betaFeedback

WorksPhysics and Astronomy

Local orbital magnetization and Chern markers from equilibrium local currents

Abstract

Local currents carry the orbital magnetization of a macroscopic sample and are key for modeling inhomogeneous magnetic materials and for understanding magnetoelectric and topological responses. However, producing realistic simulated images of local currents in inhomogeneous systems remains challenging. Building upon recent work on stochastic estimation methods for tight-binding models, we show that all matrix elements of positive semidefinite operators defined on a lattice can be obtained at once from a single set of filtered random vectors. This eliminates the need for independent simulations previously required for off-diagonal elements, which significantly speeds up the calculation of spatial maps of local currents. Further, we express bond currents in terms of loop currents that obey a discrete Poisson equation, with the current vorticity acting as the source term. Its solution yields a local orbital magnetization. In analogy with the global Středa relation, the derivative of this magnetization with respect to the chemical potential defines a new local Chern marker. Built entirely from bond currents, it requires no position operator and is not bound by the zero-sum rule of projector markers in open samples. Hence, it is free of boundary-compensation artifacts, and its sample average converges to the bulk Chern number in a controlled fashion.

Introduction

Spatially resolved local observables reveal physical information that global measurements inevitably average away. In 2D crystals, chiral edge channels [1], current vortices around defects [2, 3] and localization transitions [4, 5] are invisible to net current measurements. Scanning magnetometry has recently proven its suitability to directly probe these currents, and the magnetization they generate, with fine spatial resolution [1, 6, 7], down to the moiré scale, where orbital ferromagnets break up into domains of opposite Chern number [8]. Scanning tunneling spectroscopy has also been proposed as a probe of loop-current order, which produces weak local magnetism in cuprates, kagome metals, and at the surface of Sr2_2RuO4_4 [9].

Orbital magnetization links equilibrium local currents to topological phases [10, 11, 12, 13]. These observables are divergence-free and carry no net transport, so by the Helmholtz decomposition they manifest as the curl of a magnetization density. The Středa relation [14] links the chemical-potential derivative of the global magnetization to the Hall conductivity, which in insulators is quantized by the Chern number. This invariant is defined as a Brillouin-zone integral of the Berry curvature. Without translation symmetry, crystal momentum is no longer a good quantum number, requiring the Chern number to be defined in real space. Local Chern markers [15, 16, 17] have become the standard tool for probing topology in real space, enabling the resolution of Chern mosaics in quasi-periodic systems [8, 18] and the investigation of disorder-driven topological transitions [19, 20, 21, 22]. However, they rely on the position operator, and in a finite sample with open boundaries the marker, being the trace of a commutator of finite matrices, sums identically to zero. This behavior is guaranteed by a boundary contribution that grows with the system size and exactly compensates the bulk value. Markers and other topological indicators [23, 24, 25, 26, 27] that avoid this compensation have recently been proposed. In parallel, the local orbital magnetization has been formulated as a nearsighted property of the ground-state density matrix [28], as the field whose curl generates the equilibrium current [29], and through the field response of the local density of states [30]. On a lattice, the magnetization density of a plaquette is the response to a flux threaded through it, and differences of these densities give the bond currents [31].

Comparing local markers with scanning-probe images demands samples large enough to approach the thermodynamic limit while capturing the disorder present in experiments [8, 32, 33, 34, 35]. Since both local currents and topological markers are matrix elements of functions of the Hamiltonian, Chebyshev expansions [36, 37, 38, 39, 40, 41, 42, 43, 44, 45] serve as the natural tool, and have recently been applied to the global orbital magnetization of large disordered samples [46]. In Ref. [5], all diagonal elements of a spectral operator were sampled at once with a positive semidefinite estimator whose fluctuations are bounded by the Markov inequality. Off-diagonal elements, however, necessitate rotating disjoint pairs of sites, at the cost of two simulations (real and imaginary parts) per tiling of the lattice into pairs.

The purpose of this paper is twofold. First, we introduce an algorithmic contribution to spectral methods: We show that a single set of filtered random vectors gives every off-diagonal element of a spectral operator. This removes the multiplicative cost of the tilings without computational time or memory overhead. Consequently, a complete map of local currents requires a single Chebyshev expansion per random vector, the same as the diagonal elements alone. Second, we put forward a new local Chern marker: Using these maps we construct a local orbital magnetization built entirely from bond currents, whose chemical-potential derivative yields a local Chern marker that bypasses the position operator altogether. Because it is built from currents rather than from position-operator commutators, our marker obeys no zero-sum rule. Across an interface it interpolates between the bulk values instead of overshooting, and for a homogeneous sample its average converges to the Chern number. It is independent of the choice of origin at any temperature and energy resolution, and its steps and extrema give the sense of circulation of the current.

For pedagogical reasons, the manuscript follows the order in which the two results build on each other. We first present the numerical improvement, and then the local Chern marker that it makes possible. Section II shows that the pair rotations used for the off-diagonal elements can be applied after the spectral filter, so that one set of random vectors gives all bond currents. We then turn to the marker. Section III formalizes the loop currents on a square lattice, where their construction is most transparent, and from them a local orbital magnetization, whose chemical-potential derivative defines the local Chern marker of Sec. III A. We test our findings on a system of noninteracting electrons on a disordered square lattice in the integer quantum Hall regime, both in the mobility gaps and at the center of a Landau level (Sec. IV). We then extend the construction to a two-dimensional crystal, i.e. a lattice with a general basis and hoppings of arbitrary range (Sec. V), and apply it to a Chern domain in the Haldane model [47], where it is compared with the projector marker at a domain wall (Sec. VI).

Bond currents from one set of filtered vectors

The complete map of equilibrium local currents in tight-binding models constitutes the fundamental construction for the results presented in this work. Each bond current is the imaginary part of an off-diagonal element of a function of the Hamiltonian, weighted by a hopping. In Ref. [5], each such element was obtained by rotating its disjoint pair of sites into a basis where it appears as a difference of diagonal matrix elements. This approach demanded a separate Chebyshev expansion for every tiling of the lattice into disjoint pairs, and additionally for the real and the imaginary parts. We show here that one set of filtered random vectors gives every matrix element, and hence every bond current, while each current remains a difference of two non-negative samples whose fluctuations are set by the local spectral weight.

We consider a generic tight-binding Hamiltonian

H=∑a,btabca†cb,(1)H = \sum_{a,b} t_{ab} c_a^\dagger c_b, \tag*{(1)}

where ca†c_a^\dagger is the creation operator for an atomic-like orbital φa\varphi_a at position ra\mathbf{r}_a. Here, tabt_{ab} corresponds to either a hopping amplitude between a pair of orbitals (a≠ba \ne b) or an on-site energy (a=ba=b). The information regarding typical equilibrium properties such as energy and magnetization is encoded in the matrix elements of positive semidefinite Hermitian spectral operators, X=g(H)X = g(H), with g(…)g(\ldots) a real function of the Hamiltonian. Important examples include the Fermi–Dirac density matrix f(H)=[1+exp⁡(β(H−μ))]−1f(H) = [1+\exp(\beta(H-\mu))]^{-1} at temperature T=(kBβ)−1T=(k_{\mathrm{B}}\beta)^{-1} and the spectral window, δη(ε−H)\delta_\eta(\varepsilon-H), a normalized Gaussian of standard deviation η\eta that broadens the Dirac delta. Defining Xab=⟨a∣X∣b⟩X_{ab}=\langle a|X|b\rangle and xa=Xaax_a=X_{aa}, the equilibrium current of particles of charge ee from aa to bb can be cast as

Iab=2eℏTr⁡(XWab),(2)I_{ab} = \frac{2e}{\hbar}\operatorname{Tr}(XW_{ab}), \tag*{(2)}

where Wab=i2(tab∣a⟩⟨b∣−h.c.)W_{ab}=\frac{i}{2}(t_{ab}|a\rangle\langle b|-\mathrm{h.c.}) is the traceless Hermitian bond operator.

The stochastic estimation of the local current maps starts by generating a set of RR random vectors

∣ξn⟩=∑aeiθn,a∣a⟩,n=1,…,R,(3)|\xi_n\rangle= \sum_a e^{i\theta_{n,a}}|a\rangle,\qquad n=1,\ldots,R, \tag*{(3)}

with independent phases drawn uniformly in [0,2π)[0,2\pi). Since ξn,a∗ξn,b‾=δab\overline{\xi_{n,a}^*\xi_{n,b}}=\delta_{ab}, where the overline denotes the average over random vectors, we have that

∣vn⟩⟨vn∣‾=X,(4)\overline{|v_n\rangle\langle v_n|}=X, \tag*{(4)}

with ∣vn⟩=X∣ξn⟩|v_n\rangle=\sqrt{X}|\xi_n\rangle. From these filtered vectors, we can compute the full set of diagonal matrix elements, xa=∣vn,a∣2‾x_a=\overline{|v_{n,a}|^2}, as a mean of non-negative samples whose standard deviation is bounded by the mean [5].

Off-diagonal elements are estimated as differences of diagonal elements in a rotated basis. For a pair of orbitals {a,b}\{a,b\} and a phase ∣γ∣=1|\gamma|=1, the states

∣+γ⟩=∣a⟩+γ∣b⟩2,∣−γ⟩=−γ∗∣a⟩+∣b⟩2(5)|+\gamma\rangle= \frac{|a\rangle+\gamma|b\rangle}{\sqrt{2}},\qquad|-\gamma\rangle= \frac{-\gamma^*|a\rangle+|b\rangle}{\sqrt{2}} \tag*{(5)}

yield

γγ≡⟨+γ∣X∣+γ⟩−⟨−γ∣X∣−γ⟩=2Re⁡(γXab).(6)\gamma_\gamma\equiv\langle+\gamma|X|+\gamma\rangle-\langle-\gamma|X|-\gamma\rangle= 2\operatorname{Re}(\gamma X_{ab}). \tag*{(6)}

In the original scheme, all elements of a tiling (T\mathcal{T}) of the lattice into disjoint pairs were rotated all at once by a block-diagonal unitary UTU_{\mathcal{T}}, whose columns are the states ∣±γ⟩|\pm\gamma\rangle of each pair {a,b}\{a,b\}. The diagonal elements of UT†XUTU_{\mathcal{T}}^\dagger XU_{\mathcal{T}} were sampled as in Eq. (4), with X\sqrt{X} replaced by UT†XUTU_T^\dagger\sqrt{X}U_T. The phases γ=1\gamma=1 and γ=i\gamma=i give the real and imaginary parts of XabX_{ab}, and with tba=∣tab∣eiθbat_{ba}=|t_{ab}|e^{i\theta_{ba}} the current was assembled as

ℏeIab=∣tab∣(sin⁡θba Υ1−cos⁡θba Υi).(7)\frac{\hbar}{e}I_{ab}=|t_{ab}|\left(\sin\theta_{ba}\,\Upsilon_1-\cos\theta_{ba}\,\Upsilon_i\right). \tag*{(7)}

Each orbital belongs to exactly one pair of a tiling, so zz bonds per orbital require NT≥zN_T\ge z tilings (NT=4N_T=4 on the square lattice, NT=9N_T=9 for the Haldane model). As each tiling required its own expansion, for each random vector we had to perform 2NT2N_T independent simulations to obtain the complete map of local currents. In the following paragraphs we show how this multiplicative cost can be eliminated exactly.

The first step consists of rotating into the eigenbasis of WabW_{ab}, rather than computing the real and imaginary parts of XabX_{ab} independently. On a particular pair, {a,b}\{a,b\}, the structure of the local current is encoded in

Wab=12(0itab−itba0),(8)W_{ab}=\frac{1}{2}\begin{pmatrix}0&it_{ab}\\-it_{ba}&0\end{pmatrix}, \tag*{(8)}

with eigenvalues ±∣tab∣/2\pm|t_{ab}|/2. Its eigenvectors are the states of Eq. (5) evaluated at the bond-dependent phase

γab=−itba∣tab∣,(9)\gamma_{ab}=-i\frac{t_{ba}}{|t_{ab}|}, \tag*{(9)}

so that Wab=12∣tab∣(∣+γab⟩⟨+γab∣−∣−γab⟩⟨−γab∣)W_{ab}=\frac{1}{2}|t_{ab}|(\lvert+\gamma_{ab}\rangle\langle+\gamma_{ab}\rvert-\lvert-\gamma_{ab}\rangle\langle-\gamma_{ab}\rvert). Using Eq. (6) we obtain

Iab=eℏ∣tab∣Υγab.(10)I_{ab}=\frac{e}{\hbar}|t_{ab}|\Upsilon_{\gamma_{ab}}. \tag*{(10)}

Since the pairs of a tiling are disjoint, this rotation with a bond-dependent phase can be applied simultaneously across the lattice, halving the number of expansions required by Eq. (7).

The second step removes the multiplicative cost altogether. It follows from comparing two estimators of the same rotated diagonal element ⟨u∣X∣u⟩\langle u|X|u\rangle, with ∣u⟩=UT∣a⟩|u\rangle=U_T|a\rangle. In the original scheme, the rotation acts twice, on both sides of X\sqrt{X}, sampling the off-diagonal matrix elements contained in the tiling TT through the estimator

ϕn,u(T)=⟨ξn∣∘UT†XUT∣ξn⟩,(11)\phi_{n,u}^{(T)}=\langle\xi_n|\circ U_T^\dagger\sqrt{X}U_T|\xi_n\rangle, \tag*{(11)}

with ⟨ψ1∣∘∣ψ2⟩=ψ1∗(a)ψ2(a)\langle\psi_1|\circ|\psi_2\rangle=\psi_1^*(a)\psi_2(a) denoting element-wise multiplication, whose component aa we label by the rotated state ∣u⟩|u\rangle that it samples. Instead, the rotation can act once, after the spectral operator, on the same vector for every tiling,

χn,u(T)=⟨ξn∣∘UT†X∣ξn⟩.(12)\chi_{n,u}^{(T)}=\langle\xi_n|\circ U_T^\dagger\sqrt{X}|\xi_n\rangle. \tag*{(12)}

For a fixed unitary transformation the estimators of Eqs. (11) and (12) are both unbiased and converge to the same diagonal matrix element. Since UTU_T is unitary and UT∣ξn⟩⟨ξn∣UT†=1U_T|\xi_n\rangle\langle\xi_n|U_T^\dagger=\mathbb{1}, we have that

∣ϕn,u(T)∣2=∣χn,u(T)∣2,(13)\left|\phi_{n,u}^{(T)}\right|^2=\left|\chi_{n,u}^{(T)}\right|^2, \tag*{(13)}

while the error bound associated with the diagonal estimator of Eq. (11) carries over to χn,u(T)\chi_{n,u}^{(T)}. This is easily verified by computing its variance

Var⁡(∣χn,u(T)∣2)=⟨u∣X∣u⟩2−∑c∣⟨u∣X∣c⟩∣4,(14)\operatorname{Var}\left(\left|\chi_{n,u}^{(T)}\right|^2\right)=\langle u|X|u\rangle^2-\sum_c\left|\langle u|\sqrt{X}|c\rangle\right|^4, \tag*{(14)}

signifying that every diagonal matrix element in the rotated basis is again a mean of non-negative samples whose standard deviation does not exceed the mean.

The bond current, however, is the difference of two such non-negative samples drawn from the same random vector [Eq. (10)]. While each rotated diagonal element retains the error bound of the diagonal estimator, the relative error of the current is not bounded in the same way, depending both on the model, and on the local observable being computed. This subtraction was already present in the original scheme. The gain is therefore entirely in computational efficiency. For the computational time needed for a single map in Ref. [5], which required 2NT2N_T expansions per random vector, the present scheme averages over 2NT2N_T times more random vectors, reducing the variance of every local current field by the same factor.

Loop currents and a current-based Chern marker on the square lattice

The systematic construction of the loop current field begins with the Helmholtz decomposition of a planar current field,

I=Idiv+Icirc+Itr.(15)\mathbf{I}=\mathbf{I}_{\mathrm{div}}+\mathbf{I}_{\mathrm{circ}}+\mathbf{I}_{\mathrm{tr}}. \tag*{(15)}

Here Idiv=∇ζ\mathbf{I}_{\mathrm{div}}=\nabla\zeta, with ∇2ζ=∇⋅I\nabla^{2}\zeta=\nabla\cdot\mathbf{I}, carries the sources and sinks, Icirc=∇×(Mez)\mathbf{I}_{\mathrm{circ}}=\nabla\times(M\mathbf{e}_{z}) the circulation, and the harmonic component Itr=∇v\mathbf{I}_{\mathrm{tr}}=\nabla v, with ∇2v=0\nabla^{2}v=0, a transport current. Equilibrium currents are steady, Idiv=0\mathbf{I}_{\mathrm{div}}=0, and carry no net transport, Itr=0\mathbf{I}_{\mathrm{tr}}=0, so that

I=∇×(Mez),(16)\mathbf{I}=\nabla\times(M\mathbf{e}_{z}), \tag*{(16)}

or equivalently

(∇×I)z=−∇2M,(17)(\nabla\times\mathbf{I})_{z}=-\nabla^{2}M, \tag*{(17)}

with M=0M=0 outside a sample with open boundaries. On a tight-binding lattice, charge is defined on the nodes, while local currents flow through the connecting bonds. On the square lattice we define loop currents M(r)M(\mathbf{r}) on the plaquettes, each labeled by its bottom-left corner r=(x,y)\mathbf{r}=(x,y).

Each bond current is the difference of the loop currents of the two plaquettes that share the bond, the lattice form of Ampère’s law [31] (Figure 1),

Figure 1. Loop currents on the square lattice. Each plaquette, labeled by its bottom-left corner r=(x,y)\mathbf{r}=(x,y), carries a loop current M(r)M(\mathbf{r}) defined at the geometric center of the square. Each bond current is the difference of the loop currents on its two sides: Ix(r)=M(r)−M(x,y−1)I_x(\mathbf{r})=M(\mathbf{r})-M(x,y-1) and Iy(r)=M(x−1,y)−M(r)I_y(\mathbf{r})=M(x-1,y)-M(\mathbf{r}) [Eq. (18)]. Outside the sample (shaded), M=0M=0.

Ix(r)=M(r)−M(x,y−1),Iy(r)=M(x−1,y)−M(r),(18)\begin{aligned} I_x(\mathbf{r})&=M(\mathbf{r})-M(x,y-1),\\ I_y(\mathbf{r})&=M(x-1,y)-M(\mathbf{r}), \tag*{(18)} \end{aligned}

with Ix(r)I_x(\mathbf{r}) and Iy(r)I_y(\mathbf{r}) being the currents from r\mathbf{r} to r+ex\mathbf{r}+\mathbf{e}_x and to r+ey\mathbf{r}+\mathbf{e}_y, respectively. Equation (18) can be solved exactly only if the current is conserved at every site, which is also the condition for the loop currents to be independent of the path along which Eq. (18) is integrated [31],

∑bIab[X]=0,(19)\sum_b I_{ab}[X]=0, \tag*{(19)}

which represents the lattice statement of Idiv=0I_{\mathrm{div}}=0. In that case the loop current is obtained directly as

M(r)=∑y′≤yIx(x,y′),(20)M(\mathbf{r})=\sum_{y'\le y}I_x(x,y'), \tag*{(20)}

or equivalently, by Eq. (19),

M(r)=−∑x′≤xIy(x′,y).(21)M(\mathbf{r})=-\sum_{x'\le x}I_y(x',y). \tag*{(21)}

Summing by parts with M=0M=0 outside the sample gives

∫M d2r=12∫(r×I)z d2r,(22)\int M\,d^{2}r=\frac{1}{2}\int(\mathbf{r}\times\mathbf{I})_{z}\,d^{2}r, \tag*{(22)}

which corresponds to the orbital magnetization of a finite sample [46, 48, 49]. Hence MM is a local orbital magnetization. Unlike the integrand 12(r×I)z\frac{1}{2}(\mathbf{r}\times\mathbf{I})_{z}, which changes under a shift of origin and becomes origin-independent only once integrated, M(r)M(\mathbf{r}) is fixed by the bond currents alone. It involves no position operator, and its local

definition is independent of the choice of origin. However, stochastic currents are conserved only on average rendering Eq. (18) overdetermined. We solve it using least-squares minimization, whose normal equations are exactly the lattice form of Eq. (17), with the vorticity of the local current field taking the role of the source term.

A current-based Chern marker

When the chemical potential μ\mu lies within the bulk gap of an insulator, the Středa relation [10, 14] connects the global orbital magnetization to the Chern number CC,

∂Mz∂μ=eAhC,(23)\frac{\partial M_z}{\partial\mu}=\frac{eA}{h}C, \tag*{(23)}

where AA is the area of the sample and C=(h/e2)σxyC=(h/e^{2})\sigma_{xy}, with jx=σxyEyj_x=\sigma_{xy}E_y, being the Hall conductivity in units of e2/he^{2}/h [50]. Since Mz=∫M d2rM_z=\int M\,d^{2}r [Eq. (22)], a local version of this relation follows from the chemical-potential derivative of the loop current. Writing the single-particle density matrix as

ρ(H)=∫dϵ f(ϵ−μ)δ(ϵ−H),(24)\rho(H)=\int d\epsilon\,f(\epsilon-\mu)\delta(\epsilon-H), \tag*{(24)}

where ff denotes the Fermi–Dirac occupation, every bond current decomposes into individual energy contributions,

Iab=∫dϵ f(ϵ−μ) jab(ϵ)I_{ab} = \int d\epsilon\, f(\epsilon-\mu)\,j_{ab}(\epsilon), where the spectral bond current is given by

jab(ϵ)=2eℏIm⁡[tbaδ(ϵ−H)ab].(25)j_{ab}(\epsilon) = \frac{2e}{\hbar}\operatorname{Im}\left[t_{ba}\delta(\epsilon-H)_{ab}\right]. \tag*{(25)}

By linearity, the spectral currents obey the same continuity equation as IabI_{ab}, and Eq. (20) applied to them defines the spectral loop current m(r,ϵ)m(\mathbf{r},\epsilon) satisfying

M(r)=∫dϵ f(ϵ−μ) m(r,ϵ)(26)M(\mathbf{r}) = \int d\epsilon\, f(\epsilon-\mu)\,m(\mathbf{r},\epsilon) \tag*{(26)}

and, by Eq. (20), we have that

m(r,ϵ)=∑y′≤yjx(x,y′;ϵ).(27)m(\mathbf{r},\epsilon) = \sum_{y'\le y} j_x(x,y';\epsilon). \tag*{(27)}

At zero temperature ∂ρ/∂μ=δ(μ−H)\partial\rho/\partial\mu= \delta(\mu-H), hence

∂M(r)∂μ=m(r,μ),(28)\frac{\partial M(\mathbf{r})}{\partial\mu} = m(\mathbf{r},\mu), \tag*{(28)}

which implies that the derivative of the local magnetization is the spectral loop current at the Fermi level. Integrated over the sample it yields ∂Mz/∂μ\partial M_z/\partial\mu. Consequently, Eq. (28) is the local form of Eq. (23). We therefore define a local Chern marker

Cη(r,μ)≡he∫dϵ δη(μ−ϵ) m(r,ϵ),(29)C_\eta(\mathbf{r},\mu) \equiv\frac{h}{e}\int d\epsilon\,\delta_\eta(\mu-\epsilon)\,m(\mathbf{r},\epsilon), \tag*{(29)}

from the spectral loop current resolved within an energy resolution η\eta around the Fermi level, in units of e/he/h. Setting δη=−f′\delta_\eta=-f' yields (h/e) ∂M/∂μ(h/e)\,\partial M/\partial\mu at the inverse temperature β\beta. In a finite sample the spectrum is discrete, and inside a gap the only states near μ\mu belong to the chiral edge channels, with level spacing δϵ≃2πℏv/ℓ\delta\epsilon\simeq2\pi\hbar v/\ell for a boundary of length ℓ\ell and edge velocity vv. The marker is therefore defined in the regime

δϵ≲η≪Δ,(30)\delta\epsilon\lesssim\eta\ll\Delta, \tag*{(30)}

where the window contains many edge levels, so that the edge spectrum acts as a continuum, while excluding the bulk bands across the gap Δ\Delta. This is the same order of limits, ℓ→∞\ell\to\infty before η→0\eta\to0, under which the Středa relation (23) holds. In the opposite limit, η≪δϵ\eta\ll\delta\epsilon with μ\mu between two levels, δη(μ−H)\delta_\eta(\mu-H) vanishes and so does CηC_\eta, everywhere in the sample. Using Eq. (16), we find that

jη=(e/h) ∇×(Cη ez).(31)\mathbf{j}_\eta= (e/h)\,\nabla\times(C_\eta\,\mathbf{e}_z). \tag*{(31)}

where jη=∫dϵ δη(μ−ϵ)j(ϵ)\mathbf{j}_\eta= \int d\epsilon\,\delta_\eta(\mu-\epsilon)\mathbf{j}(\epsilon) is the spectral current resolved within η\eta, the field plotted in Figure 3. Therefore, a step ΔC\Delta C carries the net spectral current ΔC e/h\Delta C\,e/h along it, and the spectral current circulates counter-clockwise around maxima of CηC_\eta and clockwise around minima, so the sense of rotation is given by the sign of the step for an edge channel, and by the type of extremum for a vortex. Built from bond currents alone, CηC_\eta involves no position operator and is independent of the origin for any η\eta and temperature. Projector markers share this property only when PP is an exact projector. At finite temperature, P2≠PP^2 \ne P at metallic edges and walls, and a shift of origin changes the values acquired by the marker.

Figure 3. Current marker and spectral currents in an open 512×512512 \times512 square lattice in a perpendicular magnetic field, with flux Φ=1/100\Phi= 1/100 per plaquette and Anderson disorder W=0.25 tW = 0.25\,t, for a single disorder realization and three chemical potentials: (a) μ=−3.88 t\mu= -3.88\,t, in the gap above the lowest Landau level, E0<μ<E1E_0 < \mu< E_1. In (b) μ=−3.82 t≃E1\mu= -3.82\,t \simeq E_1, at the center of the second Landau level, while for (c) μ=−3.75 t\mu= -3.75\,t, in the gap above it, E1<μ<E2E_1 < \mu< E_2. Each panel shows the same 96×9696 \times96 window at the bottom-left corner of the lattice. Background: Cη(r)C_\eta(\mathbf{r}), lighter for larger ∣Cη∣\lvert C_\eta\rvert. In (a) and (c) it is shown site by site on a shared color scale from 0 to 2.4. In (b) it is averaged over the same 4×44 \times4 blocks as the currents, on a scale running from the 1st percentile of the window to 0. Values outside a scale take its end colors. Lines: streamlines of the spectral current, averaged over 4×44 \times4 blocks, with color and width set by ∣j∣\lvert\mathbf{j}\rvert relative to its maximum in each panel (currents below 1% of the maximum are not shown). On the plateaus the current flows only along the edges, counter-clockwise around the region of positive marker, and the bulk is free of current. At the center of the level the current penetrates the whole window: the bright spots of (b), local minima of CηC_\eta, are encircled by clockwise whirlpools, and the darker regions between them, local maxima, by counter-clockwise ones.

Current vortices in a quantum Hall lattice

The integer quantum Hall effect has been the paramount example of a topological system. In fact, it is where equilibrium currents, magnetization, and topology were first linked [10, 14, 51] and, recently, where equilibrium currents have been imaged [1]. While the construction detailed in Sec. III applies directly to the tight-binding square lattice, this system tests the marker in two regimes. On the one hand, in a mobility gap, it should reproduce the Chern number corresponding to each topological gap. On the other hand, inside a disorder-broadened Landau level, the spectral current invades the system’s bulk and the marker is no longer pinned to an integer value. We take the Hofstadter model [52], a square lattice with hopping −t-t and magnetic flux Φ=1/100\Phi= 1/100 per plaquette (in units of the magnetic flux quantum Φ0\Phi_0) in the Landau gauge, on a 512×512512 \times512 flake (2.6×1052.6 \times10^5 sites) with Anderson disorder. Each uncorrelated on-site energy is drawn from a uniform probability distribution defined on [−W/2,W/2][-W/2, W/2], with W=0.25tW = 0.25t. The three lowest Landau levels of the clean lattice peak at E0≃−3.94tE_0 \simeq-3.94t, E1≃−3.81tE_1 \simeq-3.81t, and E2≃−3.69tE_2 \simeq-3.69t, while the cyclotron gap is ℏωc≃0.12t\hbar\omega_c \simeq0.12t. We consider three distinct chemical potentials: μa=−3.88t\mu_a = -3.88t and μc=−3.75t\mu_c = -3.75t, in the mobility gaps below and above the second level, and μb=−3.82t\mu_b = -3.82t which lies within the center of the second Landau level. For the profiles, we also consider μ=−3.635t\mu= -3.635t, situated in the gap above the third level. We take η=5×10−3t\eta= 5 \times10^{-3}t, larger than the spacing of the edge levels, and R=2048R = 2048 random vectors per μ\mu and disorder configuration. The maps of Figure 3 are for a single disorder realization, while the profiles of Figure 2 are averaged over 16 independent configurations.

Figure 2. Profile of the current marker across an open 512×512512\times512 square lattice in the presence of a perpendicular-to-plane magnetic field, with flux Φ=1/100\Phi=1/100 per plaquette, and Anderson disorder (on-site energies drawn uniformly from [−W/2,W/2][-W/2,W/2], with W=0.25tW=0.25t). The chemical potentials μ=−3.88t\mu=-3.88t, −3.75t-3.75t and −3.635t-3.635t lie in the gaps above the first three Landau levels, ν=1,2,3\nu=1,2,3 (μa\mu_a and μc\mu_c of Figure 3). Solid lines: ⟨C⟩imp\langle C\rangle_{\mathrm{imp}}, the marker CηC_{\eta} (η=5×10−3t\eta=5\times10^{-3}t) averaged over 16 disorder configurations, along the column x=L/2x=L/2, as a function of yy. Horizontal dashed lines: C=1,2,3C=1,2,3. In the bulk the marker is flat at the filling ν\nu, to within 1%, while at the edges it falls to zero over a few magnetic lengths, ℓB=a/2πΦ≈4a\ell_B=a/\sqrt{2\pi\Phi}\approx4a.

In the mobility gap above the lowest level the local marker is uniform in the bulk, rendering the Chern number C=1C = 1 of the plateau, and above the second and third levels it is C=2C = 2 and C=3C = 3, respectively [Fig. 2]. This is the bulk value expected for the projector marker [15, 16]

CP(ra)=4πAsIm⁡⟨a∣PxQyP∣a⟩,(32)C_P(\mathbf{r}_a) = \frac{4\pi}{A_s}\operatorname{Im}\langle a|P x Q y P|a\rangle, \tag*{(32)}

with PP the ground-state projector, Q=1−PQ = 1 - P its complement and AsA_s the area per site. That CηC_\eta reproduces this value follows from the edge currents. An in-gap window carries spectral current only along the boundary, where each chiral channel carries the jump of the Chern number across the gap [53, 54, 55]. Along the cut these jumps add up from 0 outside the sample to CC, with corrections exponentially small in the distance to the nearest boundary

and of order e−Δ2/8η2e^{-\Delta^2/8\eta^2} for a gap Δ\Delta [55, 56]. The window must, however, resolve the discrete edge spectrum, which has a level spacing of δε≃2πℏv/ℓ\delta\varepsilon\simeq2\pi\hbar v/\ell for a boundary of length ℓ\ell. By Poisson summation the relative error is 2e−2π2η2/δε22e^{-2\pi^2\eta^2/\delta\varepsilon^2}, meaning the marker requires δε≃η≪Δ\delta\varepsilon\simeq\eta\ll\Delta, a condition our choice of η\eta satisfies.

Towards the edges, however, the local markers CηC_\eta and CPC_P exhibit distinct behaviors. CηC_\eta falls to zero over a few magnetic lengths [Fig. 2], corresponding to the spatial region where the spectral current flows [Fig. 3(a) and 3(c)]. It runs down the left edge and to the right along the bottom edge, counter-clockwise around the sample, as dictated by the positive sign of the marker (Sec. III A), and carries the net spectral current Ce/hCe/h, which is twice as large in (c) as in (a). The projector marker CPC_P must instead overshoot at the edge since, for an exact projector, its sum over the sample vanishes,

∑aAsCP(ra)=0,(33)\sum_a A_s C_P(\mathbf{r}_a) = 0, \tag*{(33)}

which forces its bulk value to be compensated at the boundaries. Conversely, for the current-based local Chern marker CηC_\eta, the sample average converges to the Chern number with finite-size corrections of order the perimeter times the channel depth divided by the area.

The projector marker is a local quantity when the ground-state projector PP decays exponentially in real space. This occurs whenever μ\mu lies in a gap or a mobility gap [56, 57]. At the center of a Landau level, the localization length diverges [58, 59] and PP decays as a power law. CP(ra)C_P(\mathbf{r}_a) then collects contributions from distant regions of the sample, and its fluctuations and correlation length grow as the transition is approached [19]. The current marker has no such requirement. Its input is the local current, a nearest-neighbor matrix element of the smooth window δη(μ−H)\delta_\eta(\mu-H) defined at every energy, so CηC_\eta remains a local quantity throughout the band. Its value, however, is no longer pinned to an integer, and its sum over the whole sample equals (h/e) ∂Mz/∂μ(h/e)\,\partial M_z/\partial\mu, carried by the states within η\eta of μ\mu. Within the window of Figure 3(b) it varies on the scale ℓB\ell_B, dominated by compact local minima (bright spots). Since CηC_\eta is the stream function of the spectral current, its local extrema determine the direction of circulation. The strong whirlpools around the minima rotate clockwise, while weaker whirlpools around local maxima rotate counter-clockwise, and between them the current follows open, meandering channels.

Generalization to arbitrary two-dimensional crystals

On the square lattice with nearest-neighbor hopping, every bond separates two plaquettes, and Eq. (18) assigns a loop current to each of them. For a general model, such as a lattice with a multi-site, multi-orbital basis and arbitrary-range hoppings, this approach is no longer convenient. Bonds drawn as straight segments intersect one another and divide the plane into plaquettes of diverse shapes. We therefore do not resolve the loop currents on these plaquettes. Instead, we evaluate them on a regular grid of points defined by the primitive vectors a1\mathbf{a}_1, a2\mathbf{a}_2 of the underlying Bravais lattice, thereby extending the construction of Sec. III to any two-dimensional crystal.

For integers b1,b2≥1b_1,b_2\geq1, the coarse cells are the parallelograms spanned by b1a1b_1\mathbf{a}_1 and b2a2b_2\mathbf{a}_2, with corners

cN1N2=c0+b1N1a1+b2N2a2,(34)\mathbf{c}_{N_1N_2}=\mathbf{c}_0+b_1N_1\mathbf{a}_1+b_2N_2\mathbf{a}_2, \tag*{(34)}

so that b1=b2=1b_1=b_2=1 recovers the unit cells. The offset c0\mathbf{c}_0 is chosen so that no site lies on a cell edge. For instance, on the honeycomb lattice of the Haldane model, the corners can be chosen as the hexagon centers (see Figure 4). Each cell owns the two walls w1w_1 and w2w_2 that extend from its corner cN1N2\mathbf{c}_{N_1N_2} along b1a1b_1\mathbf{a}_1 and b2a2b_2\mathbf{a}_2. The net current through a wall,

Figure 4. Coarse cells of size bi=1b_i = 1 on the Haldane lattice. Parallelograms are spanned by b1a1\mathbf{b}_1\mathbf{a}_1 and b2a2\mathbf{b}_2\mathbf{a}_2 (dashed), with corners (squares) on hexagon centers. The colored bonds cross the bottom wall w1w_1: one nearest-neighbor bond and four next-nearest-neighbor chords.

Ik(N1,N2)=∑⟨a,b⟩σab(wk)Iab,(35)I_k(N_1,N_2)=\sum_{\langle a,b\rangle}\sigma_{ab}(w_k)I_{ab}, \tag*{(35)}

collects only the bonds that cross it, where σab(wk)=±1\sigma_{ab}(w_k)=\pm1 is positive when the bond a→ba\to b crosses wkw_k from left to right, as seen walking along wkw_k away from its corner, and negative otherwise. For instance, in the Haldane model with bi=1b_i=1 (discussed below), each wall is crossed by one nearest-neighbor bond and four next-nearest-neighbor chords (Figure 4).

Analogous to the square lattice construction, the loop current is constant between bonds and jumps by IabI_{ab} across the bond a→ba\to b. Following a wall from one corner to the next therefore accumulates the jumps of all the bonds that cross it, and the corner loop currents MN1N2≡M(cN1N2)M_{N_1N_2}\equiv M(\mathbf{c}_{N_1N_2}) obey

MN1+1,N2−MN1N2=I1(N1,N2),MN1,N2+1−MN1N2=I2(N1,N2),(36)\begin{aligned} M_{N_1+1,N_2}-M_{N_1N_2}&=I_1(N_1,N_2),\\ M_{N_1,N_2+1}-M_{N_1N_2}&=I_2(N_1,N_2), \tag*{(36)} \end{aligned}

with M=0M=0 on the corners outside the sample. Going counter-clockwise around a cell, the wall currents add up to the net current leaving the sites it encloses, which vanishes by Eq. (19). (36) is therefore consistent, and MN1N2M_{N_1N_2} does not depend on the path along which the walls are summed. On the square lattice with b1=b2=1b_1=b_2=1 and c0\mathbf{c}_0 at a plaquette center, each wall is crossed by a single bond, and Eq. (36) reduces to Eq. (18).

As on the square lattice, stochastic currents are conserved only on average, and we determine the loop current field using least-squares minimization. Accordingly, when applied to the spectral currents jab(ε)j_{ab}(\varepsilon), the same construction gives the spectral loop currents mN1N2(ε)m_{N_1N_2}(\varepsilon). Through Eq. (29), these quantities define the local Chern marker Cη(cN1N2,μ)C_\eta(\mathbf{c}_{N_1N_2},\mu) at the corners of the coarse cells.

Chern domain in the Haldane model

We demonstrate the effectiveness of the construction of Sec. V on the Haldane model [47], featuring a topological domain at the center of a trivial insulator. Unlike the quantum Hall lattice of Sec. IV, this is a Chern insulator without a net magnetic field, whose crossing next-nearest-neighbor hoppings require the coarse cells. A Chern domain is also the simplest setting in which the current and projector markers must differ inside the sample, at the domain wall, and not only at its outer edge. Such domains occur in magnetic topological insulators [60], twisted graphene multilayers [7, 8, 61], and ultracold-atom realizations of the Haldane model [62].

The Hamiltonian is

H=−t1∑⟨a,b⟩ca†cb+t2∑⟨⟨a,b⟩⟩eiνabϕca†cb+∑aϵaca†ca(37)H = -t_{1} \sum_{\langle a,b\rangle} c_{a}^{\dagger}c_{b} + t_{2} \sum_{\langle\langle a,b\rangle\rangle} e^{i\nu_{ab}\phi} c_{a}^{\dagger}c_{b} + \sum_{a} \epsilon_{a} c_{a}^{\dagger}c_{a} \tag*{(37)}

where νab=sgn⁡([(rk−rb)×(ra−rk)]z)\nu_{ab} = \operatorname{sgn}\left(\left[(\mathbf{r}_{k}-\mathbf{r}_{b}) \times(\mathbf{r}_{a}-\mathbf{r}_{k})\right]_{z}\right), with kk the common nearest neighbor of aa and bb, i.e. νab=+1\nu_{ab} = +1 when an electron hopping from bb to aa turns counter-clockwise, t2=t1/3t_{2} = t_{1}/3 and ϕ=π/2\phi= \pi/2. The on-site energy ϵa=ϵs(ra)sa+Va\epsilon_{a} = \epsilon_{s}(\mathbf{r}_{a})s_{a} + V_{a} combines a staggered potential, with sa=+1s_{a} = +1 (−1-1) on sublattice A (B), and Anderson disorder VaV_{a} drawn uniformly in [−W/2,W/2][-W/2, W/2] with W=0.5t1W = 0.5t_{1}. The pristine model is a Chern insulator for ∣ϵs∣<3t1|\epsilon_{s}| < \sqrt{3}t_{1}. The sample is an L×LL \times L flake with open boundaries, consisting of L=1024L = 1024 unit cells per side (2.1×1062.1 \times10^{6} orbitals). A disk of radius r0=384ar_{0} = 384a, with a=∣a1∣a = |\mathbf{a}_{1}| the lattice constant, is defined by setting ϵs=0\epsilon_{s} = 0 for sites at distance r<r0r < r_{0} from the center of the flake (yielding a Chern insulator with band edges at ±t1\pm t_{1}), with a sharp step to the topologically trivial background, ϵs=3t1\epsilon_{s} = 3t_{1}.

The two markers are expected to behave differently at the wall. The projector marker sums to zero over the sample [Eq. (33)], requiring it to compensate for the bulk contribution of the domain within an annulus along the wall whose width ww is set by the decay length of PP. In this region, it reaches values of order −Cr0/2w-Cr_{0}/2w, which grow with the size of the domain; for w≈1.2aw \approx1.2a this gives ≈−160\approx-160, the value found in Figure 5. The current marker is subject to no such constraint. It interpolates between the bulk values, and its step carries the current Ce/hCe/h along the wall, counter-clockwise around the domain for C>0C > 0.

Figure 5. Local Chern markers for a Chern domain (ϵS=0\epsilon_S=0 for r<r0=384ar<r_0=384a) embedded in a trivial Haldane insulator (ϵS=3t1\epsilon_S=3t_1) on a 1024×10241024\times1024 flake with disorder W=0.5t1W=0.5t_1, at μ=0\mu=0. (a) Current marker CηC_\eta on the whole flake, from one set of R=4096R=4096 random vectors; the dashed line indicates the cut shown in panel (b). (b) Cut along the row n2=L/2n_2=L/2 from the center of the domain to the edge; shaded region: the domain r<r0r<r_0. Solid line: current marker at hexagon centers. Circles: projector marker, evaluated deterministically cell by cell. Inset: details of the wall region, r0±12ar_0\pm12a, on a symmetric-log scale. The projector marker overshoots to about −160-160 just inside the wall, while the current marker passes between the bulk values over a few cells.

Both markers are computed using Chebyshev expansions in KITE [40, 44, 45], for the same domain geometry. The current marker is mapped across the whole flake from a single set of 4096 random vectors and η=1×10−2t1\eta= 1 \times10^{-2}t_{1}. The projector marker, although deterministic, requires two expansions per orbital, one for P∣a⟩P|a\rangle and another for QyP∣a⟩QyP|a\rangle, where PP is a Fermi function at β=2048t1−1\beta= 2048t_{1}^{-1} and positions are measured from the center of the domain. Since a map of the whole flake would require 4L24L^{2} such expansions, we evaluate the projector marker on 23 cells of the row n2=L/2n_{2} = L/2, from the center of the disk to the edge, with spacing decreasing from r0/4r_{0}/4 in the bulk to one cell across the wall. Figure 5 shows this comparison. Deep inside both bulk regions the two markers agree with the bulk Chern numbers, C=1C = 1 and C=0C = 0. At the wall, however, the projector marker overshoots. The current marker instead interpolates smoothly between the bulk values over a few cells, corresponding to the width of the wall mode. Its statistical error decreases as R−1/2R^{-1/2}, and η\eta lies in the regime δϵ≲η≪Δ\delta\epsilon\lesssim\eta\ll\Delta of Eq. (30).

Conclusions

Our work establishes two main results. First, applying pair rotations after the spectral filter yields all bond currents from a single set of random vectors. This eliminates the 2Nτ2N\sqrt{\tau} expansions per random vector required in Ref. [5], thereby reducing the variance of each current by the same factor at equivalent computational cost. Second, current conservation maps these currents into loop currents, defined at the corners of coarse cells in arbitrary two-dimensional crystals, which, for the Fermi sea, correspond to a local orbital magnetization. Their derivative with respect to the chemical potential defines a local Chern marker that requires no position operator and no compensation at boundaries. In a disordered quantum Hall lattice it recovers the Chern number of each plateau, while at the center of a Landau level its contours trace current vortices whose sense of rotation is set by whether they encircle maxima or minima of the marker. At a Chern domain wall in the Haldane model, it transitions between the bulk values over the width of the wall mode, where the projector marker overshoots to values that are expected to grow linearly with the radius of the domain.

In incompressible regions, the marker, proportional to ∂M/∂μ\partial M/\partial\mu, is the quantity imaged by gate-modulated scanning magnetometry [7, 8], whereas in compressible regions the measured signal corresponds to ∂M/∂n\partial M/\partial n. The coarse cells can be enlarged (bi≫1b_{i} \gg1) to the resolution set by the tip height, and convolving MM with the point-spread function of the magnetometer turns the maps of Secs. IV and VI into predictions for such measurements. Two natural extensions are the subject of ongoing work. At the quantum Hall transition the loop currents form a random surface whose contours percolate, and it remains to be seen whether the spectral currents share the multifractality of the wave functions [59, 63]. Furthermore, this estimator also applies to Bogoliubov–de Gennes Hamiltonians, where the same stochastic vectors give the pairing amplitudes and supercurrents.

Acknowledgments

H.P.V. thanks J. M. Alendouro Pinho for fruitful discussions. H.P.V. and J.M.V.P.L. acknowledge support from Fundação para a Ciência e a Tecnologia (FCT, Portugal) through the Strategic Funding UID/04650/2025Centro de Física das Universidades do Minho e do Porto. H.P.V. further acknowledges support from FCT through Project No. EXPL/FIS-MAC/0953/2021 and Grant No. 2024.00560.BD. C.H.L. acknowledges support from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) within Project-ID 314695032 – SFB 1277, and from the Brazilian funding agencies FAPERJ and CNPq. H.P.V. acknowledges computational resources from the Laboratory for Advanced Computing at the University of Coimbra. A.F., J.M.V.P.L. and H.P.V. acknowledge computational resources from the University of York High-Performance Computing service, Viking.

References

References

  1. [1]A. Uri, Y. Kim, K. Bagani, C. K. Lewandowski, S. Grover, N. Auerbach, E. O. Lachman, Y. Myasoedov, T. Taniguchi, K. Watanabe, J. Smet, and E. Zeldov, Nanoscale imaging of equilibrium quantum Hall edge currents and of the magnetic monopole response in graphene, Nat. Phys. 16, 164 (2020).
  2. [2]A. Aharon-Steinberg, T. Völkl, A. Kaplan, A. K. Pariari, I. Roy, T. Holder, V. Wolf, A. Y. Meltzer, Y. Myasoedov, M. E. Huber, B. Yan, G. Falkovich, L. S. Levitov, M. Hücker, and E. Zeldov, Direct observation of vortices in an electron fluid, Nature 607, 74 (2022).DOI
  3. [3]M. L. Palm, C. Ding, W. S. Huxter, T. Taniguchi, K. Watanabe, and C. L. Degen, Observation of current whirlpools in graphene at room temperature, Science 384, 464 (2024).DOI
  4. [4]Y.-Y. Zhang, J. Hu, B. A. Bernevig, X. R. Wang, X. C. Xie, and W. M. Liu, Localization and the Kosterlitz-Thouless transition in disordered graphene, Phys. Rev. Lett. 102, 106401 (2009).pubmed.ncbi.nlm.nih.gov/19392133
  5. [5]H. P. Veiga, D. R. Pinheiro, J. P. Santos Pires, and J. M. Viana Parente Lopes, Markov inequality as a tool for linear-scaling estimation of local observables, Phys. Rev. Research 8, 033015 (2026).DOI
  6. [6]K. C. Nowack, E. M. Spanton, M. Baenninger, M. König, J. R. Kirtley, B. Kalisky, C. Ames, P. Leubner, C. Brüne, H. Buhmann, L. W. Molenkamp, D. Goldhaber-Gordon, and K. A. Moler, Imaging currents in HgTe quantum wells in the quantum spin Hall regime, Nat. Mater. 12, 787 (2013).DOI
  7. [7]C. L. