Orbital-Selective Mott Transition

• Orbital-Selective Mott Transition is a phenomenon where, due to strong electron correlations, electrons in certain orbitals become localized while others remain itinerant, resulting in coexisting local moments and metallic carriers.
• It is extensively analyzed using methods such as DMFT, slave-spin formulations, and DMRG, which reveal insights into Fermi-surface reconstruction and the two-stage collapse of coherence.
• Control parameters like orbital bandwidth, crystal-field splitting, and Hund’s coupling enable tuning between metallic states, orbital-selective Mott phases, and complete Mott insulation.

Orbital-selective Mott transition (OSMT) denotes a correlation-driven localization transition in a multi-orbital system in which electrons in one or more orbitals become Mott localized while electrons in the remaining orbitals stay itinerant. The corresponding orbital-selective Mott phase (OSMP) combines localized spin degrees of freedom with metallic quasiparticles and, at zero temperature, can be associated with a Fermi-surface reconstruction because the localized orbital drops out of the Fermi-surface volume (Vojta, 2010). In model studies, OSMT arises through several distinct routes, including bandwidth asymmetry, crystal-field splitting, orbital degeneracy differences, Hund-stabilized orbital differentiation, and, in a particularly restrictive construction, purely through distinct noninteracting densities of states even when bandwidths, crystal-field splitting, and orbital degeneracy are held equal (Song et al., 2014).

1. Definitions and diagnostic criteria

The defining criterion of an OSMT is orbital differentiation in the quasiparticle weight: for some orbital $\alpha$, $Z_\alpha \to 0$, while for other orbitals $\beta$, $Z_\beta>0$ remains finite (Mukherjee et al., 2016). In DMFT-based work, the quasiparticle weight is commonly written as

$$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$

with $\Sigma_\gamma$ the orbital-resolved self-energy (Song et al., 2014). In slave-spin formulations, the same loss of coherence is encoded by

$$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$

so that a vanishing slave-spin expectation value directly signals Mott localization (Mukherjee et al., 2016).

Phase identification is not based on $Z_\alpha$ alone. In DMFT studies of doped two-band models, an OSMP is diagnosed by pinning of the orbital charge at an integer value over a finite crystal-field window together with divergence of $\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$ as $\omega_n\to0$ (Jakobi et al., 2013). Spectrally, the localized orbital has zero weight at the Fermi level and develops Hubbard bands, while the itinerant orbital retains finite low-energy spectral weight (Song et al., 2018). In this sense, the OSMP is distinct both from an ordinary correlated metal, where all orbitals remain coherent, and from a full Mott insulator, where all orbitals are gapped.

The physical content of the OSMP is the coexistence of localized spins and itinerant electrons on the same site (Song et al., 2018). In heavy-fermion language, this is closely related to Kondo breakdown, where the strongly correlated $Z_\alpha \to 0$0 electrons localize and disappear from the Fermi surface; the review literature therefore treats Kondo-breakdown transitions as a form of orbital-selective Mott transition (Vojta, 2010).

2. Minimal Hamiltonians and control parameters

The standard theoretical framework is a multi-orbital Hubbard or Hubbard-Kanamori Hamiltonian with orbital-dependent hopping, crystal-field energies, intraorbital repulsion $Z_\alpha \to 0$1, interorbital repulsion $Z_\alpha \to 0$2, and Hund’s exchange. A representative two-orbital form is

$Z_\alpha \to 0$3

with the rotationally invariant relation $Z_\alpha \to 0$4 used throughout many studies (Mukherjee et al., 2016). Closely related Kanamori forms are employed in iron-based superconductors, ruthenates, vanadates, nickelates, and layered magnetic materials (Yu et al., 2017).

The principal control parameters are the orbital bandwidths $Z_\alpha \to 0$5, crystal-field splittings $Z_\alpha \to 0$6, Hund’s coupling $Z_\alpha \to 0$7, filling, and the degree of interorbital hybridization. Distinct noninteracting densities of states can play the same role as explicit bandwidth asymmetry: in the square-lattice two-orbital model of Song, Lee, and Zhang, one orbital is assigned quasi-1D character and the other quasi-2D character through different hopping anisotropies, while bandwidths are explicitly matched (Song et al., 2014). In realistic models for iron pnictides and chalcogenides, the problem is more subtle because the orbitals are kinetically coupled through nonzero off-diagonal hoppings, so an OSMP must emerge despite bare hybridization rather than in its absence (Yu et al., 2017).

