DFT+U+V is equivalent to DFT+U with density-dependent hybridized projectors

arXiv:2607.18071 · cond-mat.mtrl-sci, cond-mat.str-el, physics.chem-ph, physics.comp-ph · Submitted 2026-07-20 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: We're looking at the paper: "DFT+U+V is equivalent to DFT+U with density-dependent hybridized projectors".

Mira: Here's the gist of the paper. DFT+U+V is equivalent to DFT+U with density-dependent hybridized projectors The inter-site +V corrections of DFT+U+V are exactly on-site DFT+U evaluated on a redefinition of the Hubbard projectors This equivalence is rigorous: when the projectors are allowed to adapt to the density matrix, the single choice of rotation e S reproduces both the DFT+U+V energy and, through the projectors’ own dependence on the density matrix, its potential.

Kai: The Core Equivalence The central finding is that to first order in V /U, DFT+U+V is exactly on-site DFT+U evaluated on a redefinition of the Hubbard projectors.

Mira: This means the inter-site +V correction is essentially a redefinition of the Hubbard projectors.

Kai: This equivalence is established by rewriting the DFT+U energy in a basis of eigenvectors of the on-site occupation matrix n II sigma, with eigenvalues lambda I sigma, and then hybridizing each site with its V-coupled neighbors.

Paper summary: Kai: So, we just talked about how the authors are arguing that DFT plus U plus V is equivalent to DFT plus U with density dependent hybridized projectors. The core idea is that the inter site plus V corrections can be viewed as a redefinition of the on site projectors when they adapt to the density matrix >

Mira: Right. They establish this equivalence by looking at how those corrections reduce the curvature of the total energy with respect to occupations of each Hubbard subspace, specifically those eigenvalues lambda I sigma i >

Lev: So they start by showing that a two hubbard correction directly reduces that local curvature, and with a certain choice for U times one, you can actually remove that local curvature entirely from the DFT plus U energy term >

Kai: That's an ansatz, not an exact condition. They say correcting local curvature might help but it doesn't guarantee fixing the global curvature with respect to total particle number >

Mira: They justify this by looking at what happens if the rest of the system only interacts through an energy that is linear in site occupation, which happens if the on site interactions are so strong that you can almost ignore all other electron interactions or if there's a constant potential >

Lev: That’s a technical justification for why they can treat the local curvature as spurious under certain approximations >

Kai: But this paper is about showing how this connection works, even with the caveats. It's about proving that inter site corrections are tied to these projectors in a specific way >

Mira: And it shows that when you look at the hybridization of those projectors, the admixture of one state grows with V over U and also with that off diagonal occupancy term lambda I sigma i >

Lev: So, it connects the physics of how bonds form directly to how those orbital definitions are modified in a self consistent way >

Conclusion: Kai: So, looking at this paper by Linscott and Carta, they aren't claiming a huge new physical phenomenon. They’re doing something more fundamental about the tools we use for calculations. They are showing that the way we add inter site plus V corrections is really just a different mathematical description of how we select our on site projectors >

Mira: Exactly. The implication for us is that instead of treating the choice of projectors and the inter site corrections as separate knobs to turn, they are deeply intertwined degrees of freedom >

Lev: It means when we think about running this on real hardware, you can't just pick the best orbitals and then add a separate correction term without thinking about how that choice affects the whole system's energy landscape >

Kai: So what this really changes is that it pushes us to figure out what those Hubbard projectors are at their core. It suggests an "optimal" set of projectors might be found by correcting the curvature based on localized orbitals >

Mira: And it opens up a new way to think about calculating those parameters, suggesting that the parameter you compute and the parameter that appears in the functional are conjugate to different occupations >

Lev: But I'm still thinking about how this translates to practical computation. It doesn's just mean we can use any arbitrary set of orbitals; it means we have a new recipe for defining them >

Edward Linscott, Alberto Carta, Nicola Marzari

PSI Center for Scientific Computing, Theory and Data · Paul Scherrer Institute · National Centre for Computational Design and Discovery of Novel Materials (MARVEL) · Theory and Simulation of Materials (THEOS) · Cavendish Laboratory, University of Cambridge

cond-mat.mtrl-sci, cond-mat.str-el, physics.chem-ph, physics.comp-ph

Submitted: 2026-07-20

Updated: 2026-09-07

Comments: 9 pages with an additional 9 pages of supplemental material

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 92/100

The gist: DFT+U+V is equivalent to DFT+U with density-dependent hybridized projectors The inter-site +V corrections of DFT+U+V are exactly on-site DFT+U evaluated on a redefinition of the Hubbard projectors…

