---
title: Local Mean-Field Density Functional Theory
url: https://www.emergentmind.com/topics/local-mean-field-density-functional-theory
type: topic
---

# Local Mean-Field Density Functional Theory

Searching arXiv for recent and foundational papers on local mean-field density functional theory and closely related formulations.
Local mean-field density functional theory denotes a class of density-functional constructions in which inhomogeneous many-body systems are represented through local densities, local order parameters, or local effective one-body fields, while interaction effects are treated by mean-field, local-density, or auxiliary-field closures. In superfluid fermionic systems this viewpoint is realized by the Superfluid Local Density Approximation (SLDA), described as a “fully local, orbital-based density functional theory for superfluid fermions” [1204.2207]. In classical non-uniform liquids, the Local Molecular Field (LMF) equation for an effective reference potential follows directly from the standard mean-field DFT treatment of attractive forces [1211.7321]. In correlated lattice systems, local exchange-correlation potentials derived from Dynamical Mean Field Theory (DMFT), as well as density-and-pair-density functionals, provide local Kohn–Sham-like descriptions of Hubbard models [1004.2264], [1203.6873]. A rigorous local-density approximation (LDA) theorem for short-range classical gases further establishes when the free energy with prescribed profile $\rho(x)$ is approximated by $\int f_T(\rho(x))\,dx$ [2310.18028].

## 1. Common formal structure

Across its major realizations, local mean-field DFT begins from a functional of one-body information and replaces either explicit many-body correlations or nonlocal interaction terms by local functionals, effective local fields, or local auxiliary equations. In classical DFT one writes the grand potential as
\[
\Omega[\rho]=F_{\rm id}[\rho]+F_{\rm ex}[\rho]+\int dr\,\rho(r)\bigl(\phi_{\rm ext}(r)-\mu\bigr),
\]
with the Euler–Lagrange equation
\[
\ln[\Lambda^3\rho(r)]-c^{(1)}(r)+\phi_{\rm ext}(r)=\mu,
\]
where $c^{(1)}(r)=-\delta F_{\rm ex}[\rho]/\delta\rho(r)$ [1211.7321]. In quantum formulations, one instead minimizes an energy functional with respect to orbitals, quasiparticle amplitudes, or generalized densities, obtaining Kohn–Sham, Bogoliubov–de Gennes, or collective Schrödinger equations [1204.2207], [1301.0807].

The local variables differ substantially by domain, but the operational motif is recurrent: a uniform reference problem supplies either the free-energy density, the exchange-correlation energy, the pairing coupling, or a kinetic-renormalization factor; the inhomogeneous system is then treated through pointwise evaluation or local self-consistency. This suggests that “local” in this literature refers primarily to the representation of the functional and the resulting equations, not necessarily to the bare interaction itself.

| Formulation | Local variables or fields | Systems emphasized |
|---|---|---|
| LMF–DFT | $\rho(r)$, $\phi_R(r)$ | Non-uniform liquids [1211.7321] |
| Classical LDA | $\rho(x)$, $f_T(\rho)$ | Short-range classical gases [2310.18028] |
| SLDA / ASLDA | $\rho(r)$, $\tau(r)$, $\nu(r)$, $\Delta(r)$ | Unitary Fermi gas, nuclei [1204.2207], [1008.3933] |
| DMFT-based lattice DFT | $n_i$, $v_{xc}^{DMFT}(n)$ | 3D Hubbard model [1004.2264] |
| DPDFT | $n_x$, $d_x$, $z(n,d)$ | Inhomogeneous Hubbard model [1203.6873] |
| Collective-coordinate DFT | $D(q,r)=|f(q)|^2 d(q,r)$ | Symmetry breaking and configuration mixing [1301.0807] |
| Local DFT-SIR-DH | $\rho_i(x)$, $\psi_{ij}^{(n)}(x)$ | Epidemic spreading with social interactions [2509.05979] |

## 2. Classical non-uniform liquids and the LMF–DFT correspondence

The clearest classical local mean-field construction is the reformulation of Local Molecular Field Theory within DFT. The pair potential is split into a reference part and a remainder,
\[
w(r)=u_0(r)+u_1(r),
\]
and a standard mean-field approximation is made for the excess free-energy difference,
\[
F_{\rm ex}[\rho;w]-F_{\rm ex}^R[\rho;u_0]\approx \frac12\int dr\,dr'\,\rho(r)\rho(r')u_1(|r-r'|).
\]
Substitution into the Euler–Lagrange equation yields the effective reference field
\[
\phi_R(r)=\phi_{\rm ext}(r)+\int dr'\,\rho(r')u_1(|r-r'|)+{\rm const.},
\]
or, equivalently,
\[
\phi_R(r)-\phi_{\rm ext}(r)=-c^{(1)}[\rho;w](r)+c^{(1)}[\rho;u_0](r).
\]
In this form, LMF is not an external alternative to DFT but a restatement of the usual mean-field DFT treatment of attractions in terms of an effective one-body field [1211.7321].

Archer and Evans illustrated this correspondence for a hard-core Yukawa liquid adsorbed at a planar hard wall, using Rosenfeld’s Fundamental-Measure Theory as the hard-sphere reference functional. The self-consistent procedure is explicit: start from a trial $\phi_R(z)$, solve the one-dimensional Euler–Lagrange equation for the hard-sphere profile $\rho(z)$, update $\phi_R(z)$ through the LMF equation involving the laterally integrated Yukawa tail, and iterate to self-consistency. The resulting density profile and thermodynamics are identical to those obtained by direct minimization of a DFT functional with
\[
F_{\rm ex}\approx F_{HS}[\rho]+\frac12\iint \rho\rho\,u_1
\]
[1211.7321].

Near bulk liquid–gas coexistence, the same framework predicts a repulsive hump in $\phi_R(z)$ immediately outside the hard wall, of height $O(k_BT)$, generating the low-density gas film characteristic of complete drying. For sufficiently long-range attraction the tail of $\phi_R(z)$ decays monotonically, whereas for shorter-range attractions it exhibits damped oscillations, reflecting the Fisher–Widom crossover in the asymptotic decay of $g(r)-1$. The one-dimensional exactly solvable model studied in the same work shows that the choice of splitting matters quantitatively: placing only a small fraction $\lambda$ of the attractive tail into the mean-field term, with $\lambda\lesssim 0.1$, yields very accurate pressure and structure factor, whereas the standard RPA choice $\lambda=1$ does not [1211.7321]. A persistent misconception is therefore that locality alone guarantees accuracy; in the LMF setting, the decisive ingredient is the “intelligent division” of the pair potential.

## 3. Local-density approximation as a controlled limit

A complementary route to local mean-field DFT is the LDA itself. For a short-range classical gas at temperature $T$, the free energy with prescribed density profile can be approximated by the local functional
\[
G_T^{\rm LDA}[\rho]=\int f_T(\rho(x))\,dx,
\]
where $f_T(\rho_0)$ is the free energy per unit volume of the infinite homogeneous gas of density $\rho_0$ [2310.18028]. The rigorous result applies under assumptions of superstability, stability, and lower-regularity of the interaction, together with slow spatial variation of the density.

The theorem is quantitative. Introducing the local variation function
\[
\delta\rho_\ell(z):=\ell^{-1}\,\operatorname*{ess\,sup}_{x,y\in z+[-\ell/2,\ell/2]^d}|\rho(x)-\rho(y)|,
\]
one obtains upper and lower bounds for $G_T[\rho]-\int f_T(\rho(x))\,dx$ in terms of $\int \sqrt{\rho}$ and $\ell^{bp}\int(\delta\rho_\ell)^p$, multiplied by an explicit decay factor $\eta_\ell$. The proof combines a Legendre-dual representation, coarse-graining into cubes, local thermodynamic-limit comparison in “main” cubes, and quantitative Ruelle-type bounds controlling local number fluctuations and inter-cube interactions [2310.18028].

