Local Orbital Magnetization

• Local orbital magnetization is a real-space measure of orbital magnetic responses that integrates local markers to reproduce the global moment of insulating or bounded systems.
• Advanced formulations decompose the magnetization into projector-based markers, Berry-curvature contributions, and current-generating fields to elucidate boundary currents and sublattice textures.
• Methodologies combine k-space interpolations with real-space analyses to capture finite-size effects and electric-field responses in diverse crystalline, topological, and correlated materials.

Local orbital magnetization denotes a real-space-resolved orbital magnetic response whose spatial average reproduces the macroscopic orbital magnetization of an insulating or otherwise bounded electronic system. In the modern literature it appears in several closely related but non-identical forms: as a projector-based local marker built from the ground-state density matrix, as a single-site quantity decomposed into local orbital moment and local Berry-curvature terms, and as a current-generating field defined by local-flux response of the grand potential. These formulations are used to analyze boundary currents, local Chern structure, sublattice textures, orbital magnetoelectricity, and finite-size orbitronics in topological, trivial, crystalline, amorphous, fractal, and correlated systems (Bianco et al., 2013, Wang et al., 2022, Saati et al., 13 Dec 2025, Daido, 28 Jun 2026).

1. Formal definitions and decompositions

A central projector-based construction begins from the occupied-state projector

$$\hat P=\sum_{\varepsilon_n<\mu}|\phi_n\rangle\langle \phi_n|,\qquad \hat Q=1-\hat P.$$

For a non-interacting insulator, Bianco and Resta rewrote the bulk orbital magnetization as the spatial average of a local real-space density. Their local magnetization density is

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$

with

$M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$

For a Chern insulator one adds a $\mu$-dependent Středa term through the local Chern marker

$$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$

so that

$$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$

This formulation is explicitly real-space and is intended to express orbital magnetization as a genuine bulk property of insulators (Bianco et al., 2013).

A closely related decomposition, used in finite quantum anomalous Hall bilayers, writes a local marker

$$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$

with LC, IC, and BC denoting local circulation, itinerant circulation, and Berry-curvature contributions. In that notation,

$$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$

$$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$

$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$

The corresponding sample-averaged terms are

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$0

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$1

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$2

This three-term split is gauge invariant term by term and is particularly useful for diagnosing boundary effects in finite Chern systems (Wang et al., 2022).

A later microscopic theory defines the local orbital magnetization density from the local grand-potential density,

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$3

with

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$4

The local orbital magnetization is then

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$5

where $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$6 is a local orbital magnetic moment density and $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$7 is an effective local Berry-curvature density. This formulation is explicitly state resolved and extends local orbital magnetization to single-site and sublattice texture analysis (Saati et al., 13 Dec 2025).

2. Nearsightedness, bulk character, and boundary compensation

In the projector formalism, locality is tied to the exponential decay of the one-particle density matrix in a gapped system. Both $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$8 and $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$9 decay exponentially for $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$0 beyond the localization length $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$1, so each matrix element entering $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$2 depends only on a finite neighborhood of $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$3. This is the sense in which the local magnetization density is “nearsighted.” Boundary conditions change $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$4 only when either argument lies within $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$5 of the boundary, and therefore a large central-region average is insensitive to whether one uses open or periodic boundaries (Bianco et al., 2013).

The finite-size role of boundaries was examined numerically in a bilayer quantum anomalous Hall model with adjustable Chern number at half filling. For an $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$6 bilayer with open boundaries, three prescriptions were compared: a $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$7-space Brillouin-zone formula, a real-space bulk average over an inner block, and a real-space whole-sample average over all sites. In every topological sector studied, $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$8, all three methods converge to the same value up to $M_1=\frac1A\int d^dr\,\mathcal M_1(\mathbf r).$9 as $\mu$0–$\mu$1 (Wang et al., 2022).

