Mott Insulators

A Mott insulator is an interaction-driven, charge-incompressible state at a commensurate filling for which the appropriate independent-particle baseline would permit metallic charge motion. Its defining physics is a many-body obstruction to changing the charge density, not merely large resistivity, a narrow band, suppressed double occupancy, or magnetic order.

This definition is operational rather than taxonomic perfection. Real compounds can combine Coulomb repulsion with ligand hybridization, crystal-field splitting, magnetism, disorder, and structural distortion. “Mott” is most trustworthy when it identifies the dominant origin of charge incompressibility and survives tests against those alternatives. It should not be used as a decorative synonym for “strongly correlated.”

We connect the charge gap to density response, spectroscopy and local spin dynamics, then examine how bandwidth and filling tune real materials. The Hubbard Model develops the lattice Hamiltonian and controlled limits; Metals, Insulators, and Semiconductors compares the broader mechanisms of insulating behavior.

Required background. What Are Strong Correlations? supplies the distinction between interaction scales and demonstrated failure of an independent-particle description. Hubbard Model supplies hopping, onsite repulsion, and the atomic and strong-coupling limits. Metals, Insulators, and Semiconductors distinguishes band, interaction, and localization mechanisms, while Spectral Functions supplies addition/removal thresholds and spectral-weight normalization used to interpret Hubbard bands.

Charge Incompressibility

Let $E_0(N)$ be the ground-state energy in the sector with $N$ particles. The zero-temperature costs to add and remove one particle are

$$\begin{aligned} \mu_- &= E_0(N)-E_0(N-1),\\ \mu_+ &= E_0(N+1)-E_0(N). \end{aligned}$$

Their difference is the adjacent-sector charge-gap diagnostic,

$$\Delta_c(N) = \mu_+(N)-\mu_-(N) = E_0(N+1)+E_0(N-1)-2E_0(N).$$

For an ordinary Mott state, these adjacent sectors determine the chemical-potential plateau only when no multiparticle charging process has a lower cost per particle. More generally, compare $E_0(N)-\mu N$ with every other charge sector. The exact bounds are

$$\begin{aligned} \mu_-^{\mathrm{gc}}(N) &=\sup_{M<N}\frac{E_0(N)-E_0(M)}{N-M},\\ \mu_+^{\mathrm{gc}}(N) &=\inf_{M>N}\frac{E_0(M)-E_0(N)}{M-N}. \end{aligned}$$

The $N$ sector is a grand-canonical ground state when $\mu_-^{\mathrm{gc}}\leq\mu\leq\mu_+^{\mathrm{gc}}$; if the lower bound exceeds the upper one, it is never stable. A nonzero interval in the thermodynamic limit gives a density plateau. The adjacent bounds coincide with these all-sector bounds only when single-particle addition and removal control both edges. Stable thermodynamic energy density is convex, but finite-sector energies can have even–odd pairing effects: a single-particle gap alone need not imply incompressibility.

If $n=N/V$, a convenient lattice compressibility is

$$\kappa_n(T,\mu) = \left(\frac{\partial n}{\partial\mu}\right)_T.$$

We omit the conventional factor $1/n^2$ used in the thermodynamic compressibility. At zero temperature, an ideal clean Mott phase has a density plateau and $\kappa_n=0$ at its commensurate filling. For a number-conserving Hamiltonian in thermal grand-canonical equilibrium,

$$\kappa_n = \frac{\beta}{V} \left( \langle N^2\rangle-\langle N\rangle^2 \right),$$

where $\beta=1/(k_{\mathrm B}T)$. Here $N$ is the total particle number, not the occupancy of one site. At low temperature, a compressible bulk metal has $\operatorname{Var}N/V\simeq k_{\mathrm B}T\kappa_n$; a charge gap suppresses this ratio more strongly. Zero total-number variance at strictly zero temperature alone does not distinguish an insulator from a metal, and a fixed-$N$ calculation has no total-number variance by construction. Its compressibility must instead be inferred from changes across sectors or another appropriate response.

