Hubbard U Parameter in DFT+U

• Hubbard U is a material-specific on-site Coulomb interaction term that corrects self-interaction errors in standard DFT methods.
• It is determined using linear-response techniques such as DFPT, with its value sensitive to orbital projector choices and local screening effects.
• Accurate U computation is crucial for predicting electronic, magnetic, and structural properties in strongly correlated materials.

The Hubbard U parameter is a fundamental, material-specific on-site Coulomb interaction term introduced in electronic structure methodologies—from model Hamiltonians to extensions of density functional theory (DFT+U)—to remedy deficiencies in the treatment of correlation effects, particularly for localized, partially filled shells (such as d or f orbitals) in transition-metal and rare-earth systems. The accurate determination, interpretation, and practical deployment of Hubbard U underpins predictive first-principles modeling of strongly correlated materials, affecting structural, magnetic, spectral, and transport properties.

1. Formal Definition and Physical Interpretation

The Hubbard U quantifies the cost in coulombic energy for adding a second electron of the same spin to a localized orbital on the same site. In the DFT+U framework—using the rotationally invariant (Dudarev) correction in its simplest form—the total energy is modified as: $$E_{\text{DFT}+U}[\rho] = E_{\text{DFT}}[\rho] + \sum_I \frac{U^I}{2}\sum_\sigma \operatorname{Tr}\left[ n^{I\sigma}\left(1 - n^{I\sigma}\right) \right] - E_{\text{DC}}[\{n^{I\sigma}\}]$$ Here, $n^{I\sigma}_{mm'}$ are the occupation matrices for localized orbital subspaces (site I, spin σ), obtained by projection of Kohn–Sham states onto chosen atomic or Wannier projectors. The double-counting term $E_{\text{DC}}$ removes on-site contributions already present in the (semi)local DFT exchange-correlation functionals. The essential effect of $U$ is to penalize non-integer (i.e., delocalized) occupancies of these subspaces, correcting self-interaction errors and restoring the missing derivative discontinuity with respect to occupation (Himmetoglu et al., 2013).

2. First-Principles Determination: Linear Response and DFPT Framework

A rigorous, parameter-free definition of Hubbard U within DFT follows from linear-response theory. Adding a small on-site perturbation $\alpha^I$ to the external potential on site I,

$$\Delta V_{\operatorname{ext}} = \sum_{I,m} \alpha^I |\phi_m^I\rangle \langle\phi_m^I|,$$

one measures the change $\Delta n^I$ in the corresponding occupation. The fully interacting response matrix $\chi_{IJ} = \partial n^I / \partial \alpha^J$ and its noninteracting (bare) analog $\chi^0_{IJ}$ are evaluated using either constrained DFT or (preferably) density-functional perturbation theory (DFPT). The on-site parameter is then

$U^I = \left(\chi_0^{-1} - \chi^{-1}\right)_{II}$

(Himmetoglu et al., 2013, Timrov et al., 2022, Bastonero et al., 3 Mar 2025).

The DFPT implementation leverages monochromatic perturbations in a primitive cell, solving Kohn–Sham Sternheimer equations for each reciprocal-space q-point instead of constructing large supercells. This enables efficient and reproducible self-consistent U (and intersite V) calculation, with convergence and parallelization over q-points and Hubbard atoms (Timrov et al., 2022, Bastonero et al., 3 Mar 2025).

3. Computational Approximations and Projector Choices

The numerical value of U is not unique—its precise value depends on:

• Projector Type:
  • Atomic (orthogonalized) orbitals, maximally localized Wannier functions (MLWFs), or generalized Wannier/NGWFs. Results differ, so U must always be computed in the specific representation defining the correlated subspace (Himmetoglu et al., 2013, Bastonero et al., 3 Mar 2025).
• Spherical Approximation:
  • In many codes, only the isotropic Slater integral $F^0$ (“Hubbard U”) is retained, neglecting higher multipole moments; in this case, effective $U_{\text{eff}} = U - J$ (with the Hund’s J exchange parameter) may be used (Himmetoglu et al., 2013).
• Screening:
  • Linear-response U incorporates environment-dependent screening through the density response and is thus distinct from bare atomic Coulomb values; this accounts for chemical context, oxidation state, and coordination (Bastonero et al., 3 Mar 2025).
• Double Counting:
  • Energy functional includes fully localized limit (FLL, appropriate for Mott insulators) or around-mean-field (AMF, better for weakly correlated or metallic cases) versions, with different physical consequences (Himmetoglu et al., 2013).

Numerical results show substantial variance in U for the same element across oxidation states and coordination environments (e.g., Fe 3d: 4.4–6.5 eV; Mn 3d: 3.6–9 eV), necessitating context-specific, self-consistent computation (Bastonero et al., 3 Mar 2025).

