Hard-Core Fermi-Hubbard Model

• The hard-core Fermi-Hubbard model is a limit of the Fermi-Hubbard Hamiltonian where on-site double occupancy is energetically suppressed, leading to a Mott-insulating state with localized spins.
• It is implemented in optical lattices with tunable parameters that isolate the Mott phase and reveal extended antiferromagnetic correlations through controlled hopping and interaction strengths.
• Experimental signatures such as doublon suppression, modulation spectroscopy, and mapping to Heisenberg spin models underscore its significance in probing quantum magnetism and simulation.

The hard-core Fermi-Hubbard model is the strong-coupling limit of the spinful Fermi-Hubbard Hamiltonian in which on-site double occupancy is energetically suppressed, so the low-energy Hilbert space is restricted from $$\{|0\rangle,|\uparrow\rangle,|\downarrow\rangle,|\uparrow\downarrow\rangle\}$$ to a constrained sector dominated by $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$. For repulsive interactions at half filling, this is the Mott-insulating regime: charge fluctuations are frozen, local moments remain, and the low-energy theory reduces to a spin-$\tfrac12$ Heisenberg antiferromagnet with superexchange $J=4t^2/U$ (Esslinger, 2010). In contemporary cold-atom realizations, this regime is accessed in homogeneous three-dimensional optical lattices with $U/t$ up to $\sim 18$, where doublon fractions become rare and extended antiferromagnetic correlations and the antiferromagnetic phase transition can be resolved directly (Wang et al., 12 Feb 2025, Shao et al., 2024).

1. Microscopic formulation

The standard spinful Fermi-Hubbard Hamiltonian relevant to the hard-core limit is

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

or, in homogeneous bulk realizations, without a dominant site-dependent confinement term,

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

Here $t$ is the nearest-neighbor tunneling matrix element, $U$ is the on-site interaction between opposite spins, and $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$0 is the chemical potential. In optical lattices, $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$1 originates from Wannier-function overlap between neighboring wells, while $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$2 with $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$3; $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$4 is tuned through a Feshbach resonance $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$5. Three orthogonal standing waves generate

$|0\rangle,|\uparrow\rangle,|\downarrow\rangle$6

with lattice spacing $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$7 and recoil energy $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$8; increasing lattice depth exponentially suppresses $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$9, and in the lowest band the next-nearest-neighbor tunneling is suppressed by about an order of magnitude (Esslinger, 2010).

In trapped realizations, the last term produces spatially varying local chemical potential $\tfrac12$0, typically with harmonic confinement. In homogeneous flat-top plus box-trap implementations, the central region instead realizes an approximately uniform $\tfrac12$1, $\tfrac12$2, and $\tfrac12$3, which removes the dominant LDA convolution present in Gaussian lattices and makes direct comparison to homogeneous Hubbard theory possible over $\tfrac12$4 sites (Wang et al., 12 Feb 2025).

2. Strong-coupling projection and effective models

In the repulsive model, “hard-core” means $\tfrac12$5: a doublon costs energy $\tfrac12$6, so low-energy states exclude simultaneous occupation by opposite spins on the same site. This does not change fermionic statistics; it truncates the local Hilbert space by suppressing $\tfrac12$7. At half filling, the system becomes a Mott insulator: it is incompressible, double occupancy is suppressed, and each site carries a localized spin degree of freedom. The low-energy effective Hamiltonian is

$\tfrac12$8

obtained from second-order virtual hopping processes that momentarily create a doublon and favor antiferromagnetic correlations (Esslinger, 2010).

A recurrent misconception is to identify the hard-core limit with a band insulator. The review explicitly distinguishes them: both are incompressible, but the Mott insulator is interaction-driven, suppresses doublons, and can carry large spin entropy, whereas a band insulator has maximal double occupancy and interactions play a minor role (Esslinger, 2010).

Away from half filling, a plausible low-energy description is a $\tfrac12$9–$J=4t^2/U$0-type constrained-hopping model, although the review emphasizes primarily the Heisenberg limit and Mott physics. In the strong-coupling half-filled case on planar graphs, the same projection yields the $J=4t^2/U$1–$J=4t^2/U$2 model with no double occupancy, and two-particle scattering is reflection-only with no transmission, “as would be the case for classical hard spheres,” which underlies the universality construction based on hard-core scattering (Bao et al., 2014).

