Mott Insulator

The puzzle band theory could not solve

Band theory is one of the great triumphs of quantum mechanics. Treat the electrons in a crystal as nearly free particles moving through a periodic potential, and you get bands of allowed energies separated by gaps. The rule that follows is simple and powerful: a solid with a partially filled band is a metal; a solid whose bands are completely filled or completely empty is an insulator. Diamond, silicon, and table salt are band insulators with full valence bands. Copper, with its half-filled 4s band, is a metal.

Then the theory hits a wall. Nickel oxide (NiO) has one unpaired 3d electron per nickel — an odd electron count, a half-filled band. Band theory says it must be a metal. In reality NiO is a transparent green insulator with a gap of about 4 eV and a room-temperature resistivity above 10¹³ Ω·cm — more insulating than many ceramics. The same scandal appears in CoO, MnO, and a whole family of transition-metal oxides. In 1937, at a meeting in Bristol, Rudolf Peierls and Nevill Mott pinned down the culprit: band theory throws away the electron-electron repulsion, and for these materials that is exactly the term that matters most.

The traffic-jam picture

Imagine a one-dimensional lattice of atoms, each with one orbital and exactly one electron sitting on it — half filling. In the free-electron picture each electron is a wave spread across the whole crystal and current flows effortlessly. But moving an electron means, at some instant, putting two electrons on the same atom. Two electrons crammed into one orbital repel each other through the Coulomb interaction, paying an energy cost we call U, the on-site repulsion.

If U is small compared with how much energy the electrons gain by spreading out (the bandwidth W), the electrons ignore the toll and the material conducts: it is a metal. If U is large, the toll is prohibitive. Every electron stays parked on its own atom because hopping next door would force a doubly-occupied, high-energy configuration. The lattice is gridlocked. There is exactly one car per parking space and nowhere to move — a traffic jam of electrons. The material insulates, even though the band is only half full. That is a Mott insulator.

The Hubbard model

The minimal Hamiltonian that captures this competition is the Hubbard model, written down by John Hubbard in 1963:

H = −t Σ⟨i,j⟩,σ ( c†_{iσ} c_{jσ} + h.c. )  +  U Σ_i n_{i↑} n_{i↓}

Two terms, two physics:

• Hopping, −t. The kinetic term lets an electron of spin σ tunnel from site j to neighboring site i. The amplitude t sets the bandwidth W ≈ 2zt, where z is the number of nearest neighbors (the coordination number). Big t means delocalized, metallic electrons.
• On-site repulsion, U. The interaction term charges an energy U only when a site holds both a spin-up and a spin-down electron (n↑ = n↓ = 1). It is the price of double occupancy.

Everything in Mott physics is the war between these two. The single dimensionless ratio U / W (equivalently U/t) decides the winner. At half filling — one electron per site on average — the phase diagram is roughly:

Regime	Behavior	Why
U / W ≪ 1	Metal (Fermi liquid)	Kinetic energy wins; electrons delocalize freely
U / W ≈ 1	Bad metal / critical region	Strong correlations; quasiparticles barely survive
U / W > (U/W)_c	Mott insulator	Double occupancy forbidden; electrons localize one per site

For a single-band model on the Bethe lattice, dynamical mean-field theory (DMFT) puts the zero-temperature critical value at roughly U_c2 ≈ 1.2 W (with a hysteretic, first-order region between U_c1 ≈ 0.8 W and U_c2). The exact number depends on lattice and filling, but the lesson is universal: the transition happens when U becomes comparable to the bandwidth, not at some tiny perturbation.

Upper and lower Hubbard bands

Where did the gap go in spectroscopy? In the Mott insulating state the single band of band theory splits into two Hubbard bands:

• The lower Hubbard band is the energy to remove an electron from a singly-occupied site (photoemission). Its weight is the occupied states.
• The upper Hubbard band is the energy to add an electron to an already-occupied site, paying U (inverse photoemission).

The two are separated by the Mott-Hubbard gap, of order U − W. In a many-body insulator like NiO, oxygen 2p states can sit inside this gap, and the smallest charge gap is actually a charge-transfer gap (oxygen → nickel) rather than a pure Mott gap. The Zaanen–Sawatzky–Allen scheme of 1985 classifies oxides accordingly into Mott-Hubbard insulators (gap ≈ U, like V₂O₃) and charge-transfer insulators (gap ≈ Δ < U, like the cuprates and NiO). Either way, the gap is interaction-driven.

Why Mott insulators are usually magnets

Localizing one electron per site leaves the spins free. They are not inert. Second-order perturbation theory in t/U — virtual hopping where an electron briefly visits a neighbor and returns — lowers the energy only if the two neighboring spins are antiparallel, because the Pauli principle blocks the virtual move for parallel spins. This produces an effective antiferromagnetic Heisenberg coupling, the superexchange:

J = 4 t² / U     (antiferromagnetic, J > 0)
H_spin = J Σ⟨i,j⟩ S_i · S_j

So at half filling and large U, the Hubbard model reduces to the antiferromagnetic Heisenberg model. This is exactly why most Mott insulators — NiO, MnO, La₂CuO₄ — order antiferromagnetically below a Néel temperature. For La₂CuO₄ the superexchange is enormous, J ≈ 130 meV (≈ 1500 K), one of the largest known. On geometrically frustrated lattices (triangular, kagome) the antiferromagnetic bonds cannot all be satisfied at once, and instead of ordering, the spins may form a quantum spin liquid — a Mott insulator whose spins fluctuate down to absolute zero.