The technically important subtlety is the behavior of the Berry-curvature term. For nonzero Chern number, the bulk average retains

$\mu$2

whereas the whole-sample average of the BC term vanishes identically because the local Chern marker in the rim carries the opposite sign and exactly cancels the bulk. Taken alone, that would appear to spoil the whole-sample average. The numerical result is that the rim simultaneously induces a nonzero boundary contribution to $\mu$3 which exactly compensates the lost BC piece,

$\mu$4

Consequently,

$\mu$5

The same work showed that the slope $\mu$6 is captured by both real-space averaging methods, and that even very small clusters, $\mu$7, agree perfectly with the whole-sample real-space formula when compared with the first-principles definition

$\mu$8

This establishes that excising the rim is not required, even at quite small sizes, for the total orbital magnetization in that setting (Wang et al., 2022).

3. Competing local formulations and their relations

The phrase “local orbital magnetization” now covers several constructions with different defining properties.

Formulation	Defining quantity	Distinctive statement
Bianco–Resta marker	$\mu$9 from projectors $$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$0 and local Chern marker	Nearsighted bulk density, boundary-condition independent in the bulk (Bianco et al., 2013)
Local Berry-curvature theory	$$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$1	Reveals topological and geometric sublattice textures; may differ from Bianco–Resta on individual sites (Saati et al., 13 Dec 2025)
Current-generating theory	$$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$2 from local-flux response and natural local marker $$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$3	Generates equilibrium current; site-position moment equals the orbital magnetic quadrupole moment (Daido, 28 Jun 2026)

In the single-site theory, each occupied state contributes a local orbital magnetic moment density

$$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$4

and an effective local Berry-curvature density

$$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$5

The global sum of $$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$6 vanishes, $$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$7, but locally it carries topological and geometric information. For Bloch electrons this leads to a sublattice-resolved Berry curvature

$$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$8

where the “topo” piece integrates to the usual intraband Berry curvature and the “geom” piece is purely interband and sums to zero over sublattices. Numerically, the $$\mathcal F(\mathbf r)=4\pi\,\mathrm{Im}\,\langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle,$$9-space and $$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$0-space onsite magnetizations coincide, but they differ from the Bianco–Resta approach, especially in trivial insulators, because the new $$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$1 retains explicit interband and geometric content (Saati et al., 13 Dec 2025).

A different criterion is imposed by the current-generating theory. There the defining relation is

$$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$2

with

$$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$3

In two dimensions the local-flux response fixes the local orbital magnetization uniquely; in three dimensions it fixes the field only up to a longitudinal ambiguity. The same work defines a coarse-grained “natural local marker” $$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$4 which generates the current to third-derivative order and whose first spatial moment reproduces the orbital magnetic quadrupole moment of a finite sample. On the Haldane model, the cell average of the natural local marker matches the bulk magnetization exactly, while markers of Bianco–Resta, Seleznev–Vanderbilt, and Saati–Hur–Pi fail to reproduce the correct local current in the orbital quadrupole-moment state (Daido, 28 Jun 2026).

A plausible implication is that locality is not a single universal constraint. One formulation may be optimized for bulk sum rules, another for site-resolved geometric texture, and another for exact current generation. The resulting local fields can agree on macroscopic averages while differing qualitatively on individual sites.

4. Electric fields, adiabatic evolution, and magnetoelectric response

A semiclassical theory of adiabatically induced orbital magnetization treats a slowly varying Bloch Hamiltonian $$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$5 and identifies extra current terms of the form $$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$6 and $$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$7. In that framework a gauge-invariant, bulk-defined induced orbital magnetization arises only in two cases: when explicit time dependence is absent, giving the orbital magnetoelectric effect, and when one averages over a period, giving pumped orbital magnetization in insulators. The same theory also yields an electric-field-induced intrinsic orbital magnetization in two-dimensional metals and Chern insulators (Xiao et al., 2020).

