Standard Density Functional Theory (DFT) is the workhorse of modern computational materials science, but it suffers from a fundamental flaw when dealing with correlated electron systems: the self-interaction error. Standard exchange-correlation functionals, such as the Local Density Approximation (LDA) or the Generalized Gradient Approximation (GGA), do not perfectly cancel out the artificial electrostatic repulsion of an electron with its own charge density.
This unphysical self-interaction acts as a spurious force that drives electrons to spread out, favoring an artificially delocalized state. For simple metals or wide-bandgap semiconductors with delocalized valence electrons, this error is negligible. However, for materials containing highly localized valence states—such as the $3d$ orbitals of transition metals or the $4f$ orbitals of rare-earth elements—this artificial delocalization completely breaks the physics.
As noted by Jinming Dong in First-principles investigation of transition-metal-doped group-IV semiconductors, standard approximations fail to capture electron correlation in transition-metal systems. This failure often manifests as a dramatic mischaracterization of a material's fundamental nature: standard DFT frequently predicts true insulators (Mott insulators) to be metallic because it cannot properly split partially filled localized bands. Furthermore, Simon Loftager highlights in A density functional theory study of the carbon-coating effects on lithium iron borate battery electrodes that these self-interaction errors severely distort both the electronic structures and the calculated redox potentials in transition-metal-based materials.
Restoring Localization via the Hubbard U Penalty
The Hubbard-corrected density functional theory (DFT+$U$) method fixes this issue by injecting a piece of model-Hamiltonian physics into the first-principles framework. It borrows the core intuition of the tight-binding Hubbard model: if you attempt to place two electrons within the same localized atomic orbital shell, they must overcome a massive electrostatic repulsion.
To counteract the delocalization error, DFT+$U$ introduces a corrective energy term that penalizes fractional occupancies of the localized shell. The correction can be intuitively understood as a mathematical force that pushes the system toward integer occupancies. If an orbital is more than half-full, the correction lowers its energy, forcing it to become completely occupied; if it is less than half-full, its energy is raised, forcing it to empty out.
$$\Delta E_{\text{Hubbard}} = \frac{U}{2} \sum_{m} \left( n_m - n_m^2 \right)$$
When the orbital occupancy $n_m$ is exactly 0 or 1, the corrective energy term vanishes. For any fractional occupancy ($0 < n_m < 1$), the term adds a positive energy penalty scaled by the parameter $U$ (the on-site Coulomb repulsion). This penalty splits the partially filled bands into a lower Hubbard band (fully occupied states) and an upper Hubbard band (empty states), opening up the correct insulating bandgap.
By restoring the physical localization of electrons, DFT+$U$ allows researchers to accurately map magnetic landscapes and determine true ground states. For instance, Kun Cao et al. demonstrated in First-Principles Modeling of Multiferroic RMn2O5 that integrating Hubbard-$U$ corrections is critical for capturing localized states and determining magnetic exchange constants in strongly correlated materials. Similarly, Zhen-Xiong Shen et al. showed in Magnetic ground state and electron-doping tuning of Curie temperature in Fe3GeTe2: First-principles studies that this parameter corrects self-interaction errors to ensure that the predicted stability of ferromagnetic versus antiferromagnetic states aligns with experimental observations.
Extending the Correction to Inter-Site Interactions
Correlation effects are not always confined to a single atomic shell. In complex hybrid materials, multiple distinct shells may exhibit strong correlation simultaneously. B. I. Min explored this in Mott physics in the 2p electron dioxygenyl magnet O2PtF6, demonstrating that a realistic simulation requires applying on-site Coulomb parameters to multiple distinct atomic shells—such as the $2p$ orbitals of a light molecule and the $5d$ orbitals of a heavy metal—within a single calculation.
Furthermore, when significant orbital hybridization occurs between neighboring atoms (such as a transition metal $3d$ orbital and a ligand oxygen $2p$ orbital), penalizing only the on-site occupancy can lead to an incomplete physical picture. If electrons are shared covalently between sites, an over-correction on the metal site can artificially suppress bonding.
To resolve this, the extended Hubbard functional (DFT+$U$+$V$) introduces an inter-site Coulomb interaction parameter, $V$. As discussed by the authors of First principles study of dielectric properties of ferroelectric perovskite oxides with extended Hubbard interactions, this extended mechanism accounts for the interaction between neighboring atomic orbitals. By self-consistently calculating both the on-site ($U$) and inter-site ($V$) parameters, this approach accurately captures electronic correlations and covalent bonding in materials with significant orbital overlap, matching the precision of expensive hybrid functionals at a fraction of the computational cost.
Capturing Polarons and Localized Charge Transport
The ability of DFT+$U$ to enforce localization is also essential for modeling charge transport in insulating oxides. When an extra electron or hole is introduced into an insulating crystal, it can cause the surrounding atomic lattice to distort. This local structural distortion can trap the charge carrier in a self-reinforcing potential well, forming a quasiparticle known as a polaron.
Because standard DFT functionals suffer from delocalization error, they artificially spread the excess charge across the entire crystal lattice, completely failing to capture polaron formation. By applying the Hubbard $U$ penalty, the functional allows the charge to localize on a single atomic site, which in turn permits the simulated lattice to relax and form the local strain field that stabilizes the polaron.
Jón Steinar Garðarsson Mýrdal et al. leveraged this capability in Computational Investigations of the Electronic Transport in Lithium-Air Battery Materials. By using the Hubbard-corrected functional to overcome self-interaction errors, the authors successfully modeled polaron formation and hopping mobility, providing a clear physical explanation for the electronic transport limits that govern battery discharge products.