At nonzero temperature, thermally activated carriers round the plateau. Vanishing compressibility is a zero-temperature limit, not an exact property of every finite-temperature curve.

Two views of an ordinary Mott state when no multiparticle charging process lowers the plateau edges. The upper curve shows constant density over a chemical-potential interval; the lower curve shows separated electron-removal and addition sectors at a chemical potential inside that gap. Its horizontal axis is energy $\varepsilon$ relative to the chemical potential, so the center is $\varepsilon=0$. The curves are schematic; optical and transport thresholds need not coincide with the single-particle gap.

One phrase, several gaps

The phrase “Mott gap” is often used too loosely. These quantities answer different experimental questions:

Quantity	What it measures
Thermodynamic plateau	The all-sector interval $\mu_+^{\mathrm{gc}}-\mu_-^{\mathrm{gc}}$, tested by density response or sector energies.
One-particle gap	The separation between electron-removal and addition edges; spectral matrix elements and surfaces affect its measurement.
Optical gap	The lowest allowed absorption onset; excitons and selection rules can move it.
Transport activation scale	The slope in a chosen transport law; defects, hopping and contacts can dominate it.
Neutral or spin gap	An excitation at fixed total charge, potentially far softer than the charge gap.

No theorem requires these measured scales to coincide. A transport activation energy much smaller than a photoemission gap is therefore not paradoxical; it may reveal defect-assisted conduction rather than a collapsed bulk charge gap. Conversely, an optical onset below the unbound particle–hole continuum may be an exciton and does not by itself establish compressibility.

From Atomic Repulsion to Hubbard Bands

The repulsive Hubbard model supplies the cleanest minimal mechanism. At one particle per site in the atomic limit, removing a particle creates an empty site and adding one creates a doubly occupied site. Their combined charge cost is $U$. Finite hopping gives those atomic transitions dispersion and transfers their weight into broad removal and addition structures conventionally called lower and upper Hubbard bands.

That language is useful but schematic:

• the centers of broad spectral features are not automatically the thermodynamic gap edges;
• a Mott spectrum need not display two clean, symmetric peaks;
• the gap is generally not exactly $U$ once hopping, nonlocal interactions, hybridization, and multiplets matter;
• a coherent quasiparticle feature can appear on the metallic side between incoherent high-energy structures;
• orbital-selective systems may localize one orbital sector while others remain itinerant.

The ratio $U/W$, with $W$ an explicitly defined bandwidth, organizes the competition between localization and delocalization. It is not a universal phase boundary. The critical ratio depends on lattice density of states, dimensionality, frustration, filling, orbital degeneracy, Hund coupling, temperature, magnetic constraints, and the approximation used. Both parameters need a specified orbital basis and screening convention.

Separate charge and spin energy scales

Suppressing double occupancy tends to leave spin or orbital degrees of freedom active on each site. Virtual charge fluctuations then generate exchange at an energy scale much smaller than the charge cost. In the simplest large-$U$ one-band setting,

$$J \sim \frac{4t^2}{U},$$

while the charge scale remains of order $U$ after bandwidth corrections. A Mott material can consequently satisfy

$$J \ll \Delta_c,$$

with charge frozen over a broad temperature range but spins still fluctuating, ordering, or forming a more entangled state at much lower energies.

This is a separation of energy scales, not by itself fractionalization into independent spinons and holons. It has three immediate consequences:

1. a Mott insulator need not have a spin gap;
2. local moments do not imply long-range magnetic order;
3. antiferromagnetic order is not the definition of Mott localization.

Exchange Interactions explains the material mechanisms behind $J$; Effective Hamiltonians in Many-Body Systems derives the projection from charge sectors to spin or t–J descriptions.

Bandwidth and Filling Control

Two idealized routes out of a Mott insulator should be kept distinct.