Key concepts

DFT+U+V Equivalence
The paper proves that DFT+U with inter-site +V corrections is mathematically identical to DFT+U when the on-site Hubbard projectors are allowed to adapt based on the density matrix. This equivalence holds rigorously, showing that inter-site corrections are fundamentally about redefining these projectors.
Hubbard Projectors
These projectors define how electronic states are hybridized between different atomic sites in a material. The study shows that even established inter-site corrections are essentially choices made within these projectors. Optimal selection of these projectors is crucial for accurately describing the system's energy and potential.
Covalency Restoration
The +V correction acts to make the on-site correction aware of electronic covalency—the degree to which electrons are shared between atoms. This is achieved by rotating the projectors, causing a mixture of states between sites that grows with the V/U ratio and off-diagonal occupancy.

Terminology

Summary

DFT+U+V is equivalent to DFT+U with density-dependent hybridized projectors The inter-site +V corrections of DFT+U+V are exactly on-site DFT+U evaluated on a redefinition of the Hubbard projectors This equivalence is rigorous: when the projectors are allowed to adapt to the density matrix, the single choice of rotation e S reproduces both the DFT+U+V energy and, through the projectors’ own dependence on the density matrix, its potential. p. 4

Where the paper says it

This equivalence is rigorous: when the projectors are allowed to adapt to the density matrix, the single choice of rotation e Sˆ reproduces both the DFT+U+V energy and, through the projectors’ own dependence on the density matrix, its potential (Eq. In contrast, if this redefinition is attempted with frozen projectors, as in most current implementations, the equivalence becomes partial: the total energy and the spectrum then call for differently hybridized projectors (e Sˆ and e 2Sˆ , respectively), and no frozen-projector functional reproduces the DFT+U+V potential exactly (Supp.

The Core Equivalence

The central finding is that to first order in V /U, DFT+U+V is exactly on-site DFT+U evaluated on a redefinition of the Hubbard projectors. This means the inter-site +V correction is essentially a redefinition of the Hubbard projectors. p. 2

Where the paper says it

Equation (4) can be differentiated, upon which we find that, to first order, the DFT+U potential applied to these hybridized projectors reproduces the DFT+U+V potential. This is the central result of this work: in the limit V ≪ U, the inter-site +V correction is exactly a redefinition of the Hubbard projectors.

This equivalence is established by rewriting the DFT+U energy in a basis of eigenvectors of the on-site occupation matrix n II sigma, with eigenvalues lambda I sigma, and then hybridizing each site with its V-coupled neighbors. p. 1
Where the paper says it

This Letter demonstrates that to first order in V /U, DFT+U+V is exactly on-site DFT+U evaluated on a redefinition of the Hubbard projectors. To prove this, we first rewrite the DFT+U energy in the basis {|φ Iσ i ⟩} of eigenvectors of the on-site occupation matrix n IIσ , with eigenvalues λ Iσ i .

The generator S is built from the off-diagonal density-matrix elements that quantify the covalency between sites I and J. p. 1
Where the paper says it

1d–1f alongside the original projectors (panels a–c). We then hybridize each site with its V -coupled neighbors by rotating |φ Iσ i ⟩ −→ e Tˆ |φ Iσ i ⟩, where Tˆ is an as-yet-unknown anti-Hermitian generator that induces a weak hybridization between neighboring sites, built from the Hubbard parameters and the intersite occupations Λ IJσ ij = ⟨φ Iσ i |ρˆ σ |φ Jσ j ⟩ (the off-diagonal density-matrix elements that quantify the covalency between sites I and J).

Mechanism of Covalency Restoration

The role of a +V correction is thus to make the on-site correction aware of covalency. The authors show that the rotation acts on the projectors as phi I sigma i> → e Tphi I sigma i> -X J X j Lambda JI sigma ij U I (1 - 2lambda I sigma i) phi J sigma j>, where the admixture of phi J sigma j> grows with V IJ /U and with the off-diagonal occupancy Lambda IJ sigma ij. p. 2

Where the paper says it

the 2p orbitals of oxygen in a transition metal oxide). The admixture of |φ Jσ j ⟩ grows with V IJ /UI and with the off-diagonal occupancy Λ IJσ ij , and is smallest when orbital i is near integer occupation.