Filling is equally important. Some studies focus on half-filled models, where Mott criticality is most direct (Song et al., 2014, Song et al., 2018). Others emphasize non-integer fillings close to half filling, where strain or doping shifts selected orbitals toward half filling and thereby drives selective localization (Mukherjee et al., 2016, Wang et al., 2015). In iron-based models, vacancy order, chemical substitution, and disorder provide additional tuning parameters that reshape the OSMP boundaries (Yu et al., 2012, Liu et al., 2016).

3. Microscopic mechanisms

Early routes to OSMT invoked three ingredients: different orbital bandwidths, large crystal-field splitting, or differing orbital degeneracies. Song, Lee, and Zhang constructed a half-filled two-orbital Hubbard model in which all three are explicitly removed and showed that distinct noninteracting densities of states alone suffice to produce an OSMP, provided the full Hund’s coupling is active (Song et al., 2014). For isotropic Hund’s coupling, $Z_\alpha \to 0$8, and DOS anisotropy parameter $Z_\alpha \to 0$9, they found a two-stage collapse of quasiparticle weights: for $\beta$0 both orbitals are metallic, at $\beta$1 one orbital localizes while the other remains metallic, and at $\beta$2 both orbitals become insulating. In the anisotropic Hund’s case, $\beta$3 and $\beta$4, the two orbitals instead undergo a single simultaneous Mott transition at $\beta$5 (Song et al., 2014).

The mechanism identified in that work is not decoupling of interorbital charge degrees of freedom. Interorbital charge fluctuations are strongly suppressed both at a single Mott transition and in the OSMP, so charge suppression alone does not distinguish the selective phase (Song et al., 2014). The operative process is the formation of local spin triplet states followed by a two-stage breakdown of the Kondo effect. With full Hund’s coupling, the $\beta$6 triplet channel remains available and allows antiparallel-spin fluctuations in the still-metallic orbital, delaying its localization to a larger interaction strength; with Ising-only Hund’s coupling, this channel is absent and the Kondo singlet in the second orbital collapses simultaneously (Song et al., 2014).

Bandwidth-controlled studies sharpen the role of Hund’s anisotropy. In DMFT for a half-filled two-orbital model with $\beta$7 and $\beta$8, the OSM phase occurs between $\beta$9 and $Z_\beta>0$0 for full Hund’s coupling, but between $Z_\beta>0$1 and $Z_\beta>0$2 for Ising Hund’s coupling (Song et al., 2018). The wide orbital is then a Fermi liquid in the full-Hund case, because a low-energy peak survives in the local dynamical spin susceptibility, but a non-Fermi liquid in the Ising case, because that low-energy peak disappears and the Kondo scale collapses to zero (Song et al., 2018).

In multiorbital models with explicit interorbital hybridization, the mechanism is often formulated differently. Yu and Si developed a Landau free-energy functional

$Z_\beta>0$3

with $Z_\beta>0$4 generated by the bare kinetic hybridization and its renormalization by intersite spin correlations (Yu et al., 2017). In this picture, intersite spin correlations are crucial to the renormalization of the bare hybridization amplitude toward zero, thereby allowing one orbital to decouple dynamically and localize even though the microscopic Hamiltonian contains nonzero orbital mixing (Yu et al., 2017).

Doping introduces a further mechanism. In equal-bandwidth multiband Hubbard models with crystal-field splitting, doping electrons into the lifted band can leave the lower bands pinned at half filling over a finite $Z_\beta>0$5 window. Hund’s coupling then increases the total local moment, and the resulting frozen moments enhance the effective correlation in the half-filled sector until the lower orbitals undergo a Mott transition while the doped band remains metallic (Wang et al., 2015). In Cu-substituted iron-based superconductors, Hund’s coupling first drives the $Z_\beta>0$6 orbital toward half filling and opens a Mott-Hubbard gap once $Z_\beta>0$7, after which impurity-induced disorder produces localized in-gap states and an orbital-selective insulating phase (Liu et al., 2016).