Bandwidth control

At fixed commensurate filling, pressure, strain, bond angles, or chemical substitution can increase orbital overlap and hence $W$, decreasing an effective $U/W$. This route can connect an incompressible state to a correlated metal without intentionally changing the carrier count. In practice, pressure also changes crystal fields, hybridization, phonons, and sometimes structure; “bandwidth control” is a model coordinate, not a guarantee that only $W$ moved.

In paramagnetic single-site dynamical mean-field theory for the half-filled single-band Hubbard model, a low-temperature region of metallic and insulating self-consistent solutions terminates at a finite-temperature critical endpoint. A Landau description of that endpoint is standard within this framework (Kotliar, Lange and Rozenberg, 2000). Magnetic ordering is suppressed in this construction, which can be useful when frustration lowers its scale. It is not a universal phase diagram for every solid: magnetic order, lattice distortion, charge order, nonlocal correlations, or disorder can preempt or reshape the paramagnetic transition.

Experiments on pressure-tuned $\mathrm{V_2O_3}$ and $\kappa$-type BEDT–TTF organic salts have revealed hysteresis, crossovers, and critical behavior consistent with important parts of this picture. They also reveal spatial inhomogeneity, elastic coupling, and material-specific exponents. Agreement with one model feature should be stated as such rather than promoted to proof of a universal Mott endpoint.

Filling control

Leaving the plateau changes the density from its commensurate value; the equation of state determines whether the density changes continuously or jumps. For hole doping of a one-band half-filled state, write

$$\delta = 1-n.$$

A homogeneous doped metallic branch is compressible, but other commensurate insulating states or phase separation can intervene. Even when metallic, the doped system is not a rigidly shifted version of the noninteracting band. Mobile carriers propagate through a constrained spin and orbital background, and spectral weight can move between energy windows separated by the original interaction scale. Chemical substitution additionally introduces disorder, local strain, and electrostatic changes, whereas electrostatic gating or cold-atom preparation can more nearly isolate filling.

Bandwidth- and filling-controlled transitions can therefore end in different metals and need not share a critical theory.

Beyond the One-Band Mott–Hubbard Picture

Charge-transfer insulators

Many transition-metal compounds contain ligand $p$ states near the correlated metal $d$ states. Let $\Delta$ denote a ligand-to-metal charge-transfer energy and $U_d$ the interaction cost within the $d$ shell. In the Zaanen–Sawatzky–Allen organization:

• a Mott–Hubbard insulator has low-energy removal and addition edges dominated primarily by different $d$-electron configurations, with $U_d$ setting the basic charge cost;
• a charge-transfer insulator has the lowest removal states predominantly on ligands, so $\Delta$ rather than $U_d$ controls the smallest gap.

Hybridization means that actual eigenstates are mixtures, not ionic labels. In late transition-metal oxides, ligand holes and even negative effective charge-transfer energies can make formal oxidation-state counting especially misleading. The correct classification follows the orbital character and energetics of the many-body addition and removal states.

The parent copper oxides are more accurately described as strongly hybridized charge-transfer insulators than as literal realizations of a one-band atomic Mott gap. A low-energy one-band or t–J description can still be useful after a justified downfolding; it should not erase the oxygen degrees of freedom from statements about spectroscopy or charge transfer.

Multiorbital and structural entanglement

Crystal-field splittings, Hund coupling, spin–orbit coupling, intersite repulsion, and electron–lattice coupling can cooperate with localization. Rare-earth nickelates and $\mathrm{VO_2}$, for example, exhibit strong electronic effects but also decisive lattice, bonding, or charge-disproportionation physics. Calling either a “pure Mott transition” without qualifying the structural evidence overstates what the label resolves.

Filling constraints

For a clean periodic system with local interactions, conserved charge and ordinary lattice translations, a unique, symmetry-preserving, short-range-entangled gapped ground state is generally compatible only with an integer filling per primitive unit cell. At fractional filling, a fully gapped phase must evade that conclusion through translation-symmetry breaking, topological order, or another assumption failure. Additional spin and time-reversal symmetries can impose stronger filling restrictions; magnetic translations in a background flux require a different formulation. This Lieb–Schultz–Mattis–Oshikawa reasoning (Oshikawa, 2000) does not prove that an integer-filled insulator is Mott; it constrains which insulating outcomes are possible before a material mechanism is assigned.

Experimental Diagnosis

A material interpretation connects charge response to spectra, spin dynamics and structure.

Establish the noninteracting baseline

Determine the primitive and any ordered unit cells, electron count, orbital character, spin–orbit scale, and plausible band structure. A half-filled band in an oversimplified cell is not evidence if a structural dimerization or antiferromagnetic unit-cell enlargement already permits a conventional gap. Compare nonmagnetic, magnetic, and distorted baselines when the relevant orders are close in energy.

Measure charge incompressibility

Direct bulk electronic compressibility is difficult in solids because long-range Coulomb forces, lattice elasticity, and contacts complicate density response. Chemical-potential shifts and the electronic contribution to capacitance can constrain the density response after geometrical and electrostatic contributions are treated. The disappearance of quantum oscillations alone does not measure incompressibility. Ultracold lattice fermions allow in situ density and local-compressibility measurements, compared with a finite-temperature equation of state and a treatment of the trap (Duarte and collaborators, 2015).

Reconstruct addition and removal sectors

Angle-resolved photoemission spectroscopy measures occupied removal spectra; inverse photoemission and suitable x-ray probes access addition character; scanning tunneling microscopy and spectroscopy sample a surface-weighted local spectrum. A bulk gap claim should account for surface reconstruction, charging, matrix elements, energy resolution, and whether the same chemical potential aligns the two sectors. Spectral Functions supplies the common Lehmann and sum-rule language.

Use optical sum rules, not only an onset

Optical conductivity can follow weight over a much wider energy window than dc transport. Across pressure, temperature, or doping, transfer between a Drude response, mid-infrared structures, and high-energy interband or Hubbard features is stronger evidence than fitting one activation slope. The integration cutoff and subtraction of phonons and interband backgrounds must be reported because partial spectral weights are cutoff dependent. Sum Rules explains the exact moment constraints; Kubo Formula develops the response calculation.

Separate charge, spin, and structure

Magnetic susceptibility, NMR, neutron scattering, and resonant x-ray probes can establish local moments and their correlations. Diffraction, Raman, and elastic measurements test whether a structural or charge-ordering transition accompanies the gap. A Mott interpretation is strongest when the charge gap or local-moment regime persists where long-range order has melted and when tuning evolves charge, spin, and structure consistently.

Require a parameter-consistent model

A realistic calculation should use the same structure, filling, orbital window, and interaction convention across spectra and thermodynamics. Density-functional theory plus dynamical mean-field theory can connect material bands to local dynamics; cluster extensions and other many-body methods test nonlocal correlations. Fitting $U$ to one gap leaves many alternatives. Predictions for spectral-weight transfer, moments, compressibility and their evolution under tuning provide independent tests.

Mott, Slater, Band, and Anderson Limits

The labels identify mechanisms, not mutually exclusive appearances.

Mechanism	Distinguishing physical feature
Band insulation	Filled one-particle bands with a gap; no broken symmetry is required.
Slater-like insulation	Magnetic band reconstruction, with a gap closely tied to the order.
Mott-like insulation	Interaction-driven charge incompressibility; magnetic order is optional.
Charge-transfer insulation	Correlated metal–ligand transitions determine the lowest charge cost.
Anderson localization	Disorder localizes available states; a spectral gap need not exist.

Mott and Slater behavior can form a continuum when the same interactions create both moments and order. A gap that grows below the Néel temperature need not be purely Slater, and a large $U$ does not make every ordered insulator purely Mott. The useful question is which observables require local interaction physics above and beyond symmetry-induced band folding.

