Half-Filled Single-Band Hubbard Model

• The half-filled single-band Hubbard model is a fundamental lattice Hamiltonian that captures electron correlation effects, including Mott transitions and antiferromagnetism.
• It employs analytic and numerical techniques such as DMFT, AFQMC, and strong-coupling methods to delineate spectral features and phase coexistence boundaries.
• The model’s flexibility across different lattices and symmetry conditions clarifies how band topology and interaction strengths govern metal-insulator transitions.

The half-filled single-band Hubbard model is defined by a one-orbital lattice Hamiltonian of the form

$$H=-t\sum_{\langle i,j\rangle,\sigma}\left(c^\dagger_{i\sigma}c_{j\sigma}+\mathrm{h.c.}\right)+U\sum_i n_{i\uparrow}n_{i\downarrow}-\mu\sum_i n_i,$$

with average density $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$ at half filling; on particle-hole-symmetric bipartite lattices this is commonly implemented by $\mu=U/2$ (Sherman, 2014). In the SU$(2M)$-symmetric generalization, the ordinary spin-$\tfrac12$ one-band model is exactly the $M=1$ case, so the familiar SU(2) problem appears as the first member of a broader family of half-filled Hubbard systems (Blümer et al., 2012). Across the works considered here, the half-filled single-band problem is the common setting in which paramagnetic Mott transitions, local-moment formation, antiferromagnetism, and momentum-resolved excitation structure are analyzed in one, two, and three dimensions.

1. Hamiltonian, half-filling, and basic observables

In the standard formulation, the model contains a nearest-neighbor kinetic term with hopping amplitude $t$, an onsite repulsion $U>0$, and either an explicit chemical-potential term or an equivalent canonical half-filling constraint (Sherman, 2015). Half filling means one electron per site on average, $n=1$, or equivalently half occupation of the full single-particle Hilbert space for a single orbital per site (Ixert et al., 2014). On bipartite particle-hole-symmetric lattices, several works use $\mu=U/2$ as the half-filled condition, including square-lattice strong-coupling treatments, ionic extensions, and slave-particle analyses of the cubic-lattice problem (Sherman, 2014).

The core observables used to characterize the half-filled problem recur across methods. For Mott physics, the quasiparticle weight

$n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$0

or its Matsubara estimate

$n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$1

distinguishes metallic and insulating solutions in DMFT (Blümer et al., 2012). The double occupancy

$n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$2

tracks charge fluctuations and enters directly in the interaction energy $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$3 in the one-band case (Blümer et al., 2012). Magnetic diagnostics include the local spin susceptibility, the effective local moment $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$4, staggered structure factors, and local moments of the form $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$5 (Mazitov et al., 2021).

A persistent technical point is that the phrase “half-filled single-band Hubbard model” does not specify the lattice, dimensionality, or approximation. The square lattice, Bethe lattice, simple cubic lattice, decorated honeycomb lattice, and vortex-full square lattices all realize half-filled single-band Hubbard physics, but their weak-coupling band structures and strong-coupling spin sectors differ substantially (Nourse et al., 2022).

2. Paramagnetic Mott transition and coexistence structure

Within DMFT, the half-filled one-band problem exhibits the standard paramagnetic Mott metal-insulator transition: a correlated metal at weak coupling, a paramagnetic Mott insulator at strong coupling, and for $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$6 a first-order coexistence region bounded by $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$7 and $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$8, ending at a second-order critical endpoint $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$9 (Blümer et al., 2012). A crucial distinction is that these coexistence boundaries are not the same as the actual first-order line $\mu=U/2$0; locating the latter requires a free-energy comparison (Blümer et al., 2012).

For the single-band case $\mu=U/2$1 in the Bethe-lattice unit-variance convention, the fitted SU$\mu=U/2$2 scaling formulas give $\mu=U/2$3, $\mu=U/2$4, $\mu=U/2$5, and $\mu=U/2$6 (Blümer et al., 2012). The broader significance of the SU$\mu=U/2$7 analysis is that the half-filled SU(2) model is not exceptional: after rescaling with $\mu=U/2$8, $\mu=U/2$9, and $(2M)$0, the coexistence boundaries for different orbital degeneracies collapse onto a nearly universal phase diagram, with the single-band problem as the $(2M)$1 endpoint (Blümer et al., 2012).

Cluster self-energy-functional approaches recover the same low-temperature first-order Mott phenomenology in two dimensions. Finite-temperature CDIA for the half-filled square-lattice model yields a discontinuous paramagnetic metal-insulator transition with a coexistence region persisting down to zero temperature and a critical endpoint near $(2M)$2 (Seki et al., 2018). In that framework, the metallic spectrum contains a central quasiparticle band flanked by Hubbard bands, whereas the insulating solution loses the coherent low-energy feature (Seki et al., 2018).

Strong-coupling diagram techniques on the half-filled square lattice also produce a Mott transition, but with approximation-dependent critical scales and a more local description of the self-energy. One such calculation finds $(2M)$3 for the nearest-neighbor square-lattice model, while a related strong-coupling treatment reports $(2M)$4 with $(2M)$5 for the square-lattice bandwidth (Sherman, 2014). These approaches reproduce broad Hubbard-band formation and a self-consistent Mott gap, but their reliability is best for $(2M)$6 and temperatures high enough that magnetic ordering and spin correlations are suppressed (Sherman, 2015).

3. Local moments, antiferromagnetism, and thermal scales

The half-filled one-band problem supports more than one thermal crossover, and the local-correlation scales need not coincide. In DMFT for the half-filled two-dimensional square-lattice model, the local charge susceptibility $(2M)$7 first increases as coherence develops, then reaches a local maximum $(2M)$8, which marks the beginning of local-moment formation; on further cooling, $(2M)$9 reaches a local minimum $\tfrac12$0, which coincides with the minimum of double occupancy and the low-temperature edge of the $\tfrac12$1 plateau and is interpreted as full local-moment formation (Mazitov et al., 2021). Below $\tfrac12$2, screening sets in, but full Kondo/Fermi-liquid behavior occurs only below the smaller scale $\tfrac12$3 (Mazitov et al., 2021).

This decomposition refines the usual half-filled Hubbard narrative. The onset of local moments, the temperature where moments are fully formed, and the temperature where they are fully Kondo screened are distinct. Near the Mott transition, the line $\tfrac12$4 approaches the Mott endpoint and tracks the quasiparticle-peak width through $\tfrac12$5, while the fingerprint criterion based on the lowest Matsubara-frequency block tends to overestimate $\tfrac12$6 away from the transition (Mazitov et al., 2021).

Antiferromagnetic order adds further structure. Finite-temperature VCA on the half-filled square lattice finds a Néel temperature that peaks around $\tfrac12$7 and then decreases as $\tfrac12$8 in strong coupling, but the existence of a finite $\tfrac12$9 in two dimensions is explicitly identified as a cluster mean-field artifact rather than a literal statement about the thermodynamic 2D model (Seki et al., 2018). The same study constructs a crossover diagram in the $M=1$0-plane separating Slater-type and Mott-type insulators on thermodynamic grounds (Seki et al., 2018).

In three dimensions, a slave-fermion/Schwinger-boson treatment of the half-filled cubic-lattice Hubbard model rewrites the low-energy sector as an effective antiferromagnetic Heisenberg model with a reduced effective spin

$M=1$1

so that increasing $M=1$2 enhances empty and doubly occupied sites, reduces $M=1$3, and suppresses $M=1$4 (Karchev, 2012). In that approach, the localized-electron regime corresponds to $M=1$5, the critical effective spin is $M=1$6, and the quantum critical point where $M=1$7 occurs at $M=1$8 (Karchev, 2012).

Dilution by setting $M=1$9 on a fraction of sites further separates magnetic and insulating scales. On the simple cubic lattice at half filling and $t$0 on the correlated sites, the metal-insulator crossover scale $t$1 drops rapidly as the interacting-site fraction $t$2 is reduced from 1 to 0.7, whereas $t$3 remains nearly unchanged over the same interval (Chakraborty et al., 2021). For $t$4, $t$5 and $t$6 coincide and are suppressed together, and long-range AF order survives down to $t$7, well below the classical simple-cubic site-percolation threshold $t$8, because the nominally noninteracting sites develop induced local moments and participate in the staggered network (Chakraborty et al., 2021).

4. Single-particle spectra and excitation structure

The half-filled single-band Hubbard model displays sharply different excitation phenomenology in one and two dimensions. In the pure one-dimensional limit, exact-diagonalization spectra with twisted boundary conditions reproduce the standard half-filled Hubbard signature of spin-charge separation: the spectral function contains spinon and holon bands, which on finite systems appear as interlaced stripes that evolve toward distinct continua in the thermodynamic limit (Xu et al., 2024). In that setting, the pure Hubbard chain corresponds to the $t$9, $U>0$0 limit of the extended Peierls-Hubbard model, and the spinon/holon structure is gradually lost when dimerization or strong nearest-neighbor repulsion is added (Xu et al., 2024).

The one-dimensional Holstein-Hubbard analysis is consistent with the same half-filled Hubbard benchmark on the repulsive side: in the Mott phase the charge sector is gapped while spin excitations remain gapless, exactly as in the ordinary repulsive half-filled Hubbard chain (Hohenadler et al., 2012). That result is used there as the reference point for identifying the additional Luther-Emery metallic phase and Peierls phase generated by retardation and electron-phonon coupling (Hohenadler et al., 2012).

In two dimensions, the low-energy structure is strongly momentum dependent. AFQMC plus constrained stochastic analytic continuation for the half-filled square lattice at $U>0$1 finds that the minimum single-particle gap occurs at the four $U>0$2 points, while the X points $U>0$3 lie slightly higher in energy, with

$U>0$4

in the thermodynamic extrapolation (Schumm et al., 3 Apr 2025). Around $U>0$5, the dispersion is anisotropic and quadratic; around X, it is quartic and produces a DOS singularity $U>0$6, so the DOS consists of a nearly flat ledge between $U>0$7 and $U>0$8 followed by a sharp peak at $U>0$9 (Schumm et al., 3 Apr 2025). The fraction of quasiparticle states in the ledge is summarized as $n=1$0 (Schumm et al., 3 Apr 2025).

Local strong-coupling approximations produce a different but still informative spectral viewpoint. In the half-filled square-lattice strong-coupling diagram technique, the spectral maxima remain close to the Hubbard-I pole positions

$n=1$1

while the self-consistent treatment broadens them into finite-width continua and opens a Mott gap when $n=1$2 (Sherman, 2014). This makes explicit that much of the gross dispersion is already Hubbard-I-like, whereas the main gains of the self-consistent strong-coupling treatment are spectral broadening, weight redistribution, and a finite critical interaction (Sherman, 2014).

5. Lattice deformations and related one-band variants

Several one-band deformations of the half-filled problem clarify which aspects of the standard model are robust and which depend sensitively on band structure or explicit symmetry breaking. In the half-filled ionic Hubbard model, a staggered potential $n=1$3 converts the weak-coupling limit into a band insulator rather than a metal. DMFT on the Bethe lattice finds that for finite $n=1$4 and at zero temperature, antiferromagnetic order turns on beyond a threshold $n=1$5 through a first-order transition; just after AF order appears there is a half-metal phase, followed at larger $n=1$6 by an AF insulator, and at finite temperature the CTQMC phase diagram contains a line of tricritical points separating first- and second-order transition surfaces (Bag et al., 2015). In the SU($n=1$7) ionic generalization, the Gutzwiller/RMFT treatment yields a critical interaction $n=1$8 where charge fluctuations freeze into sublattice occupations $n=1$9 and $\mu=U/2$0, the quasiparticle weight vanishes, and the low-energy theory becomes an SU($\mu=U/2$1) Heisenberg model with conjugate representations; for $\mu=U/2$2, the usual half-filled ionic Hubbard model is recovered (Wang et al., 2022).

If staggered potentials and spin-asymmetric hopping are both present, the weak- and strong-coupling phases reorganize further. In the mass-imbalanced ionic Hubbard chain, mean-field theory finds that period-two charge order and alternating spin density coexist for all $\mu=U/2$3; charge order dominates below a critical $\mu=U/2$4, staggered spin order above $\mu=U/2$5, and the transition is weakly first-order because a would-be critical metallic point is avoided by a jump of one gap parameter from positive to negative (Sekania et al., 2017).

Other lattice geometries produce qualitatively different half-filled one-band Mott transitions. On the decorated honeycomb lattice, where the half-filled noninteracting Fermi energy lies in a flat-band-derived degeneracy, RISB yields two distinct first-order interaction-driven insulating states: a $\mu=U/2$6-dimer valence-bond solid Mott insulator when inter-triangle correlations dominate, and a broken-$\mu=U/2$7-symmetry antiferromagnetic insulator when intra-triangle frustration dominates (Nourse et al., 2022). The paper explicitly argues that neither transition is well described by either the Brinkman-Rice or Slater paradigms (Nourse et al., 2022). On the family of vortex-full square lattices with $\mu=U/2$8-flux per plaquette, QMC and CUTs find only four phases—semi-metal, band insulator, Néel order, and valence-bond solid—and report no evidence for a spin-liquid phase anywhere in the half-filled ground-state phase diagram (Ixert et al., 2014).

These variants are not the same model as the translationally invariant nearest-neighbor square-lattice Hubbard problem, but they sharpen a general lesson: at half filling, the one-band Hubbard interaction does not determine the phase structure by itself. The noninteracting band topology, explicit symmetry-breaking fields, and the availability of dimer or flat-band limits can qualitatively reorganize the weak-coupling state and the route into the Mott regime (Ixert et al., 2014).

6. Interpretive boundaries, recurrent misconceptions, and methodological limits

A central misconception in the half-filled literature is to identify Mott insulating behavior with vanishing double occupancy. In the symmetric half-filled periodic Anderson-Hubbard model, exact diagonalization on finite one-dimensional chains always finds finite $\mu=U/2$9-site double occupancy $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$00, mirroring the exact one-dimensional half-filled Hubbard result that double occupancy remains finite for all finite $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$01, whereas the Gutzwiller treatment predicts a Brinkman-Rice transition with $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$02 once both orbitals are interacting (Hagymasi et al., 2012). The broader implication is that a Mott insulator need not be characterized by a Brinkman-Rice collapse of double occupancy (Hagymasi et al., 2012).

A second recurring issue is the distinction between local and nonlocal physics. Paramagnetic DMFT captures local Mott coexistence and local-moment/Kondo scales, but it excludes nonlocal self-energy effects and, when constrained to homogeneous paramagnetic solutions, also excludes ordered states by construction (Blümer et al., 2012). In the two-dimensional square-lattice case, DMFT therefore yields a sharp local picture of preformed moments and screening, but the results are explicitly interpreted as local-correlation physics most directly relevant to frustrated lattices or paramagnetic states (Mazitov et al., 2021).

Method-specific caveats are equally important. The half-filled square-lattice model is sign-problem free in AFQMC and in the nearest-neighbor Hubbard limit, which is what allows numerically exact imaginary-time data to be obtained for the 2D spectral problem (Schumm et al., 3 Apr 2025). By contrast, adding Holstein phonons spoils the exact sign protection of the half-filled nearest-neighbor Hubbard model in DQMC, even though the average sign can recover again at strong Holstein coupling when $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$03 (Johnston et al., 2013). Strong-coupling diagram techniques, although useful for Hubbard-band formation and approximate critical scales, are local approximations that may require artificial broadening or interpolation over unphysical structures to restore causality or remove narrow spurious gaps (Sherman, 2015).

Finally, rigorous constructions need careful interpretation. A decorated multi-band Hubbard model can be driven, in the limits $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$04 and band-gap $n=\langle n_{i\uparrow}+n_{i\downarrow}\rangle=1$05, to a low-energy sector with a single surviving finite-energy band at half filling, and that effective one-band problem supports rigorous saturated ferromagnetism for sufficiently small hole number on closely packed lattices (Tanaka et al., 2016). But the same work explicitly states that this is not a theorem for the generic nearest-neighbor one-band Hubbard model on an arbitrary lattice (Tanaka et al., 2016). The half-filled single-band Hubbard model is therefore best understood as a family of closely related problems whose universal content—Mott localization, moment formation, and strong-coupling spin physics—must continually be distinguished from lattice-specific, solver-specific, and approximation-specific features.