The attractive branch supplies a complementary hard-core limit. For $J=4t^2/U$3 and large $J=4t^2/U$4, opposite-spin fermions bind into on-site doublons; these pairs “can be regarded as hard-core bosons,” with pair hopping dominated by second-order tunneling $J=4t^2/U$5 and repulsive nearest-neighbor interaction also proportional to $J=4t^2/U$6 (Esslinger, 2010). In the mass-imbalanced extended Fermi-Hubbard model with nearest-neighbor interaction $J=4t^2/U$7, the doublon subspace is governed by

$J=4t^2/U$8

with

$J=4t^2/U$9

At the resonance point $U/t$0, doublons behave like free hard-core bosons; under a tilt they exhibit Bloch oscillations with period $U/t$1, while bound doublon pairs yield period $U/t$2 (Zhang et al., 2024).

3. Realization in optical lattices and homogeneous traps

The canonical realization uses ultracold fermions in optical lattices. Deepening the lattice reduces $U/t$3; tuning the scattering length through a Feshbach resonance increases $U/t$4; together these drive the system into the large-$U/t$5 regime. Three-dimensional simple-cubic lattices, 2D planes, 1D tubes, superlattices, and triangular lattices are all obtained by modifying lattice depths along selected axes, so dimensionality and frustration are directly engineered rather than emergent (Esslinger, 2010).

The homogeneous realization reported in 2025 combines a flat-top optical lattice with a cylindrical optical box. Each axis is formed from two counter-propagating 1064 nm flat-top beams, producing an almost uniform central disk of diameter $U/t$6, while blue-detuned 532 nm ring and line beams isolate the central region. The resulting 3D simple-cubic sample contains $U/t$7 lattice sites, with a flat density profile over $U/t$8 diameter. At lattice depth $U/t$9, with $\sim 18$0, the hopping is $\sim 18$1; the interaction is tuned over $\sim 18$2, corresponding to $\sim 18$3 (Wang et al., 12 Feb 2025).

This homogeneous geometry is consequential for the hard-core regime. In harmonic traps, metallic, Mott-insulating, and band-insulating shells coexist. In the flat-top plus box-trap system, by contrast, the bulk Hamiltonian is essentially uniform, so the metal-to-Mott crossover, Pomeranchuk physics, and AFM correlations can be read out without dominant trap averaging (Wang et al., 12 Feb 2025).

4. Phases and experimental signatures

In trapped repulsive systems, the phase structure is shell-like. Low atom number gives a metallic cloud; at intermediate atom number a Mott core with density $\sim 18$4 appears at the center and metallic shells survive outside; when the central chemical potential exceeds $\sim 18$5, double occupancy rises again and eventually a band insulator with two atoms per site forms in the core. The defining hard-core signature is the suppression of doubly occupied sites in the Mott region, not merely incompressibility (Esslinger, 2010).

Double occupancy is the central diagnostic. In early optical-lattice experiments, a rapid lattice-depth ramp froze motion, a Feshbach resonance shifted the energy of doubly occupied sites, and an RF pulse transferred one partner to an empty spin state, making doublons countable by absorption imaging. In the Mott regime, double occupancy below $\sim 18$6 was measured, with estimated hole fraction below $\sim 18$7 and $\sim 18$8; cloud-size measurements simultaneously indicated an incompressible core (Esslinger, 2010).

In the homogeneous implementation, the doublon fraction is defined by

$\sim 18$9

The measurement freezes tunneling by ramping from $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$0 to $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$1 in $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$2, then uses a Gaussian-shaped microwave pulse of total duration $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$3 to drive $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$4, producing distinct resonance peaks for atoms on singly occupied and doubly occupied sites. At half filling and entropy per particle $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$5, $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$6 decreases smoothly with increasing $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$7; around $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$8, $$\hat H = -t \sum_{\langle i,j\rangle,\sigma} \left( \hat c^{\dagger}_{i\sigma} \hat c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i \hat n_{i\uparrow} \hat n_{i\downarrow} + \sum_{i,\sigma} (V_i-\mu)\hat n_{i\sigma},$$9, and for the coldest samples at $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$0, $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$1 reaches $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$2 (Wang et al., 12 Feb 2025).

Modulation spectroscopy probes the high-energy sector excluded by the hard-core constraint. Periodic modulation $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$3 creates doublon-hole excitations resonantly at $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$4, directly measuring the Hubbard gap and the separation between lower and upper Hubbard bands. Doublon lifetimes scale exponentially with $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$5, consistent with the need to redistribute energy $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$6 into many low-energy excitations (Esslinger, 2010).

