DFT+U Energy Functional Overview

Updated 14 February 2026

DFT+U Energy Functional is a correction to standard DFT that adds an orbital-selective penalty to enforce near-integer occupancies in localized orbitals.
The methodology employs advanced Hubbard corrections, double-counting schemes, and linear response techniques to mitigate self-interaction and static correlation errors.
Recent advances include orbital-resolved functionals, hybrid approaches, and DFPT integration, offering improved band gap predictions and accurate treatment of strongly correlated systems.

Density Functional Theory (DFT) with an on-site Hubbard $U$ correction—commonly referred to as DFT+U—extends standard Kohn–Sham DFT to more accurately treat systems with localized and strongly correlated electrons. The DFT+U functional introduces an orbital-selective penalty to the DFT total energy that is designed to recover piecewise linearity with respect to fractional occupation, thereby mitigating self-interaction errors typical of (semi)local functionals in correlated subspaces. Over the past two decades, DFT+U has evolved from model-based constructions to rigorously derived, fully first-principles, and symmetry-respecting formulations. Modern functionals also address static correlation, inter-site interactions, and double-counting ambiguities. This article presents a comprehensive technical overview of DFT+U energy functionals, focusing on their foundational principles, explicit mathematical structure, connection to spectral and hybrid DFT functionals, and recent advances addressing many-body and multiorbital effects.

1. Formulation of the DFT+U Energy Functional

The DFT+U total energy is typically written as the sum of a baseline (semi)local or hybrid DFT functional and a subspace-localized Hubbard correction: $E_{\mathrm{DFT}+U} = E_{\mathrm{DFT}} + E_U - E_{\rm dc}$ where $E_{\mathrm{DFT}}$ denotes the standard Kohn–Sham DFT total energy, $E_U$ is the on-site correction accounting for correlations among localized orbitals, and $E_{\rm dc}$ is a double-counting term to subtract the interaction already included in $E_{\mathrm{DFT}}$ .

1.1. Rotationally Invariant Dudarev Construction

A widely adopted, rotationally invariant form is due to Dudarev et al.: $E_{U} = \sum_{I,\sigma} \frac{U^{I}}{2} \mathrm{Tr} \left[ n^{I\sigma} - (n^{I\sigma})^2 \right]$ Here, $I$ runs over atomic sites with correlated subspaces (typically $d$ or $f$ shells), $\sigma$ is the spin, and $n^{I\sigma}$ is the on-site occupation matrix: $n^{I\sigma}_{mm'} = \sum_{i} f_{i}^{\sigma} \langle \psi_{i}^{\sigma} | \phi_{m}^{I} \rangle \langle \phi_{m'}^{I} | \psi_{i}^{\sigma} \rangle$ where $\{|\phi_{m}^{I}\rangle\}$ are local (projector) orbitals and $f_{i}^{\sigma}$ the occupations of Kohn–Sham eigenstates $|\psi_{i}^{\sigma}\rangle$ (Kirchner-Hall et al., 2021, O'Regan et al., 2010, O'Regan et al., 2011). This quadratic penalty enforces idempotency (occupancies near $0$ or $1$) and cures spurious curvature of $E$ with respect to occupation.

1.2. Double Counting: FLL and AMF Schemes

The double-counting correction, $E_{\rm dc}[n]$ , subtracts the average interaction already present in the DFT exchange-correlation functional. It is implemented in variants such as the fully localized limit (FLL): $E_{\rm dc}^{\rm FLL} = \frac{U}{2} N(N-1) - \frac{J}{2} N \left( \frac{N}{2} - 1 \right)$ where $N = \mathrm{Tr}_m[n] = \sum_{m \sigma} n^{\sigma}_{mm}$ and $J$ is the intra-atomic Hund’s coupling (Ryee et al., 2017).