This result sharpens the status of LDA in local mean-field theories. It is not merely a heuristic replacement of a nonlocal functional by a pointwise one; for slowly varying profiles it becomes an asymptotic statement with controlled errors. The same analysis indicates where the approximation should fail or need refinement: nonlocal corrections must be tied to density variation and are naturally expressed as gradient terms of the form
\[
\int K(\rho)|\nabla\rho|^2\,dx,
\]
whose appearance is described in the discussion of refined cluster-expansion or gradient-expansion arguments [2310.18028].

## 4. Superfluid fermions and the Superfluid Local Density Approximation

In quantum many-body physics, the canonical local mean-field density functional is SLDA. Its basic local densities are
\[
\rho(r)=\sum_n |v_n(r)|^2,\qquad
\tau(r)=\sum_n |\nabla v_n(r)|^2,\qquad
\nu(r)=\sum_n u_n(r)v_n^*(r),
\]
and the total energy is written as
\[
E[\rho,\nu]=\int d^3r\,\mathcal E[\rho(r),\tau(r),\nu(r)].
\]
The local superfluid energy density contains kinetic, particle-hole, and pairing parts. In the unitary Fermi gas (UFG), the unrenormalized SLDA functional takes the form
\[
\mathcal E_{SLDA}^{UFG}
=\frac{\hbar^2}{m}\left[
\frac{\alpha}{2}\tau(r)
+\beta\,\frac{3(3\pi^2)^{2/3}}{10}\rho^{5/3}(r)
+\gamma\,\frac{|\nu(r)|^2}{\rho^{1/3}(r)}
\right]
+V_{\rm ext}(r)\rho(r),
\]
while in nuclear applications the particle-hole and pairing parts are generalized to isospin-dependent functionals involving $\rho_n,\rho_p,\tau_n,\tau_p,\nu_n,\nu_p$ [1204.2207].

A central structural feature is ultraviolet renormalization. Both $\tau(r)$ and $\nu(r)$ diverge, but the combination
\[
\left(\frac{\hbar^2}{2m_{\rm eff}(r)}\right)\tau(r)-\Delta(r)\nu^*(r)
\]
is finite, with $\Delta(r)=\delta E/\delta\nu^*(r)$ the local pairing field. Because the pairing energy is proportional to $|\nu|^2$, SLDA yields the local gap equation
\[
\Delta(r)=-g_{\rm eff}(r)\,\nu(r),\qquad
g_{\rm eff}(r)^{-1}=g_0^{-1}-\Lambda(r),
\]
so the order parameter is purely local after renormalization [1204.2207].

Variation of $E-\mu N$ gives the Kohn–Sham–Bogoliubov equations
\[
\Bigl(-\nabla\cdot\frac{\hbar^2}{2m_{\rm eff}(r)}\nabla + U(r)-\mu\Bigr)\,u_n(r)
+\Delta(r)\,v_n(r)=E_n\,u_n(r),
\]
\[
\Delta^*(r)\,u_n(r)
-\Bigl(-\nabla\cdot\frac{\hbar^2}{2m_{\rm eff}(r)}\nabla + U(r)-\mu\Bigr)\,v_n(r)
=E_n\,v_n(r),
\]
with $U(r)=\delta E/\delta\rho(r)$. In the UFG, the three dimensionless constants $(\alpha,\beta,\gamma)$ are fixed by matching infinite-matter QMC data for the energy per particle, the pairing gap, and the quasiparticle spectrum. A related detailed SLDA fit reports $\xi=0.40(1)$, $\eta=\Delta_0/\varepsilon_F=0.50(2)$, $\alpha=1.094(17)$, $\beta=-0.526(18)$, and $\gamma^{-1}=-0.091(9)$ for the homogeneous system, while the 2012 review summarizes the homogeneous limit as recovering QMC energy with $\xi\approx0.37$ and gap $\Delta\approx0.5\,\varepsilon_F$ [1008.3933], [1204.2207].