4. Variants and Extensions: J, V, and Dynamical U

Beyond the classic DFT+U paradigm:

• DFT+U+J: Hund’s exchange J is included explicitly to stabilize spin alignment and capture physics of noncollinear magnets and Hund’s metals:

$$E_{U+J} = \sum_{I,\sigma} \frac{U^I-J^I}{2}\operatorname{Tr}[n^{I\sigma}(1-n^{I\sigma})] + \frac{J^I}{2}\operatorname{Tr}[n^{I\sigma} n^{I-\sigma}]$$

(Himmetoglu et al., 2013)

• DFT+U+V: Inter-site V corrections (computed via off-diagonal response functions) address nonlocal correlations, relevant for charge-transfer excitations and bond localization in covalent semiconductors:

$$E_{V} = -\frac{1}{2} \sum_{I\neq J,\sigma} V^{IJ} \operatorname{Tr}[ n^{IJ\sigma} n^{JI\sigma}] + \frac{1}{2} \sum_{I\neq J} V^{IJ} n^I n^J$$

(Timrov et al., 2022, Tancogne-Dejean et al., 2019)

• Frequency-Dependent U(ω): Dynamical screening effects are captured using constrained random-phase approximation (cRPA), yielding a frequency-dependent U, with the static limit recovered as $ω \rightarrow 0$ (Himmetoglu et al., 2013).
• Noncollinear and Relativistic Formulations: Modern implementations generalize U to fully relativistic ultrasoft pseudopotentials and noncollinear spin context, essential for spin-orbit-coupled and complex magnetic systems (Binci et al., 2023).

5. Material Dependence, Self-Consistency, and High-Throughput Protocols

Self-consistent determination of U is critical for predictive accuracy; the response matrices must be recomputed on the DFT+U ground state (not only on plain DFT) whenever the underlying electronic structure, geometry, or oxidation state changes appreciably (Himmetoglu et al., 2013, Bastonero et al., 3 Mar 2025). Workflows such as aiida-hubbard automate this iterative process, embedding geometry relaxation, projector update, and self-consistent U (and V) calculation with robust error recovery, enabling high-throughput screening and data-driven approaches (Bastonero et al., 3 Mar 2025).

Statistical analysis of large materials databases reveals that:

• U increases with oxidation state (e.g., Fe2+ to Fe3+ shifts U by ≈0.5 eV), but with significant scatter due to local geometry.
• U differs by up to several eV for the same atom between different chemical locations or crystal fields.
• V decays monotonically with atomic separation, typically 0.2–1.6 eV for transition metal–oxygen bonds, with standard deviations ≈0.2 eV at fixed distance (Bastonero et al., 3 Mar 2025).

Transferability of U across compounds or environments is limited; material-specific, context-aware calculation is required for quantitatively reliable property prediction.

6. Impact on Materials Properties and Applications

Proper treatment of Hubbard U is essential for:

• Electronic Structure: U opens and tunes band gaps, regularizes self-interaction errors, and corrects band-edge placement. In prototypical materials (e.g., Gd₂FeCrO₆, UO₂), U must be tuned (sometimes in concert with negative U on O-2p) to reproduce both the experimental lattice parameter and the insulating band gap (Das et al., 2021, Payami, 2023).
• Magnetism: U stabilizes correlated magnetic order, controls local moments, and enables accurate magnetic ground state prediction, as shown for rare-earth metals under pressure (e.g., Tb: U decreases from ≈4.55 to 3.5 eV from ambient to 65 GPa; accurate reproduction of FM/AFM transitions) (Burnett et al., 2024).
• Phonons and Lattice Dynamics: DFPT generalizations with U allow analytic computation of forces, stresses, and vibrational spectra in correlated insulators (Himmetoglu et al., 2013, Binci et al., 2023).
• Defect and Polaron Physics: The use of U, especially when determined to enforce piecewise linearity of total energy in localized subspaces, is crucial for correct polaron formation energies and charge localization. Linear-response U can overestimate formation energies compared to piecewise-linearity-constrained values (Falletta et al., 2022).

The effect of U on observable properties is strongly property and material-dependent; its systematic inclusion is now routine and often essential in high-throughput discovery and design workflows for batteries, catalysts, and electronic materials (Bastonero et al., 3 Mar 2025).

7. Automated and Machine Learning Approaches

Recent advances incorporate machine learning models that, trained on descriptors from atomic occupation matrices, oxidation states, and chemical environments, predict self-consistent Hubbard U (and V) to accuracies within a few percent of fully DFPT values. Such models enable rapid, high-throughput assignment of U (and V) in large-scale databases or screening studies, bypassing the cost of iterative DFPT without sacrificing realism or transferability (Uhrin et al., 2024).

Automated and reproducible infrastructures such as aiida-hubbard orchestrate the entire calculation (structure, correlated subspace, DFPT response, self-consistency, provenance), supporting robust, scalable, and reproducible materials research (Bastonero et al., 3 Mar 2025).

---

References:

• (Himmetoglu et al., 2013) Hubbard-corrected DFT energy functionals: the LDA+U description of correlated systems
• (Timrov et al., 2022) HP -- A code for the calculation of Hubbard parameters using density-functional perturbation theory
• (Bastonero et al., 3 Mar 2025) First-principles Hubbard parameters with automated and reproducible workflows
• (Payami, 2023) DFT+U study of UO$_2$: Correct lattice parameter and electronic band-gap
• (Tancogne-Dejean et al., 2019) Parameter-free hybrid functional based on an extended Hubbard model: DFT+U+V
• (Falletta et al., 2022) Hubbard $U$ through polaronic defect states
• (Das et al., 2021) First-principles calculation of the electronic and optical properties of Gd$_{2}$FeCrO$_{6}$ double perovskite: Effect of Hubbard U parameter
• (Burnett et al., 2024) First-Principles Calculation of Hubbard U for Terbium Metal under High Pressure
• (Uhrin et al., 2024) Machine learning Hubbard parameters with equivariant neural networks
• (Binci et al., 2023) Noncollinear DFT+$U$ and Hubbard parameters with fully-relativistic ultrasoft pseudopotentials