4. Theoretical methods and quantitative diagnostics

Single-site DMFT remains the central nonperturbative framework. In the DOS-driven two-orbital model, each orbital maps onto an independent two-impurity Anderson problem coupled only by on-site interactions, and the zero-temperature impurity problem is solved by exact diagonalization with 6 bath orbitals per impurity (Song et al., 2014). Other DMFT studies employ Hirsch-Fye quantum Monte Carlo for doped two-band models with crystal-field splitting (Jakobi et al., 2013), hybridization-expansion CT-QMC at $Z_\beta>0$8 with typical sampling $Z_\beta>0$9 sweeps per DMFT iteration in equal-bandwidth multiband models (Wang et al., 2015), and CT-HYB in realistic DFT+DMFT calculations for Fe-based materials (Bai et al., 2022).

Slave-spin approaches provide a complementary mean-field description of selective coherence loss. In the U(1) slave-spin formalism, the physical electron is factorized as $$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$0, with a local constraint tying the slave spin to the spinon occupation (Mukherjee et al., 2016). This method has been used both for strain-engineered VO$$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$1 (Mukherjee et al., 2016) and for five-orbital models of iron pnictides and chalcogenides (Yu et al., 2017, Yu et al., 2012). Variational Monte Carlo with non-magnetic Jastrow-Slater wave functions reaches a similar phase structure in two dimensions and diagnoses orbital selectivity through the small-$$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$2 behavior of orbital density structure factors, with $$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$3 for a metal and $$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$4 for an insulator (Tocchio et al., 2015).

Methods beyond local mean-field and impurity descriptions expose additional structure. DMRG in a one-dimensional three-orbital Hund’s metal reveals multiple OSMPs and a quantum phase transition between them that is driven by charge fluctuations and the emergence of free spinless fermions rather than by magnetic rearrangement (Rincon et al., 2014). Cluster exact diagonalization plus cluster perturbation theory has been used to model the orbital-selective metal skin induced by alkali dosing of Ca$$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$5RuO$$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$6, where selective hybridization with a surface $$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$7 level generates coherent in-gap states (Horio et al., 2023). D-TRILEX, a diagrammatic extension of DMFT retaining local one- and two-particle vertices while adding nonlocal dual corrections, has been used to study the effect of spatial magnetic fluctuations on the OSMT in three dimensions (Stepanov, 2022).

Recent work also introduced explicitly quantum-information-based diagnostics. In a half-filled non-hybridized two-band Hubbard model, the single-site von Neumann entanglement entropy $$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$8 displays singular behavior at both the metal$$Z_\gamma=\Bigl[1-\frac{\partial \mathrm{Re}\,\Sigma_\gamma(\omega)}{\partial \omega}\Big|_{\omega\to 0}\Bigr]^{-1},$$9OSM and OSM$\Sigma_\gamma$0Mott transitions and distinguishes the order of those transitions (Song et al., 2018). A later DMFT study proposed the local two-qubit fidelity, defined from the orbital impurity density matrix, as a sharply resolved marker of the two successive critical interactions; in that formulation, non-semi-integer values of the fidelity inside the OSMP diagnose Hund’s-coupling-induced quantum entanglement, whereas no such entanglement appears for $\Sigma_\gamma$1 (Niu et al., 2023).

5. Phase diagrams, competing tendencies, and nontrivial variants

The simplest phase diagrams contain three regimes: a metal at small interaction, an OSMP at intermediate interaction, and a full Mott insulator at larger interaction. This structure appears in unequal-bandwidth two-band models in DMFT (Song et al., 2018), in non-magnetic variational calculations on the square lattice (Tocchio et al., 2015), and in five-orbital slave-spin studies of alkaline iron selenides (Yu et al., 2012). In K$\Sigma_\gamma$2Fe$\Sigma_\gamma$3Se$\Sigma_\gamma$4, for example, the five-orbital model at filling $\Sigma_\gamma$5 and $\Sigma_\gamma$6 is metallic for $\Sigma_\gamma$7, enters an OSMP at $\Sigma_\gamma$8 when $\Sigma_\gamma$9 vanishes, and becomes a full Mott insulator for $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$0 (Yu et al., 2017). In the vacancy-ordered variant, both critical interaction scales are reduced because the effective bandwidths shrink (Yu et al., 2012).