Locality in SLDA is an approximation, not an exact microscopic statement. The pairing channel is taken to be zero-range, justified because typical Cooper-pair sizes in dilute or nuclear matter greatly exceed the microscopic interaction range, and gradient corrections are omitted at LDA level because numerics and QMC benchmarks show them to be small in the UFG [1204.2207]. The asymmetric extension ASLDA promotes $\alpha,\beta,\gamma$ to polarization-dependent functions and is used when $n_a\neq n_b$, while the time-dependent extension TD-SLDA introduces a time-dependent action and a Galilean-invariant replacement of $\tau_\sigma$ by $\tilde\tau_\sigma$ to describe non-adiabatic dynamics, superfluid-to-normal transitions, and modes inaccessible in Landau–Ginzburg, Gross–Pitaevskii, or quantum hydrodynamics [1008.3933].

## 5. Strongly correlated lattice formulations

For the three-dimensional Hubbard model, a local mean-field density-functional scheme can be built by extracting the exchange-correlation potential from DMFT. The lattice DFT energy functional is
\[
E[n]=T_0[n]+\frac14\sum_i U_i n_i^2+E_{xc}[n]+\sum_i\epsilon_i n_i,
\]
and in LDA one sets
\[
E_{xc}[n]\approx \sum_i E_{xc}^{\rm hom}(n_i),\qquad
v_{xc}(n)=\frac{dE_{xc}^{\rm hom}(n)}{dn}.
\]
Here $E_{xc}^{\rm hom}(n)$ is taken from the homogeneous 3D Hubbard model via DMFT,
\[
E_{xc}^{\rm hom}(n)=E_{DMFT}(n)-T_0(n)-\frac14Un^2.
\]
When $U$ exceeds a critical $U_c^{Mott}\approx14$, $E_{xc}^{\rm hom}(n)$ develops a cusp at $n=1$, and the derivative $v_{xc}^{DMFT}(n)$ jumps by a finite amount $\Delta_{xc}(U)>0$, providing the DFT signature of the Mott–Hubbard metal–insulator transition at half-filling [1004.2264].

The time-dependent counterpart uses the adiabatic Local Density Approximation. The Kohn–Sham orbitals satisfy
\[
\bigl[\hat T+\hat v_{KS}(\tau)\bigr]\varphi_\kappa(\tau)=i\,\partial_\tau\varphi_\kappa(\tau),
\]
with
\[
v_{KS}(i,\tau)=\frac12U_i n_i(\tau)+v_{xc}^{DMFT}(n_i(\tau))+\epsilon_i+w_i(\tau).
\]
This construction is explicitly memory-free: the static $v_{xc}^{DMFT}(n)$ is reused at the instantaneous density. For a $5^3$-site cluster with a single interacting impurity at the center, the agreement with exact real-time densities is reported as excellent for moderate $U$ and smooth pulses; near half-filling and across the discontinuity at $n=1$, non-local and memory effects grow important. In Bloch-oscillation simulations on a $33\times5\times5$ lattice, the ALDA-DMFT dynamics reproduces $\omega\approx F$ and interaction-induced beat notes at $\omega\pm U$, but does not capture damping because genuine memory effects are absent [1004.2264].

