Reproducibility of Hybrid Density Functional Calculations for Equation-of-State Properties and Band Gaps

Yuyang Ji, Peize Lin, Xinguo Ren, Lixin HeThe Journal of Physical Chemistry A 2022 · DOI · arXiv

Abstract: Hybrid density functional (HDF) approximations usually deliver higher accuracy than local and semilocal approximations to the exchange-correlation functional, but this comes with drastically increased computational cost. Practical implementations of HDFs inevitably involve numerical approximations─even more so than their local and semilocal counterparts due to the additional numerical complexity arising from treating the exact-exchange component. This raises the question regarding the reproducibility of the HDF results yielded by different implementations. In this work, we benchmark the numerical precision of four independent implementations of the popular Heyd-Scuseria-Ernzerhof (HSE) range-separated HDF on describing key materials' properties, including both properties derived from equations of state (EOS) and band gaps of 20 crystalline solids. We find that the energy band gaps obtained by the four codes agree with each other rather satisfactorily. However, for lattice constants and bulk moduli, the deviations between the results computed by different codes are of the same order of magnitude as the deviations between the computational and experimental results. On the one hand, this means that the HSE functional is rather accurate for describing the cohesive properties of simple insulating solids. On the other hand, this also suggests that the numerical precision achieved with current major HSE implementations is not sufficiently high to unambiguously assess the physical accuracy of HDFs. It is found that the pseudopotential treatment of the core electrons is a major factor that contributes to this uncertainty.

1. Introduction

Density Functional Theory (DFT) serves as the primary computational tool in materials science. While standard semilocal functionals, such as the Perdew-Burke-Ernzerhof (PBE) functional, have undergone rigorous benchmarking for reproducibility, Hybrid Density Functionals (HDFs) present a more complex challenge. HDFs like the Heyd-Scuseria-Ernzerhof (HSE) functional incorporate a portion of exact Hartree-Fock exchange to mitigate the band gap underestimation common in simpler methods. However, the increased numerical complexity and computational cost of HDFs lead different software packages to employ various numerical strategies, including different basis sets and integration schemes. This study addresses whether these different implementations yield consistent results for the same material properties.

2. Methods and Computational Details

The benchmark evaluates four widely-used DFT implementations: VASP, FHI-aims, Quantum ESPRESSO (QE), and ABACUS. These codes represent a diverse range of methodologies, including Projector Augmented Wave (PAW), all-electron (AE), and Norm-Conserving Pseudopotential (NCPP) approaches, paired with either Plane-Wave (PW) or Numerical Atomic Orbital (NAO) basis sets.

Numerical settings and methods for the four DFT implementations used in the benchmark.

The study utilizes the $\Delta$-gauge metric to quantify differences between energy-volume curves across codes. The dataset consists of 20 binary crystalline solids, primarily semiconductors and insulators with face-centered cubic (fcc) structures, such as Silicon (Si), Gallium Arsenide (GaAs), and Magnesium Oxide (MgO). Comparisons against experimental values for lattice constants, bulk moduli, and band gaps were performed, with careful accounting for zero-point motion (ZPM) effects.

3. Results and Discussion

3.1 Numerical Precision and Reproducibility

The average $\Delta$-values for the HSE functional were found to be two to three times larger than those for the PBE functional. This indicates that HDFs are significantly less reproducible across different software implementations than standard Generalized Gradient Approximations (GGAs). For structural properties, the Mean Absolute Deviation (MAD) between codes for lattice constants ranged from 0.01 to 0.03 Å. This discrepancy is comparable to the deviation between HSE predictions and experimental data, suggesting that current numerical precision limits our ability to further refine the physical accuracy of these functionals.

3.2 Impact of Core Electron Treatment

A key finding is that the treatment of core electrons is a more significant source of discrepancy than the choice of basis set. Results from the all-electron code (FHI-aims) and the PAW-based code (VASP) showed close agreement. In contrast, the NCPP-based codes (QE and ABACUS) formed a distinct group. The common practice of using PBE-generated pseudopotentials for HSE calculations—a "functional mismatch"—appears to be a primary driver of numerical uncertainty in structural properties.

3.3 Robustness of Band Gaps

In contrast to structural properties, band gap predictions were remarkably consistent across all four codes. The MAD between implementations was approximately 0.1 eV, while the error relative to experimental values was roughly 0.6 eV. This suggests that for electronic structure calculations, the choice of software implementation is not the limiting factor in achieving accurate results with the HSE functional.

4. Conclusion

While HSE provides consistent results for band gaps, its reproducibility for structural properties is currently limited by numerical implementations, particularly the use of mismatched pseudopotentials. To improve the precision of HDF calculations, the development of HDF-specific norm-conserving pseudopotentials is essential. Establishing a definitive community standard for HDFs, similar to the original $\Delta$-value project for PBE, will be necessary to push the boundaries of chemical accuracy in computational materials science.

Summary updated

Abstract

Hybrid density functional (HDF) approximations usually deliver higher accuracy than local and semilocal approximations to the exchange-correlation functional, but this comes with drastically increased computational cost. Practical implementations of HDFs inevitably involve numerical approximations─even more so than their local and semilocal counterparts due to the additional numerical complexity arising from treating the exact-exchange component. This raises the question regarding the reproducibility of the HDF results yielded by different implementations. In this work, we benchmark the numerical precision of four independent implementations of the popular Heyd-Scuseria-Ernzerhof (HSE) range-separated HDF on describing key materials' properties, including both properties derived from equations of state (EOS) and band gaps of 20 crystalline solids. We find that the energy band gaps obtained by the four codes agree with each other rather satisfactorily. However, for lattice constants and bulk moduli, the deviations between the results computed by different codes are of the same order of magnitude as the deviations between the computational and experimental results. On the one hand, this means that the HSE functional is rather accurate for describing the cohesive properties of simple insulating solids. On the other hand, this also suggests that the numerical precision achieved with current major HSE implementations is not sufficiently high to unambiguously assess the physical accuracy of HDFs. It is found that the pseudopotential treatment of the core electrons is a major factor that contributes to this uncertainty.
BibTeX
@article{ji2022reproducibility,
  author = {Yuyang Ji and Peize Lin and Xinguo Ren and Lixin He},
  title = {{Reproducibility of Hybrid Density Functional Calculations for Equation-of-State Properties and Band Gaps}},
  journal = {The Journal of Physical Chemistry A},
  year = {2022},
  doi = {10.1021/acs.jpca.2c05170},
}
Figure