For static systems under a perpendicular electric field, the real-space bilayer Chern analysis introduces

$$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$8

with

$$\mathcal M_2(\mathbf r)=\frac{\mu}{\phi_0}\,\mathcal F(\mathbf r), \qquad M(\mu)=\frac1A\int d^dr\,\bigl[\mathcal M_1(\mathbf r)+\mathcal M_2(\mathbf r)\bigr].$$9

The local density is correspondingly

$$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$0

In the absence of $$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$1, bulk and whole-sample averages are both correct in the thermodynamic limit. When $$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$2 and $$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$3, however, the bulk-only average acquires a gauge-dependent shift under $$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$4,

$$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$5

or, in multilayers,

$$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$6

The whole-sample average is free of this gauge shift and is therefore the reliable real-space prescription for Chern insulators under finite electric field (Wang et al., 2023).

The same finite-field framework defines the orbital magnetoelectric polarizability

$$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$7

with

$$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$8

The numerical finding is that the peaks of both the total OMP $$m(\mathbf r)=m_{\mathrm{LC}}(\mathbf r)+m_{\mathrm{IC}}(\mathbf r)+m_{\mathrm{BC}}(\mathbf r),$$9 and the Chern–Simons OMP $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$0 track the strongest field-induced change in the integrated Berry curvature,

$$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$1

This supports the concrete statement that the stronger the response of Berry curvature to electric field, the stronger is the OMP and the CSOMP (Wang et al., 2023).

Adiabatic driving provides a distinct extension. In a two-band honeycomb toy model with a circular optical phonon, the pumped orbital magnetization per cycle is of order the nuclear magneton for typical phonon frequencies, and the predicted signal is comparable to pumped spin magnetization via strong Rashba spin-orbit coupling (Xiao et al., 2020). This suggests a direct link between local orbital magnetization theory and phonon angular momentum.

5. Model systems, textures, and finite geometries

Honeycomb two-band models provide a controlled setting for sublattice-resolved local orbital magnetization. In the topological Haldane phase with Chern number $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$2, one finds $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$3 in the gap, interpreted as orbital ferromagnetism, and the slope $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$4 is quantized, with each sublattice carrying half the Chern number. In the trivial Haldane phase, $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$5 in the bands and $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$6 in the trivial gap, giving an almost antiferromagnetic pattern, while each sublattice still shows a linear slope due to the geometric Berry curvature. In the modified Haldane model, $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$7 and the texture is orbital ferrimagnetic; the nonuniversal slopes in the trivial gap come entirely from $$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$8 (Saati et al., 13 Dec 2025).

Finite topological structures highlight the role of boundary-localized states. Quantum spin Hall nanoislands support robust orbital edge magnetism when the highest occupied Kramers doublet is singly occupied. For Dirac edge states the orbital moment grows linearly with size,

$$m_{\mathrm{LC}}(\mathbf r)\equiv \mathrm{Im}\,\langle \mathbf r|P\,x\,Q\,H\,Q\,y\,P|\mathbf r\rangle,$$9

reaching $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$0 at $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$1. The moment remains nearly full up to $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$2 and is only weakly affected by Anderson disorder up to $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$3, edge-positional disorder, island shape, or crystallographic orientation (Potasz et al., 2015).

Fractal geometries alter orbital magnetization through their boundary hierarchy. In Sierpinski carpets, higher generations create a dense set of edge states and produce a staircase profile with oscillations in magnetization as a function of chemical potential. In Sierpinski triangles, self-similarity creates distinct fractal-induced spectral gaps that appear as constant plateaus in the magnetization; these structures are strongly sensitive to edge termination. In both cases the local-marker and direct-definition methods coincide numerically, and the site-resolved marker concentrates on external and internal contours (Lage et al., 16 Oct 2025).

The Haldane model also serves as a benchmark for the current-generating local theory. In a topological phase with $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$4, $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$5, $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$6, and $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$7, the three-point local magnetization inside a bulk unit cell forms piecewise constant domains whose jumps are fixed by bond currents, while Gaussian smearing of the natural local marker yields edge magnetization and circulating edge currents consistent with the exact bond current (Daido, 28 Jun 2026).