A more elaborate local many-body embedding is the self-energy self-consistent DFT+DMFT scheme. There the full Green’s function is
\[
G(r,r';i\omega_n)=\langle r|[i\omega_n+\mu-\hat H_{KS}-\Delta\hat\Sigma(i\omega_n)]^{-1}|r'\rangle,
\]
and the DMFT self-energy is linearized near $\omega=0$ and inserted into the Kohn–Sham exchange-correlation potential for the correlated subspace. The scheme sets $\Sigma^{dc}(k)=\Sigma^{lin}(k)$, which removes the usual double-counting ambiguity. In SrVO$_3$, the method yields $Z\approx0.48$, keeps the $t_{2g}$ quasiparticle bandwidth and Hubbard peaks at essentially the same positions as in one-shot calculations, and pushes the O $2p$ bands down to $-5.0\,{\rm eV}$ (6 orbitals) and $-6.1\,{\rm eV}$ (3 orbitals), in excellent agreement with photoemission, without any ad hoc $p-d$ shift [1904.02967]. Although this framework is more elaborate than an ordinary LDA, it remains local in the DMFT sense: the correlated physics enters through a local self-energy and its local feedback into the charge-density cycle.

## 6. Beyond ordinary density: pair density and collective coordinates

One route beyond ordinary local-density functionals is to enlarge the local variable set. In the adaptive pair-density functional theory of the inhomogeneous Hubbard model, the basic variables are the local density $n_x$ and the local double occupancy $d_x=\langle n_{x\uparrow}n_{x\downarrow}\rangle$. After a Legendre transform in the site-dependent interaction strengths $U_x$, one obtains the exact functional
\[
E_{Uv}[\{d_x,n_x\}]=T[\{d_x,n_x\}]+\sum_x U_x d_x+\sum_x v_x n_x.
\]
The interacting kinetic energy is written through a reduction factor,
\[
\rho_{x,x\pm1}=q_{x,x\pm1}[\{n,d\}]\,\rho^0_{x,x\pm1},
\]
leading to Kohn–Sham-like equations with effective hoppings $t_{x,x\pm1}^{eff}=t\,q_{x,x\pm1}[\{n,d\}]$. Under the site-factorized ansatz
\[
q_{x,x\pm1}\simeq z(n_x,d_x)\,z(n_{x\pm1},d_{x\pm1}),
\]
the functional becomes formally identical to the Gutzwiller approximation, but the renormalization function $z(n,d)$ is fixed from the exact uniform Bethe-ansatz solution through
\[
z^2(\rho,d)\,T_0(\rho)\equiv T^{BA}(\rho,d)
\]
[1203.6873].

The numerical consequences are substantial. For a periodic one-dimensional chain of length $L=16$ with $N=6$ electrons and an alternating binary potential $v_x=(-1)^xV$, the per-site energy error remains below a few percent in DPDFT-FA even for $V\approx4t$, whereas the LDA error can exceed $50\%$ in the strongly correlated regime $U/t\gg1$ and large $V$. In the single-electron impurity problem, exact physics requires $d_x\equiv0$; LDA instead produces a spurious self-interaction energy saturating at $U/4$, while DPDFT-FA drives $d_x\to0$ and hence $E_{SI}\to0$ as the impurity potential deepens [1203.6873]. The improvement is therefore tied to local adaptivity of pair density, not merely to a better scalar energy fit.

A second extension generalizes locality into a mixed single-particle/collective space. Defining the generalized density
\[
D(q,r)=N\int d^{3N}R\,\delta^{(3)}(r-r_1)\,\Psi^*(R)\,\hat P(q,R)\,\Psi(R),
\]
one factorizes it as
\[
D(q,r)=|f(q)|^2\,d(q,r),
\]
where $f(q)$ is a collective wave function and $d(q,r)$ a normalized density slice satisfying $\int d^3r\,d(q,r)=N$. The exact energy can then be written as
\[
E[f,d]=\int d^n q\, f^*(q)\Bigl[-\frac12\sum_{\mu\nu}\partial_\mu \mathcal A_{\mu\nu}(q)\partial_\nu+\mathcal U(q)\Bigr]f(q),
\]
up to the rearrangement term. Minimization yields a local Schrödinger equation for $f(q)$ and, at each $q$, Kohn–Sham equations
\[
\Bigl[-\frac12\Delta+v_{\rm ext}(r)+v_s(q;r)-\sum_\mu \lambda_\mu Q_\mu(r)\Bigr]\phi_i(q;r)=\varepsilon_i(q)\phi_i(q;r),
\]
with $v_s(q;r)=\delta\mathcal U^{ic}[f,d](q)/\delta d(q,r)$ [1301.0807]. The translational example recovers $\mathcal A_{\mu\nu}(q)=\delta_{\mu\nu}/N$ and clean separation of center-of-mass and intrinsic dynamics, showing that locality can be formulated in an augmented collective manifold rather than only in ordinary real space.