1.3. Generalizations: Orbital-resolved and Two-parameter Functionals

Extensions include orbital-resolved functionals,

$E_U = \sum_{I,\sigma,i} \frac{U^I_{i}}{2} \lambda_i^{I\sigma}(1 - \lambda_i^{I\sigma})$

where $\lambda_i^{I\sigma}$ are occupation matrix eigenvalues, and $U_{i}^I$ are mode-specific Hubbard parameters (Macke et al., 2023). Two-parameter extensions decouple linear and quadratic corrections,

$E_{U_1 U_2} = \sum_{I \sigma} \left[ \frac{U_1^I}{2} \mathrm{Tr}(n^{I\sigma}) - \frac{U_2^I}{2} \mathrm{Tr}(n^{I\sigma} n^{I\sigma}) \right]$

allowing independent tuning of total-energy slope and curvature for improved enforcement of Koopmans’ condition and piecewise linearity (Moynihan et al., 2016, Moynihan et al., 2017).

2. Piecewise Linearity, Self-Interaction, and Spectral Corrections

The DFT+U functional is fundamentally motivated by the requirement that the exact total energy should be piecewise linear in fractional occupation of any localized subspace, with the correct derivative discontinuity at integer filling (the Perdew–Parr–Levy–Balduz condition). Semi(local) DFT typically displays convex curvature (delocalization error), underestimating gaps and misplacing correlated levels. The DFT+U penalty removes this spurious curvature by enforcing

$\frac{\partial^2 E_{\mathrm{DFT}+U}}{\partial n^2} \to 0$

with $U$ chosen such that the total energy becomes linear between adjacent integer occupations, effectively mimicking the derivative discontinuity and improving band gap predictions (Kirchner-Hall et al., 2021, Burgess et al., 2022, Burgess et al., 2024).

Recent functional forms (BLOR, mBLOR) explicitly enforce the flat-plane condition both for total charge and for spin polarization, correcting not only the many-electron self-interaction error (MSIE) but also static correlation error (SCE) within localized shells: $E_{\mathrm{BLOR}} = \frac{U^\uparrow + U^\downarrow}{4} \operatorname{Tr}[\hat N - \hat N^2] + \frac{J}{2} \operatorname{Tr}[\hat M^2 - \hat N^2] + \ldots$ with $\hat N$ and $\hat M$ the total subspace occupancy and magnetization operators (Burgess et al., 2022, Burgess et al., 2024).

3. Physical Interpretation and Determination of $U$

3.1. Linear Response and Density Functional Perturbation Theory

The effective interaction parameter $U$ is most rigorously obtained from the difference of bare and screened responses of the correlated occupation to a subspace-local potential: $U^I = [ (\chi_0^{-1} - \chi^{-1}) ]_{II}$ where $\chi_{IJ} = dN^I/d\alpha^J$ is the response of occupation on site $I$ to a potential shift $\alpha^J$ (Kirchner-Hall et al., 2021, Moynihan et al., 2017). In practice, linear response (Cococcioni–de Gironcoli), DFPT-based monochromatic perturbations, or self-consistent variational optimization algorithms determine $U$ ab initio. Consistent treatment guarantees comparability of energies across geometries and chemical compositions (O'Regan et al., 2010, Macke et al., 2023).

3.2. Many-body and Spectral Foundations

Recent work rationalizes DFT+U in the context of spectral- and Koopmans-compliant functionals. Conceptually, DFT+U acts as an orbital-selective spectral correction, analogous to a model GW-type self-energy but at much lower computational cost. The quadratic form of $E_U$ removes the dominant self-interaction error for local subspaces, while parameteric generalizations can simultaneously fix total-energy curvature and frontier orbital eigenvalues (Moynihan et al., 2016, Janesko, 2023).

The mBLOR functional (many-body BLOR) corrects both one-particle and true correlated many-body errors by constructing a correction surface $E(N,M)$ that is explicitly double-counting free and parameter free, defined only by the measured curvatures of the base functional (Burgess et al., 2024).