Tuning across the Mott transition

Because the transition is set by U/W, you can drive a material across it by changing either quantity:

• Pressure. Squeezing the lattice increases orbital overlap, raising t and W, lowering U/W. V₂O₃ is metallized by pressure. Conversely, expanding the lattice (chemical pressure, e.g. Cr doping in (V₁₋ₓCrₓ)₂O₃) pushes it into the insulator.
• Temperature. V₂O₃ shows a first-order metal-insulator transition near 150–160 K where the resistivity jumps by 7–8 orders of magnitude — one of the cleanest Mott transitions known.
• Doping (filling control). Adding or removing electrons moves the system off half filling. The doped holes can move without paying U, so even a few percent doping melts the Mott insulator into a (often superconducting) conductor.
• Field / dimensionality. Gating in field-effect transistors and strain in thin films can tip oxides across the transition, the basis of proposed "Mott-tronic" switches and neuromorphic devices using VO₂.

Mott insulator versus band insulator versus Anderson insulator

	Band insulator	Mott insulator	Anderson insulator
Band filling	Full / empty bands	Half-filled band	Partially filled
Cause of gap	Periodic lattice potential	Electron repulsion U	Disorder localization
Need interactions?	No	Yes — defining feature	No (single-particle)
Magnetism	Usually nonmagnetic	Often antiferromagnetic	Can be either
Examples	Diamond, Si, NaCl	NiO, V₂O₃, La₂CuO₄	Doped/disordered semiconductors
Turn off the cause →	Still insulator	Becomes a metal	Becomes a metal

Real materials and numbers

Material	Gap / transition	Notes
NiO	Charge-transfer gap ≈ 4.0 eV	Antiferromagnet, T_N ≈ 525 K; classic Mott/charge-transfer insulator
V₂O₃	First-order MIT near 150 K	Resistivity jumps ~7 orders of magnitude; canonical Mott transition
VO₂	MIT near 340 K (67 °C)	Used in switches/thermochromics; Mott vs Peierls debate
La₂CuO₄	Charge-transfer gap ≈ 2 eV	Antiferromagnet T_N ≈ 320 K; parent of high-Tc cuprates
κ-(BEDT-TTF) organics	Pressure-tuned MIT	Quasi-2D organic Mott systems; candidate spin liquids
Cold-atom Hubbard simulators	U/t tunable, T ~ nK	Optical-lattice fermions realize the Hubbard model directly

Doped Mott insulators and high-Tc superconductivity

The reason Mott insulators are not a museum curiosity is the cuprates. The undoped parent compounds — La₂CuO₄, YBa₂Cu₃O₆ — are antiferromagnetic charge-transfer Mott insulators. Replace a little lanthanum with strontium (La₂₋ₓSrₓCuO₄) and you inject holes into the copper-oxygen planes. At a few percent doping the antiferromagnetism collapses; near optimal doping (x ≈ 0.16) the material superconducts, with critical temperatures up to ~135 K in the mercury cuprates. The strange-metal phase above T_c, the pseudogap, and the d-wave pairing all live in the neighborhood of the parent Mott state. Whether superconductivity emerges from the doped Mott insulator — and exactly how — remains one of the deepest open questions in physics. The Hubbard model on a square lattice, the same two-line Hamiltonian above, is believed to contain the answer, and is still being attacked with quantum Monte Carlo, DMFT, tensor networks, and cold-atom simulators.

How physicists actually compute it

The Mott transition is intrinsically non-perturbative — you cannot expand around either the free electrons or the atomic limit cleanly through the transition. The breakthrough was dynamical mean-field theory (Metzner & Vollhardt 1989; Georges, Kotliar, Krauth, Rozenberg 1996), which maps the lattice problem onto a single impurity in a self-consistent bath and becomes exact in infinite dimensions. DMFT reproduces the three-peak spectral function of the correlated metal — two Hubbard bands plus a central quasiparticle peak — and shows that peak collapsing as U grows, signalling the Mott transition. Combined with density-functional theory (DFT+DMFT), it is the workhorse for real correlated materials today.

Common misconceptions

• "It's just a band insulator with a small gap." No. A band insulator insulates even for non-interacting electrons. A Mott insulator has a partially filled band and would be a metal without interactions. The gap is many-body.
• "The electrons are gone." The electrons are still there — exactly one per site. They simply cannot move because moving requires double occupancy. The charge is frozen; the spin is not.
• "Mott insulators are nonmagnetic like normal insulators." Quite the opposite. Localized spins plus superexchange make most Mott insulators antiferromagnets.
• "It's the same as a Peierls insulator." A Peierls transition opens a gap by lattice distortion (a structural, single-particle effect). The Mott gap is from interactions. VO₂ is famously argued about precisely because both may contribute.
• "U just has to be nonzero." U must beat the bandwidth W. Sodium has plenty of Coulomb repulsion yet is a good metal because its band is wide. The criterion is U ≳ W, not U > 0.
• "Doping a little keeps it insulating." Even small departures from half filling let carriers move without paying U, so light doping typically collapses the Mott insulator into a conductor.