This makes precise the common intuition that +V “restores covalency” by applying +U to hybridized projectors that already contain the bond’s covalent character. p. 3
Where the paper says it

This makes precise the common intuition that +V “restores covalency”: to first order, +V adds no physics beyond the occupationcurvature penalty of +U. What changes is where that penalty is measured: now, it is applied to hybridized projectors that already contain the bond’s covalent character, rather than to bare projectors, where the same correction would disrupt that covalency.

Constraints and Limitations

The equivalence holds exactly in the limit V << U, but it breaks down when the denominator in Eq. (5) vanishes, which occurs when V couples two levels for which U I(1-2lambda I sigma i) = U J(1-2lambda J sigma j). p. 9

Where the paper says it

WHERE THE PERTURBATIVE ROTATION BREAKS DOWN The explicit form of the generator (Eq. (5)) assumes that the denominator is nonvanishing: it fails whenever V couples two levels for which U I (1−2λ Iσ i ) = U J (1−2λ Jσ j ).

In such degenerate cases, the remedy is to apply degenerate perturbation theory, where the inter-site coupling dictates the projectors outright. p. 4
Where the paper says it

Even so, an ordinary DFT+U calculation with hybridized frozen projectors may stand in — approximately — for DFT+U+V, bringing the far broader ecosystem of DFT+U implementations and property extensions to bear. The message to practitioners of DFT+U and DFT+U+V is that the choice of projectors, the choice of Hubbard parameters, and the choice of whether to apply inter-site corrections are deeply intertwined and should never be treated as independent degrees of freedom.

Furthermore, while freezing projectors does not affect energy equivalence in a fixed density matrix, it leads to a mismatch in the potential when compared to DFT+U+V because no single choice reproduces both the spectrum and the energy. p. 4
Where the paper says it

This equivalence is rigorous: when the projectors are allowed to adapt to the density matrix, the single choice of rotation e Sˆ reproduces both the DFT+U+V energy and, through the projectors’ own dependence on the density matrix, its potential (Eq. In contrast, if this redefinition is attempted with frozen projectors, as in most current implementations, the equivalence becomes partial: the total energy and the spectrum then call for differently hybridized projectors (e Sˆ and e 2Sˆ , respectively), and no frozen-projector functional reproduces the DFT+U+V potential exactly (Supp.

Implications for Theory Development

The result sharpens the question of what the Hubbard projectors fundamentally are. The authors conclude that even established inter-site correction is, at heart, a choice of projectors. p. 4

Where the paper says it

The message to practitioners of DFT+U and DFT+U+V is that the choice of projectors, the choice of Hubbard parameters, and the choice of whether to apply inter-site corrections are deeply intertwined and should never be treated as independent degrees of freedom. More deeply — with DFT+U+V now embedded in community codes and automated high-throughput workflows [36, 59–62], increasingly the default extension of DFT+U — this work sharpens the question of what the Hubbard projectors fundamentally are, by revealing that even the established inter-site correction is, at heart, a choice of projectors.

This suggests that determining how projectors might be chosen optimally is now all the more pressing. p. 4
Where the paper says it

More deeply — with DFT+U+V now embedded in community codes and automated high-throughput workflows [36, 59–62], increasingly the default extension of DFT+U — this work sharpens the question of what the Hubbard projectors fundamentally are, by revealing that even the established inter-site correction is, at heart, a choice of projectors. Determining how projectors might be chosen optimally is now all the more pressing.

p. 4
Where the paper says it

More deeply — with DFT+U+V now embedded in community codes and automated high-throughput workflows [36, 59–62], increasingly the default extension of DFT+U — this work sharpens the question of what the Hubbard projectors fundamentally are, by revealing that even the established inter-site correction is, at heart, a choice of projectors. Determining how projectors might be chosen optimally is now all the more pressing.

The equivalence does not retroactively justify computing V as the off-diagonal elements of the response matrices used to compute U. p. 3
Where the paper says it

S10 for details). In other words, the equivalence we have uncovered does not retroactively justify computing V as the off-diagonal elements of the response matrices used to compute U as is common practice [16]; instead, it compounds the pre-existing discrepancy between the site-diagonal occupations the response matrices measure and the inter-site blocks that EV penalizes.