## 7. Benchmarks, extensions, and recurring limitations

Local mean-field DFT methods are sustained by benchmark comparisons to controlled calculations. In the UFG, SLDA reproduces QMC energies in spherical traps with $(N_\uparrow,N_\downarrow)$ up to $11$ and errors $\lesssim5\%$; sample values include $(2,1)$ with QMC $=4.281\pm0.004$ and SLDA $=4.417$ ($3.2\%$), $(3,2)$ with QMC $=7.61\pm0.01$ and SLDA $=7.602$ ($0.1\%$), and $(6,6)$ with QMC $=21.278\pm0.05$ and SLDA $=21.13$ ($0.7\%$). Time-dependent SLDA has been used for Anderson–Higgs oscillations of $\Delta(t)$, Bogoliubov sound modes, vortex crossing and reconnection, quantum turbulence, and quantum shock waves and domain walls in colliding clouds. In nuclear systems, the same framework describes odd–even mass staggering across 200+ nuclei with a single volume pairing strength, vortex cores in neutron matter, and giant dipole resonances in deformed, open-shell heavy nuclei in excellent agreement with experiment and with no adjustable parameters [1204.2207].

The same localization logic has recently been exported beyond traditional condensed-matter and nuclear settings. A local mean-field density functional theory model for epidemic spreading replaces convolution kernels for social interactions by a sum-of-exponentials approximation and auxiliary Debye–Hückel equations,
\[
-\Delta\psi_{ij}^{(n)}+\kappa_{ij}^{(n)\,2}\psi_{ij}^{(n)}=\rho_j(x),
\]
thereby converting a nonlocal DDFT system into a local DFT-SIR-DH PDE system. Its linear stability analysis yields both the classical reaction-driven instability condition $\lambda S^*>\gamma$ and a cross-interaction condition for energy-driven instability; in a simplified self-repulsive case, the theory proves the existence of a unique global-in-time classical solution remaining nonnegative and bounded [2509.05979]. This suggests that the local mean-field DFT idiom is portable wherever convolution-dominated interactions can be recast through local auxiliary fields.

Several limitations recur across the literature. In classical LMF, the split $w=u_0+u_1$ is not innocuous, and the one-dimensional solvable model shows that treating too much attraction in mean field leads to unphysical van der Waals loops or negative compressibility [1211.7321]. In SLDA, omitted gradient corrections are small in the UFG at LDA level, but the approximation is expected to degrade in very tight traps or strong inhomogeneities [1204.2207]. In ALDA-DMFT, the absence of memory precludes damping of Bloch oscillations [1004.2264]. In DPDFT, the factorized-$q$ ansatz is approximate except in the infinite-coordination Gutzwiller limit [1203.6873]. The rigorous LDA theorem also clarifies that locality has a definite asymptotic regime—slowly varying densities and short-range interactions—rather than universal validity [2310.18028].

Taken together, these developments define local mean-field density functional theory not as a single formalism but as a family of mathematically related strategies: one starts from local densities or generalized local variables, imports uniform-system many-body information, and closes the inhomogeneous problem through local effective fields, local self-energies, local pair densities, or local auxiliary PDEs. The common payoff is computational tractability with direct access to structure, thermodynamics, and, in several cases, real-time dynamics; the common challenge is to control the information lost when genuinely nonlocal or history-dependent correlations are compressed into a local representation.

Source: https://www.emergentmind.com/topics/local-mean-field-density-functional-theory