Crystal-field splitting enriches this structure rather than merely shifting it. In the doped two-band Hubbard model treated by DMFT plus QMC, moderate doping and suitable $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$1 stabilize not only the usual narrow-band OSMP but also a wide-band OSMP, in which the wide band is pinned at integer occupancy while the narrow band remains metallic (Jakobi et al., 2013). In equal-bandwidth multiband models with crystal-field splitting, a large OSMP region emerges once the filling exceeds a critical value, with approximate boundaries such as $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$2 in the two-band model at $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$3 and $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$4 (Wang et al., 2015). These results show that crystal-field tuning can be as effective as explicit bandwidth disparity in generating partial localization.

Finite temperature introduces further complexity. In a two-orbital model with Ising-type Hund’s coupling, the OSMP$$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$5Mott-insulator boundary is a first-order line with positive slope, described as a slope-reversed Mott transition (Kim et al., 2015). For $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$6, the critical endpoint lies at approximately $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$7 and $$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$8, and increasing Hund’s coupling lowers the endpoint until the first-order line turns into a finite-temperature crossover (Kim et al., 2015). In the same regime, the wide orbital develops frozen local moments, visible as a long-$$Z_\alpha = |\phi_\alpha|^2 = |\langle S^+_{i\alpha\sigma}\rangle|^2,$$9 plateau in the imaginary-time spin-spin correlator (Kim et al., 2015).

Not all orbital selectivity is a simple metal$Z_\alpha$0OSMP$Z_\alpha$1Mott sequence. In a one-dimensional three-orbital Hund’s model at $Z_\alpha$2, DMRG finds a quantum phase transition between distinct orbital-selective Mott states: OSMP1, with one localized and two metallic orbitals, gives way at stronger coupling to OSMP2 or OSMP3, depending on filling, and the strong-coupling itinerant sector acquires the universal Luttinger exponent $Z_\alpha$3 characteristic of free spinless fermions (Rincon et al., 2014). This transition is preempted by charge fluctuations rather than by a reorganization of magnetic order (Rincon et al., 2014).

A central controversy concerns the robustness of OSMT once nonlocal collective fluctuations are included. In a half-filled two-orbital Hubbard-Kanamori model on the three-dimensional cubic lattice, single-site DMFT yields the standard OSMT, with the narrow band localizing first and the wide band remaining metallic over an intermediate temperature range (Stepanov, 2022). However, D-TRILEX finds that strong antiferromagnetic fluctuations drive all orbital spin-susceptibility components to diverge at the same temperature, thereby preempting the OSMT and replacing the orbital-selective pocket with a single N\'eel transition (Stepanov, 2022). This suggests that the local-theory phase diagram may be unstable in regimes dominated by strong spatial magnetism.

6. Materials realizations and experimental manifestations

The ruthenates provide a longstanding reference point. In Ca$Z_\alpha$4Sr$Z_\alpha$5RuO$Z_\alpha$6, early OSMT proposals emphasized unequal bandwidths of the Ru $Z_\alpha$7 and $Z_\alpha$8 bands, but the DOS-driven analysis of Song, Lee, and Zhang shows that even if the bandwidths are equal, the quasi-2D $Z_\alpha$9 density of states versus the quasi-1D $\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$0 density of states can drive an OSMP (Song et al., 2014). A different realization appears in alkali-metal-dosed Ca$\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$1RuO$\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$2: ARPES shows that surface dosing produces a single-band metal skin, while homogeneous electron doping in DMFT gives a three-band metal with all $\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$3, so the observed state is attributed instead to an orbital-selective Mott-insulator breakdown driven by selective hybridization with the surface alkali $\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$4 level (Horio et al., 2023).