Computational Details and Parameters

The authors computed Hubbard parameters from first-principles using density-functional perturbation theory, yielding an on-site U Ni = 6.95 eV on each Ni 3d site and an intersite V Ni-O = 0. p. 0

Where the paper says it

Inter-site +V corrections were introduced to counter this and are now widely used, but a formal justification has been lacking. Here we show that — to first order in V /U — intersite corrections are exactly equivalent to on-site DFT+U evaluated on a density-matrix-dependent redefinition of the Hubbard projectors, hybridized with those of neighboring sites, providing insight into the explicit mechanism by which V affects covalency.

77 eV between each Ni 3d site and the 2p orbitals of its six neighboring oxygens. p. 1
Where the paper says it

covalency) that might be occurring between these orbitals and those of neighboring atoms (because effective hybridization between orbitals requires them to have similar energies) [34, 35]. This is not a corner case: away from the most ionic compounds, metal–ligand hybridization is the rule rather than the exception, and a purely on-site correction cannot see it.

The expansion parameter is controlled by the spectral norm S, which for typical magnitudes V and U is of order V /U. p. 2
Where the paper says it

(5) (For bulk systems, Sˆ takes the k-dependent form given in Supp. ) The expansion is controlled by the norm of the generator ∥Sˆ∥, which, for typical magnitudes V and U of the inter-site couplings and on-site potentials, is of order V /U (see Supp.

For bulk NiO, the norm is nearly independent of k: S(k) ranges between 0. p. 2
Where the paper says it

(5) (For bulk systems, Sˆ takes the k-dependent form given in Supp. ) The expansion is controlled by the norm of the generator ∥Sˆ∥, which, for typical magnitudes V and U of the inter-site couplings and on-site potentials, is of order V /U (see Supp.

0778 and 0.0782 across the Brillouin zone. p. 0
Where the paper says it

Beyond reinterpreting the formalism, these results sharpen the questions of how V should be computed and what the Hubbard subspaces fundamentally are. Density-functional theory (DFT) is a cornerstone of firstprinciples modeling across physics, chemistry, and materials science [1–5], but it fails for so-called “strongly correlated” materials — loosely, those whose ground-state wavefunction has no single dominant electronic configuration — for instance, wrongly predicting many transition metal oxides (TMOs) to be metallic [6, 7].

The illustrative Hubbard projectors and potentials were obtained from DFT+U+V calculations performed with Quantum ESPRESSO. p. 7
Where the paper says it

Umari, and R. Wentzcovitch, QUANTUM ESPRESSO: A modular and opensource software project for quantum simulations of materials, J.

How the training signal is built

The authors propose a self-consistent way to parametrize V by enforcing linearity of the total energy in the occupations of these hybridized projectors, n II [S] = X i sigma = n II [hat I] + X i sigma + O(V 2/U 2). In this case, one would measure the total-energy curvature via response to perturbations of the form dv J sigma e Sphi J sigma j>⟨phi J sigma je-S>, which would give rise to response matrices chi IJ [Sˆ] = dnII [Sˆ]dv J ext, and one could follow the conventional linear-response strategy to calculate U for the hybridized subspace. p. 3

Where the paper says it

If +V is really +U on the hybridized projectors, the selfconsistent way to parametrize it would be to enforce linearity of the total energy in the occupations of these hybridized projectors, n II [Sˆ]. But that occupation responds through inter-site channels — both the response of the inter-site blocks n IJσ and the response to intersite perturbations — that the conventional response matrices do not contain (see Supp.

Where the gains come from

The quality of the frozen-projector approximation is not special to the Γ-point shown in Fig. p. 4

Where the paper says it

22 (d) V (0) + V (2) Ni (maj) Ni (min) O O (e) V (0) + V (2) VU + V 10 0 10 1 10 2 0 10 2 10 1 10 0 FIG. The frozen-projector approximation reproduces the exact DFT+U+V potential in bulk antiferromagnetic nickel oxide to within a small residual (panels a–c), which can be almost entirely accounted for by a single analytic second-order term (panels d, e).