6. First-principles and correlated-material implementations

In crystalline solids, a practical first-principles route uses maximally localized Wannier functions to interpolate the modern-theory formula for orbital magnetization on very fine $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$8 meshes. The gauge-invariant formulation expresses the orbital magnetization in terms of traces over small Wannier-space matrices, typically requiring only on the order of $$m_{\mathrm{IC}}(\mathbf r)\equiv -\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,H\,P\,y\,Q|\mathbf r\rangle,$$9 Wannier functions. Real-space decomposition can then be carried down to individual Wannier functions or unit cells, yielding a local magnetization density

$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$0

with

$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$1

In practice, coarse meshes such as $$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$2–$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$3 are adequate for building the Wannier representation, while the anomalous-Hall-type terms in the itinerant contribution often require fine meshes of order $$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$4 for sub-percent accuracy (Lopez et al., 2011).

Applications to magnetic oxides show that the separation between local and itinerant pieces can be numerically small in the total moment even when the momentum-resolved kernel is highly structured. In insulating perovskite transition-metal oxides, the modern theory splits the orbital magnetization into a local circulation part and an itinerant circulation part, with the site-diagonal contribution

$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$5

For YTiO$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$6, LaMnO$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$7, and orthorhombic and monoclinic YVO$$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$8, the itinerant correction $$m_{\mathrm{BC}}(\mathbf r)\equiv -\frac{e}{2\pi c}\,c(\mathbf r), \qquad c(\mathbf r)\equiv 4\pi\,\mathrm{Im}\,\langle \mathbf r|Q\,x\,P\,y\,Q|\mathbf r\rangle.$$9 is $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$00–$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$01 times smaller than $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$02, so that $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$03 captures more than $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$04 of the net ferromagnetic orbital moment. At the same time, $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$05 follows the behavior of the Chern-invariant kernel and is suppressed only after strong Brillouin-zone cancellation (Nikolaev et al., 2013).

Correlated materials can show the opposite trend, with dynamical correlations greatly enhancing the local-circulation channel. In layered ferromagnetic VI$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$06, the band-theory decomposition

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$07

was generalized through a DMFT Green’s-function expression. The reported values per V are

Method	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$08	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$09
DFT+U	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$10	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$11
DMFT $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$12	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$13	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$14
DMFT $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$15	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$16	$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$17

The total orbital magnetization rises from $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$18 in DFT+U to $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$19 and $$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$20 in DMFT. The interpretation given is that dynamical correlations localize the V-$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$21 electrons more strongly, boost the atomic-like circulating currents, and suppress or reverse the itinerant intercell contribution (Zhou et al., 2020).

A further theoretical extension emphasizes boundary anomalies. By defining an extended velocity

$$\mathcal M_1(\mathbf r)=\frac{e}{\hbar c}\,\mathrm{Im}\Bigl\{ \langle \mathbf r|\hat P\,\hat x\,\hat Q\,\hat H\,\hat Q\,\hat y\,\hat P|\mathbf r\rangle -\langle \mathbf r|\hat Q\,\hat x\,\hat P\,\hat H\,\hat P\,\hat y\,\hat Q|\mathbf r\rangle \Bigr\},$$22

a non-Hermitian reformulation attributes additional boundary contributions to orbital magnetization, predicts an emergent covariant derivative in the one-band approximation, and identifies a many-band boundary term that can become locally giant near band crossings in the presence of Hall voltage or surface charge imbalance (Kyriakou et al., 2018). A plausible implication is that the treatment of boundaries remains an active fault line between bulk-only, whole-sample, and explicitly current-based local theories.

Local orbital magnetization is therefore best regarded as a family of rigorously defined real-space objects, each tied to a specific physical criterion: nearsighted bulk averaging, site-resolved Berry and geometric texture, exact current generation, or interacting-material response. Their shared macroscopic limit is well established, but their local nonequivalence has become an important part of the subject rather than a defect in it (Bianco et al., 2013, Saati et al., 13 Dec 2025, Daido, 28 Jun 2026).