Iron-based superconductors host some of the most extensively modeled OSMPs. In K$\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$5Fe$\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$6Se$\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$7, the five-orbital slave-spin description identifies the $\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$8 orbital as the first to localize and places the OSMP as a bridge between the superconducting metallic regime and the Mott-insulating parent compound (Yu et al., 2012). More generally, the multiorbital models for iron pnictides and chalcogenides exhibit an OSMP anchored by the collapse of $\lvert \mathrm{Im}\,\Sigma_\alpha(i\omega_n)\rvert$9 while $\omega_n\to0$0 remains finite, with the OSMP window widening as $\omega_n\to0$1 increases (Yu et al., 2017). In Cu-substituted iron-based superconductors, disorder does not merely broaden spectra: it fills the $\omega_n\to0$2 Mott gap with strongly localized bound states and, together with Hund’s coupling, generates an orbital-selective insulating phase in which $\omega_n\to0$3 is Mott-localized while $\omega_n\to0$4 are Anderson-localized (Liu et al., 2016).

Strain and dimensional confinement provide direct external tuning knobs. In epitaxial rutile VO$\omega_n\to0$5/TiO$\omega_n\to0$6 films, hard x-ray photoelectron spectroscopy and V $\omega_n\to0$7-edge x-ray absorption spectroscopy show strain-induced modulation of electron correlations and increased orbital anisotropy, while U(1) slave-spin calculations argue that c-axis elongation narrows the $\omega_n\to0$8 band, raises $\omega_n\to0$9 relative to $Z_\alpha \to 0$00, drives $Z_\alpha \to 0$01, and can access an OSMT at non-integer filling close to half filling (Mukherjee et al., 2016). In monolayer LaNiO$Z_\alpha \to 0$02, ARPES observes orbital-selective decoherence of spectral density as thickness is reduced, with the spectral weight of the $Z_\alpha \to 0$03 band vanishing much faster than that of the $Z_\alpha \to 0$04 band; DFT+DMFT attributes this to localization of $Z_\alpha \to 0$05 electrons along the $Z_\alpha \to 0$06 axis in the monolayer limit (Sohn et al., 23 Sep 2025).

Layered magnetic and strongly crystal-field-split systems reveal additional variants. In Fe$Z_\alpha \to 0$07GeTe$Z_\alpha \to 0$08, neutron scattering and thermodynamic measurements show that antiferromagnetic spin fluctuations coexist with ferromagnetism, and realistic charge-self-consistent DFT+DMFT attributes this to an OSMT in which the plane-perpendicular $Z_\alpha \to 0$09 orbital remains itinerant while the narrower $Z_\alpha \to 0$10 sector localizes (Bai et al., 2022). In La$Z_\alpha \to 0$11O$Z_\alpha \to 0$12Fe$Z_\alpha \to 0$13Se$Z_\alpha \to 0$14, enhanced crystal-field splitting and smaller orbital-resolved kinetic energies place the material near orbital-selective metallization, and both slave-spin mean field and DMFT indicate that doping or uniaxial pressure can drive an orbital-selective Mott state in which only one or a few orbitals are metallized (Giovannetti et al., 2014). Song, Lee, and Zhang further proposed VOCl under pressure as a clean two-orbital candidate because the crystal-field splitting between its lowest $Z_\alpha \to 0$15 and $Z_\alpha \to 0$16 bands is negligible and the bandwidths are similar (Song et al., 2014).

Across these materials, the recurring experimental signatures are orbital-dependent loss of coherence, selective redistribution of spectral weight rather than uniform gap collapse, and the coexistence of local moments with itinerant carriers. ARPES, HAXPES, XAS, neutron scattering, transport, and thermodynamic probes therefore access complementary aspects of the same phenomenon: selective Mottness in a multiorbital setting (Mukherjee et al., 2016, Bai et al., 2022, Horio et al., 2023). A plausible implication is that the most robust experimental identification of OSMT requires combining orbital resolution with a direct measure of whether coherence is lost in only a subset of channels.