4. Hybrid DFT, DFT+U+V, and Functional Unification

4.1. Relation to Hybrid and Extended-Hubbard Functionals

DFT+U and hybrid functionals occupy adjacent positions on a spectrum of exchange-correlation approximations. Ivádý et al. formally demonstrated that a hybrid functional acting on a restricted subspace of atomic-like orbitals can be recast in the +U form with an effective

$U^\mathrm{hyb}_I = \alpha (F^0_I - J^0_I)$

where $\alpha$ is the mixing parameter and $F^0_I, J^0_I$ are Slater integrals for atom $I$ (Ivády et al., 2014). The addition of an on-site potential $V_w$ allows fine-tuning the local correlation strength, curing overscreening and enabling rigorous unification of hybrid-DFT and DFT+U treatments.

4.2. Intersite (DFT+U+V) and Orbital-resolved Corrections

The DFT+U+V formalism includes both on-site ( $U$ ) and intersite ( $V$ ) corrections: $E_{U+V} = \sum_I \frac{U_I^\mathrm{eff}}{2} \sum_{m,\sigma} (n^{I\sigma}_{mm} - \sum_{m'} n^{I\sigma}_{mm'} n^{I\sigma}_{m'm}) - \sum_{I < J} \frac{V^{IJ}}{2} \sum_{mm'\sigma} n^{IJ,\sigma}_{mm'} n^{JI,\sigma}_{m' m}$ such terms restore energy barriers for intersite charge fluctuations and accurately predict gaps and ordering in low-dimensional and charge-ordered systems (Tancogne-Dejean et al., 2019). Orbital-resolved functionals further diversify the correction, deploying distinct $U_i$ parameters by eigenmode, crucial for distinguishing $t_{2g}$ vs. $e_g$ or strongly hybridized vs. localized orbitals (Macke et al., 2023).

5. Practical Considerations, Symmetry, and Extensions

5.1. Projector Choice and Subspace Optimization

Accurate construction of the occupation matrices $n^{I\sigma}$ depends sensitively on the projector set used. Early approaches utilized atomic or hydrogenic orbitals, while subsequent research established that self-consistent nonorthogonal generalized Wannier functions or Löwdin-orthogonalized atomic orbitals yield more physically meaningful Hubbard subspaces, higher transferability, and improved convergence of $U$ (O'Regan et al., 2010, O'Regan et al., 2011).

5.2. Spin, Symmetry, and Relativistic Effects

Noncollinear magnetism and strong spin-orbit interaction require generalization of DFT+U to fully spinorial occupation matrices and Hamiltonians, with explicit treatment of the US-PP augmentation terms and orthonormalization derivatives (Binci et al., 2023). All rotationally-invariant forms are designed to be basis-independent and symmetry-preserving, ensuring that the corrections do not depend on the choice of the local orbital basis.

5.3. Implementation in Perturbative, Linear-Scaling, and Time-dependent Regimes

DFT+U has been embedded in density functional perturbation theory (DFPT) for self-consistent calculation of vibrational, dielectric, and response properties, requiring careful handling of the additional U-dependent variational and Pulay terms (Floris et al., 2019). Efficient algorithms for linearly scaling DFT+U with system size have been demonstrated for NGWF-based methods (O'Regan et al., 2011).

6. Exact Conditions, Recent Advances, and Open Questions

The evolution of DFT+U energy functionals has increasingly adhered to rigorous exact constraints: (i) piecewise linearity in $N$ and $M$ for flat-plane compliance, (ii) rotational invariance, (iii) extensivity under subsystem dissociation, (iv) double-counting freedom, and (v) satisfaction of the derivative discontinuity. The BLOR and mBLOR functionals provide parameter-free, double-counting-free forms derived directly from the curvature of the base functional and exhibit benchmark performance for both total energies and band gaps in strongly correlated systems (Burgess et al., 2022, Burgess et al., 2024). Multi-parameter forms (e.g., $U_1/U_2$ ) further enable simultaneous enforcement of correct gap, total energy, and ionization potential (Moynihan et al., 2016, Moynihan et al., 2017).

A current frontier lies in unifying these advanced corrective functionals with hybrid and GW approaches, ensuring compatibility with time-dependent phenomena and excited-state properties, and extending self-consistent $U/J/V$ determination even for complex, low-symmetry, or open-shell systems. The construction of DFT+U-type functionals that are simultaneously Koopmans-compliant, fully many-body accurate, and computationally tractable remains an important ongoing research direction.