Representative materials and simulators

$\mathrm{V_2O_3}$ and chromium-doped variants provide classic pressure- and composition-tuned examples involving paramagnetic metal, paramagnetic insulator and antiferromagnetic phases. Orbital changes, coexistence and lattice coupling remain important (Limelette and collaborators, 2003).

In $\kappa$-(BEDT–TTF)$_2X$ organic salts, molecular dimers form narrow bands and modest pressure can cross a metal–insulator boundary. Triangular geometry, intradimer charge structure and elasticity affect the transition and its critical behavior (Kagawa, Miyagawa and Kanoda, 2005). Cuprates instead highlight metal–oxygen charge transfer and the correlated phases reached by doping.

Optical-lattice atoms offer adjustable Hubbard parameters and spatial density measurements. Finite entropy, trap averaging and the conversion from local profiles to homogeneous thermodynamics require explicit treatment. Their simplicity makes them useful tests of a model, while real solids bring additional orbital and lattice couplings.

Doping a Mott Insulator

In a rigid-band semiconductor, doping primarily changes which pre-existing one-particle states are occupied. In a doped Mott system, changing the occupancy also changes the number and character of allowed local transitions. Even the atomic counting is non-rigid: removing a fraction $\delta$ of particles creates empty sites that accept either spin at low energy, while addition to singly occupied sites still accesses a high-energy doubly occupied sector. Finite hopping mixes these sectors and produces dynamical spectral-weight transfer.

At strong coupling, the projected t–J Model captures mobile holes constrained against double occupancy and coupled to exchange. It is a controlled descendant of the Hubbard model only within a specified hierarchy such as $t/U\ll1$ and with generated longer-range terms tracked. It is not a universal phenomenology for every doped oxide.

Doping may lead to a correlated Fermi liquid, pseudogap, strange metal, charge or spin order, phase separation frustrated by long-range Coulomb interactions, superconductivity, or coexistence among these tendencies. Which outcome occurs depends on dimension, frustration, longer-range hopping and interactions, phonons, disorder, and orbital composition. “Doped Mott insulator” defines a starting constraint, not a solved phase diagram.

The connection to high-temperature superconductivity is therefore precise but limited. Cuprate superconductivity emerges after doping a charge-transfer-correlated parent, and the redistribution of spectral weight and persistence of short-range spin correlations show that parent-state physics remains relevant. Those facts do not by themselves select a pairing glue, establish a particular gauge theory, or prove that all unconventional superconductors share the same mechanism.

Exercises

1. Addition and removal chemical potentials

A finite system has ground-state energies, in electronvolts,

$$E_0(N-1)=0.0,\qquad E_0(N)=4.2,\qquad E_0(N+1)=10.1.$$

Compute $\mu_-$, $\mu_+$, and $\Delta_c$. Show that adding a term $a+bN$ to every sector energy leaves $\Delta_c$ unchanged.

Solution

The removal-side and addition-side chemical potentials are

$$\mu_- = 4.2-0.0 = 4.2\ \mathrm{eV},$$

and

$$\mu_+ = 10.1-4.2 = 5.9\ \mathrm{eV}.$$

Therefore

$$\Delta_c = \mu_+-\mu_- = 1.7\ \mathrm{eV}.$$

Under $E_0(M)\mapsto E_0(M)+a+bM$, the second difference becomes

$$\Delta_c' = \Delta_c + \bigl[a+b(N+1)\bigr] + \bigl[a+b(N-1)\bigr] - 2(a+bN) = \Delta_c.$$

The arbitrary zero of energy and a uniform shift of chemical-potential convention cancel, as a physical charge gap must.

2. A single-particle gap without a plateau

Consider a finite model with only five allowed sectors and energies

$$E_0(N-2)=E_0(N)=E_0(N+2)=0, \qquad E_0(N-1)=E_0(N+1)=\epsilon>0.$$

Compute the adjacent-sector gap and the all-sector plateau bounds for $N$. What does the comparison teach?

Solution

The adjacent sectors give $\mu_-=-\epsilon$, $\mu_+=\epsilon$ and $\Delta_c=2\epsilon$. But two-particle addition and removal each cost zero per particle. Consequently,

$$\mu_-^{\mathrm{gc}}=\max(-\epsilon,0)=0, \qquad \mu_+^{\mathrm{gc}}=\min(\epsilon,0)=0.$$

The $N$ sector is a ground state only at the single point $\mu=0$, where it is degenerate with $N\pm2$. It has no finite plateau despite its positive adjacent-sector gap. This finite example illustrates an even–odd charging effect, not a bulk phase diagnosis; a thermodynamic conclusion still requires size scaling and all relevant sectors.

3. Atomic addition and removal sectors

For one Hubbard site,

$$\mathcal K_{\mathrm{at}} = U n_\uparrow n_\downarrow - \mu(n_\uparrow+n_\downarrow),$$

take $0<\mu<U$, so the singly occupied states are ground states. Find the removal and addition costs. What happens at the particle–hole-symmetric choice $\mu=U/2$?

Solution

The empty, singly occupied, and doubly occupied grand energies are

$$\mathcal K_0=0,\qquad \mathcal K_1=-\mu,\qquad \mathcal K_2=U-2\mu.$$

Removing the particle from a singly occupied ground state costs

$$\varepsilon_- = \mathcal K_0-\mathcal K_1 = \mu,$$

whereas adding the opposite spin costs

$$\varepsilon_+ = \mathcal K_2-\mathcal K_1 = U-\mu.$$

Their sum, equivalently the separation between atomic removal and addition transitions, is $U$. At $\mu=U/2$, the two edges are symmetric:

$$\varepsilon_-=\varepsilon_+=\frac{U}{2}.$$

Thus $U$ is the full atomic charge gap, while each edge lies $U/2$ from the symmetric chemical potential. Confusing those quantities creates a factor-of-two error.

4. Three experimental gaps

A compound has a $1.8\ \mathrm{eV}$ separation between bulk-sensitive addition and removal edges, an optical onset at $1.4\ \mathrm{eV}$, and dc resistivity fitted over a narrow range by $\rho\propto e^{0.25\,\mathrm{eV}/(k_{\mathrm B}T)}$. Give one physically consistent interpretation. Which number should be called the thermodynamic charge gap?

Solution

The $1.8\ \mathrm{eV}$ spectral separation may approximate the one-particle fundamental gap. An optically allowed bound exciton can lie below the unbound electron–hole continuum, giving a $1.4\ \mathrm{eV}$ onset. Defects or impurity bands can support activated or hopping transport at a much smaller scale, producing the $0.25\ \mathrm{eV}$ fit.

None of these numbers is automatically the thermodynamic charge gap. The incompressible interval is the difference of the all-sector chemical-potential bounds in the bulk thermodynamic limit. Bulk single-particle addition/removal edges may estimate it if chemical-potential alignment, surface effects, and excitonic binding are controlled and no cheaper multiparticle charging route exists. A density plateau or total-energy comparison across the relevant charge sectors directly tests that additional condition. The transport slope alone is the least reliable identification.

5. Distinguishing Mott-like and Slater-like behavior

Material A becomes antiferromagnetic at $T_N$. Above $T_N$, spectroscopy shows a broad charge gap and local moments; below $T_N$, band folding appears and the gap increases by 15%. Material B is metallic above $T_N$; below $T_N$, a gap opens continuously with the staggered moment and disappears when the order is suppressed. Classify the dominant limits and state what remains uncertain.

Solution

Material A is consistent with a Mott-dominant interpretation in this comparison: charge localization and local moments already exist without long-range order, while antiferromagnetism reconstructs and enlarges an existing gap. The ordered phase can still contain a Slater contribution.

Material B is Slater-like because the insulating gap tracks the ordering symmetry and vanishes with it. This does not imply weak interactions; interactions may generate the magnetism and renormalize the bands.

