4 more|arXiv (Cornell University)|Apr 23, 2015

Machine Learning in Materials Science3 citations

TL;DR

This paper demonstrates that hybrid functionals—commonly used in density functional theory—produce fundamentally inaccurate predictions for metallic systems, including suppressed density of states at the Fermi level, exaggerated bandwidths, overestimated electron-phonon coupling, and inflated magnetic moments. These errors stem from the unphysical treatment of Fock exchange in hybrid functionals, rendering them unsuitable for metals despite their success in semiconductors.

ABSTRACT

The repercussions of an inaccurate account of electronic states near the Fermi level EF by hybrid functionals in predicting several important metallic properties are investigated. The diffculties in- clude a vanishing or severely suppressed density of states (DOS) at EF, significantly widened valence bandwidth, greatly enhanced electron-phonon (el-ph) deformation potentials, and an overestimate of magnetic moment in transition metals. The erroneously enhanced el-ph coupling calculated by hybrid functionals may lead to a false prediction of lattice instability. The main culprit of the problem comes from the simplistic treatment of the exchange functional rooted in the original Fock exchange energy. The use of a short-ranged Coulomb interaction alleviates some of the drawbacks but the fundamental issues remain unchanged.

Motivation & Objective

• To investigate the reliability of hybrid functionals in predicting fundamental electronic and physical properties of metals.
• To identify the root cause of systematic errors in electronic structure calculations using hybrid functionals for metallic systems.
• To evaluate the impact of inaccurate electronic structure near the Fermi level on derived properties such as electron-phonon coupling, lattice stability, and magnetism.
• To challenge the widespread assumption that hybrid functionals universally improve electronic structure predictions across all material classes.

Proposed method

• Performed density functional theory (DFT) calculations using the Quantum Espresso package with PBE, PBE0, and HSE hybrid functionals.
• Employed norm-conserving pseudopotentials and plane-wave basis sets with energy cutoffs tailored to each system (180 Ry for transition metals, 50 Ry for MgB2, 20 Ry for Na).
• Treated all 3s, 3p, and 3d electrons as valence in transition metals to ensure accurate electronic description.
• Calculated density of states (DOS) with high k-point sampling (40×40×40) and small Gaussian smearing (0.01 eV) to resolve fine features near EF.
• Computed electron-phonon deformation potentials and total energy as a function of atomic displacement to assess lattice stability.
• Conducted spin-polarized calculations for Fe and Ni to compare magnetic moments predicted by PBE and HSE06 functionals against experiment.

Experimental results

Research questions

• RQ1 Why do hybrid functionals like PBE0 and HSE fail to accurately describe the electronic structure of metals near the Fermi level?
• RQ2 How does the inclusion of Hartree-Fock exchange in hybrid functionals distort the density of states and bandwidth in metals?
• RQ3 To what extent do hybrid functionals overestimate electron-phonon coupling and predict false lattice instabilities in metals?
• RQ4 Why do hybrid functionals overestimate magnetic moments in transition metals like Fe and Ni despite reducing self-interaction error?
• RQ5 Can the use of a short-ranged Coulomb interaction (as in HSE) fully resolve the fundamental flaws of hybrid functionals in metallic systems?

Key findings

• The PBE0 functional predicts a vanishing density of states at the Fermi level in sodium, with a pronounced dip even after smearing, indicating a fundamental failure in describing metallic states.
• The PBE0-calculated bandwidth in sodium is 4.2 eV, significantly wider than the PBE result (3.15 eV) and the free-electron model (3.15 eV), indicating unphysical band broadening.
• The HSE functional predicts a softened E2g phonon mode in MgB2 and a double-well potential energy curve under distortion, suggesting a false lattice instability not present in PBE results.
• The HSE06 functional overestimates the magnetic moment of iron by 29% (2.87 μB vs. 2.22 μB experimentally), and nickel by 61% (1.00 μB vs. 0.62 μB), indicating severe overestimation of magnetism.
• The electron-phonon deformation potential is significantly overestimated by hybrid functionals, which may lead to incorrect predictions of superconducting or structural instabilities.
• The root cause of these errors lies in the simplistic treatment of Fock exchange energy, which incorrectly localizes electronic states and distorts the electronic structure near EF, a problem not fully resolved by screening in HSE functionals.