2 of the main text. p. 5
Where the paper says it

6 APPENDIX Here we establish the central result of the main text — that, to first order in V /U, DFT+U+V is the DFT+U energy functional evaluated on a densitymatrix-dependent redefinition of the Hubbard projectors. The DFT+U and DFT+U+V corrections are most transparent in the basis {|φ Iσ i ⟩} that diagonalizes the onsite occupation matrices n IIσ, with eigenvalues λ Iσ i .

p. 5
Where the paper says it

6 APPENDIX Here we establish the central result of the main text — that, to first order in V /U, DFT+U+V is the DFT+U energy functional evaluated on a densitymatrix-dependent redefinition of the Hubbard projectors. The DFT+U and DFT+U+V corrections are most transparent in the basis {|φ Iσ i ⟩} that diagonalizes the onsite occupation matrices n IIσ, with eigenvalues λ Iσ i .

The authors conclude that even an ordinary DFT+U calculation with hybridized frozen projectors may stand in — approximately — for DFT+U+V, bringing the far broader ecosystem of DFT+U implementations and property extensions to bear. p. 4
Where the paper says it

In contrast, if this redefinition is attempted with frozen projectors, as in most current implementations, the equivalence becomes partial: the total energy and the spectrum then call for differently hybridized projectors (e Sˆ and e 2Sˆ , respectively), and no frozen-projector functional reproduces the DFT+U+V potential exactly (Supp. Even so, an ordinary DFT+U calculation with hybridized frozen projectors may stand in — approximately — for DFT+U+V, bringing the far broader ecosystem of DFT+U implementations and property extensions to bear.

The message to practitioners is that the choice of projectors, the choice of Hubbard parameters, and the choice of whether to apply inter-site corrections are deeply intertwined and should never be treated as independent degrees of freedom. p. 4
Where the paper says it

Even so, an ordinary DFT+U calculation with hybridized frozen projectors may stand in — approximately — for DFT+U+V, bringing the far broader ecosystem of DFT+U implementations and property extensions to bear. The message to practitioners of DFT+U and DFT+U+V is that the choice of projectors, the choice of Hubbard parameters, and the choice of whether to apply inter-site corrections are deeply intertwined and should never be treated as independent degrees of freedom.

The paper sharpens the question of what the Hubbard projectors fundamentally are by revealing that even established inter-site correction is, at heart, a choice of projectors. p. 4
Where the paper says it

The message to practitioners of DFT+U and DFT+U+V is that the choice of projectors, the choice of Hubbard parameters, and the choice of whether to apply inter-site corrections are deeply intertwined and should never be treated as independent degrees of freedom. More deeply — with DFT+U+V now embedded in community codes and automated high-throughput workflows [36, 59–62], increasingly the default extension of DFT+U — this work sharpens the question of what the Hubbard projectors fundamentally are, by revealing that even the established inter-site correction is, at heart, a choice of projectors.

p. 4
Where the paper says it

The message to practitioners of DFT+U and DFT+U+V is that the choice of projectors, the choice of Hubbard parameters, and the choice of whether to apply inter-site corrections are deeply intertwined and should never be treated as independent degrees of freedom. More deeply — with DFT+U+V now embedded in community codes and automated high-throughput workflows [36, 59–62], increasingly the default extension of DFT+U — this work sharpens the question of what the Hubbard projectors fundamentally are, by revealing that even the established inter-site correction is, at heart, a choice of projectors.

The authors suggest that an optimal set of projectors for DFT+U might be found by correcting the curvature with respect to localized orbitals. The paper also shows that the second-order correction V tilde(2)sigma instead couples projectors on distinct Hubbard sites I! p. 4
Where the paper says it

Computational details can be found in Supp. To conclude, the inter-site +V corrections of DFT+U+V are equivalent — to first order in V /U — to an on-site DFT+U correction evaluated on a densitymatrix-dependent set of Hubbard projectors hybridized with those of neighboring sites.

= J, with a strength set by V acting through an intermediate site K. p. 0
Where the paper says it

Typically, these orbitals are atom-centered, fixed, localized, orthonormal (as in the “ortho-atomic” scheme employed in this work), and correspond to the 3d or 4f subshell of a transition metal/lanthanide [15]. Meanwhile, U I is the Hubbard parameter for site I, and defines the strength of the Hubbard correction that will be applied to this particular subspace.

The first-order potential nonetheless takes exactly this U form is not automatic. p. 2
Where the paper says it

Equation (4) can be differentiated, upon which we find that, to first order, the DFT+U potential applied to these hybridized projectors reproduces the DFT+U+V potential. This is the central result of this work: in the limit V ≪ U, the inter-site +V correction is exactly a redefinition of the Hubbard projectors.

Pruning redundant reasoning tokens

The authors show that the second-order correction V tilde(2)sigma = -X IJij X Kk V IK Lambda IK sigma ik S KJ sigma kj phi I i><phi J j. This term is no longer a Dudarev-type +U correction, and the first-order term V tilde(0)sigma retained the on-site, projector-diagonal form of +U. p. 3

Where the paper says it

Figure 2 compares the exact DFT+U+V potential (panel a) with the first-order approximation V ˆ˜ (0)σ (panel b): the two agree to within a small residual (panel c), and even this residual is predicted analytically by the higher-order term V ˆ˜ (2)σ (panels d, e) — albeit at a cost: V ˆ˜ (2)σ no longer has the on-site +U form (see Supp. Note that the numerical superscript in V ˆ˜ (n)σ counts powers of Sˆ, and the first-order correction V ˆ˜ (1)σ is precisely the term that the defining relation of Sˆ eliminates.

The quality of the frozen-projector approximation is not special to the Γ-point shown in Fig. 2 of the main text. The paper sharpens the question of what the Hubbard projectors fundamentally are by revealing that even established inter-site correction is, at heart, a choice of projectors. This suggests that determining how projectors might be chosen optimally is now all the more pressing. The equivalence does not retroactively justify computing V via Eq. p. 3
Where the paper says it

S10 for details). In other words, the equivalence we have uncovered does not retroactively justify computing V as the off-diagonal elements of the response matrices used to compute U as is common practice [16]; instead, it compounds the pre-existing discrepancy between the site-diagonal occupations the response matrices measure and the inter-site blocks that EV penalizes.

(S3) because the parameter that is computed and the parameter that appears in the functional are conjugate to different occupations. p. 0
Where the paper says it

Meanwhile, U I is the Hubbard parameter for site I, and defines the strength of the Hubbard correction that will be applied to this particular subspace. Because the correction depends on the KohnSham orbitals via the projected density matrix, DFT+U is an orbital-dependent, generalized Kohn-Sham theory.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Adaptive Hubbard Projector Selection: The AI system can dynamically select Hubbard projectors based on density matrix information to minimize energy curvature, as shown by to first order in V /U, DFT+U+V is exactly on-site DFT+U evaluated on a redefinition of the Hubbard projectors.

  2. Covalency-Aware Potential Calculation: The system can calculate inter-site corrections using response matrices that include off-diagonal blocks, moving beyond the conventional approach where the off-diagonal elements of the same response matrices used to compute U as is common practice are used.

  3. Optimized Parameter Computation: The AI can compute Hubbard parameters U and V using linear response theory, specifically by measuring the total energy curvature via response to perturbations of the form dvJextXj sigma eShatphiJsigma j⟩⟨phiJsigma j instead of relying solely on site-diagonal occupation responses.

  4. Frozen Projector Approximation for Speed: The system can utilize frozen projectors, which allows for a faster calculation where the frozen contribution to the potential is deltaEU [alphaShat(rho)] = Vhat sigma U + alpha Vhat sigma V + O(V 2/U), despite it being short of the DFT+U+V potential by the on-site operator What sigma.

  5. Higher Order Correction Incorporation: The system can incorporate second-order corrections, represented by the second-order correction Vtilde (2)sigma = -XIJij Xkkk VIKLambdaIKsigmaik S K Jsigma kj - S I Ksigmaik V KJLambdaKJsigma kj, to improve accuracy beyond the first order.

Abstract

Hubbard-corrected density-functional theory (DFT+ U) is a popular tool for first-principles modeling of materials with localized d or f electrons, but its on-site corrections tend to over-localize charge and break covalent bonds. Inter-site +V corrections were introduced to counter this and are now widely used, but a formal justification has been lacking. Here we show that -- to first order in V/U -- inter-site corrections are exactly equivalent to on-site DFT+ U evaluated on a density-matrix-dependent redefinition of the Hubbard projectors, hybridized with those of neighboring sites, providing insight into the explicit mechanism by which V affects covalency. If the projectors are held frozen, as is common practice, the equivalence partially breaks down. Beyond reinterpreting the formalism, these results sharpen the questions of how V should be computed and what the Hubbard subspaces fundamentally are.

Related papers