For both materials, one should still test structure, disorder, orbital character, and thermodynamic charge response. The evidence assigns a dominant limit, not an exact decomposition of the gap into additive “Mott” and “Slater” pieces.

6. Atomic spectral-weight transfer under hole doping

In the large-$U$ atomic counting limit, suppose a fraction $\delta$ of sites is empty and $1-\delta$ is singly occupied, with no double occupancy. Count the spin-summed electron-removal weight, low-energy electron-addition weight, and high-energy addition weight per site. Verify the total addition sum rule.

Solution

Each singly occupied site contains one removable electron, so the removal weight is

$$W_{\mathrm{rem}} = 1-\delta.$$

Each empty site accepts either spin without paying $U$, giving low-energy addition weight

$$W_{\mathrm{add}}^{\mathrm{low}} = 2\delta.$$

Each singly occupied site accepts only the opposite spin into the high-energy doubly occupied sector, so

$$W_{\mathrm{add}}^{\mathrm{high}} = 1-\delta.$$

The total addition weight is therefore

$$W_{\mathrm{add}} = 2\delta+(1-\delta) = 1+\delta.$$

This equals the number of empty spin orbitals per site:

$$2-(1-\delta)=1+\delta.$$

Even before finite hopping introduces dynamical transfer, doping creates low-energy addition weight proportional to $2\delta$, not merely $\delta$. This is the simplest demonstration that a doped Mott spectrum is not a rigidly shifted band.

For related mechanisms in narrow moiré bands, see Correlated Insulators in Moiré Systems. Computational Quantum Matter explains how to test a material model with controlled numerical calculations.

References

• D. P. Arovas, E. Berg, S. A. Kivelson, and S. Raghu, “The Hubbard model,” Annual Review of Condensed Matter Physics 13, 239–274 (2022).
• D. N. Basov, R. D. Averitt, D. van der Marel, M. Dressel, and K. Haule, “Electrodynamics of correlated electron materials,” Reviews of Modern Physics 83, 471–541 (2011).
• P. M. Duarte, R. A. Hart, T.-L. Yang, et al., “Compressibility of a fermionic Mott insulator of ultracold atoms,” Physical Review Letters 114, 070403 (2015).
• A. Georges, G. Kotliar, W. Krauth, and M. J. Rozenberg, “Dynamical mean-field theory of strongly correlated fermion systems and the limit of infinite dimensions,” Reviews of Modern Physics 68, 13–125 (1996).
• M. Imada, A. Fujimori, and Y. Tokura, “Metal-insulator transitions,” Reviews of Modern Physics 70, 1039–1263 (1998).
• F. Kagawa, K. Miyagawa, and K. Kanoda, “Unconventional critical behaviour in a quasi-two-dimensional organic conductor,” Nature 436, 534–537 (2005).
• G. Kotliar, E. Lange, and M. J. Rozenberg, “Landau theory of the finite temperature Mott transition,” Physical Review Letters 84, 5180–5183 (2000).
• P. A. Lee, N. Nagaosa, and X.-G. Wen, “Doping a Mott insulator: Physics of high-temperature superconductivity,” Reviews of Modern Physics 78, 17–85 (2006).
• P. Limelette, A. Georges, D. Jérome, et al., “Universality and critical behavior at the Mott transition,” Science 302, 89–92 (2003).
• P. Limelette, P. Wzietek, S. Florens, et al., “Mott transition and transport crossovers in the organic compound $\kappa$-(BEDT–TTF)$_2$Cu[N(CN)$_2$]Cl,” Physical Review Letters 91, 016401 (2003).
• M. Oshikawa, “Commensurability, excitation gap, and topology in quantum many-particle systems on a periodic lattice,” Physical Review Letters 84, 1535–1538 (2000).
• J. Zaanen, G. A. Sawatzky, and J. W. Allen, “Band gaps and electronic structure of transition-metal compounds,” Physical Review Letters 55, 418–421 (1985).