5. Antiferromagnetism, Pomeranchuk physics, and criticality

The hard-core Fermi-Hubbard model is not exhausted by doublon suppression; its central low-energy content is quantum magnetism. In the 3D repulsive model at half filling, the Néel transition interpolates between a weak-coupling SDW instability and the strong-coupling Heisenberg regime. In the latter, the paper uses the known Heisenberg result

$$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$7

so the AFM transition is set by superexchange rather than by the Hubbard scale $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$8 (Shao et al., 2024).

The homogeneous 2025 experiment inferred this crossover already from doublon thermodynamics. At fixed $$H = -t \sum_{\langle i,j \rangle,\sigma} \left( c^{\dagger}_{i\sigma} c_{j\sigma} + \mathrm{h.c.} \right) + U \sum_i n_{i\uparrow} n_{i\downarrow} - \mu \sum_i (n_{i\uparrow}+n_{i\downarrow}).$$9, $t$0 is non-monotonic: for high entropy $t$1, cooling reduces $t$2; around $t$3, $t$4 reaches a minimum; for $t$5, $t$6 increases upon further cooling, identifying the Pomeranchuk effect; and for $t$7, $t$8 turns downward again, signaling the development of extended antiferromagnetic correlations. The same qualitative structure appears at $t$9, while at $U$0 thermally induced doublons are so suppressed that the remaining doublons are primarily quantum fluctuations $U$1 (Wang et al., 12 Feb 2025).

The AFM phase transition itself was reported in 2024 in a 3D uniform optical lattice of approximately $U$2 sites. The spin structure factor at

$U$3

was measured via spin-sensitive Bragg scattering and defined as

$U$4

Near criticality, the data were fitted with

$U$5

for $U$6, $U$7, or $U$8, consistent with the Heisenberg universality class. At half filling and near optimal coupling, the measured spin structure factor reached $U$9, and the AFM dome in interaction space was bounded by $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$00 and $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$01. At $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$02, the critical entropy was $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$03; at $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$04, $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$05. Doping suppressed AFM order rapidly, with fitted critical fillings $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$06 and $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$07 (Shao et al., 2024).

6. Computational formulations and broader significance

The hard-core limit is also a central organizing principle in quantum simulation and Hamiltonian complexity. A qudit formulation based on a Qudit Fermionic Mapping encodes the full spinful Hubbard site in one ququart with local basis $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$08; the hard-core limit is then enforced either by large $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$09 so that the fourth level is energetically inaccessible, or by restricting to the $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$10 subspace. This preserves nearest-neighbor connectivity and makes the on-site interaction a single-qudit operation (Vezvaee et al., 2024).

Ground-state preparation protocols have also been adapted to 2D hardware. QETU-based preparation of the $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$11 Fermi-Hubbard model on a 9-qubit grid uses controlled time evolution, fermionic swap networks, and a 2D topology tailored to the Jordan-Wigner layout; the same circuit structure remains applicable in the hard-core limit, where on-site $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$12 terms are removed or assigned a very large penalty (Müller et al., 2024). A distinct “Fermi Machine” construction maps Hubbard models to multi-component noninteracting fermions and, in the large-$|0\rangle,|\uparrow\rangle,|\downarrow\rangle$13 limit, can enforce perfect exclusion of double occupancy through the hidden-fermion structure (Imada, 2024).

At the complexity-theoretic end, planar-graph Hubbard dynamics at strong coupling and half filling reduce to the $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$14–$|0\rangle,|\uparrow\rangle,|\downarrow\rangle$15 model, where only reflection and no transmission occur in scattering events; this reflection-only hard-core dynamics suffices to encode universal quantum computation (Bao et al., 2014). For parameter identification rather than state preparation, a Hamiltonian-learning protocol for bounded-degree Fermi-Hubbard graphs achieves $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$16 total evolution time and uses only simple one- or two-site fermionic manipulations, which is directly relevant to calibrating analog hard-core Hubbard simulators (Ni et al., 2023).

Alternative solid-state simulators reach the same qualitative regime. A semiconductor triple-dot array realizing an extended Fermi-Hubbard Hamiltonian achieved $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$17 and $|0\rangle,|\uparrow\rangle,|\downarrow\rangle$18, allowing detailed characterization of the collective Coulomb blockade transition as the finite-size analogue of the interaction-driven Mott metal-to-insulator transition (Hensgens et al., 2017). This suggests that the hard-core Fermi-Hubbard model is best viewed not as a single limiting Hamiltonian but as a unifying low-energy description of constrained charge motion, Mott insulation, superexchange magnetism, and, in attractive or extended variants, hard-core doublon dynamics across cold-atom, solid-state, and digital-quantum platforms.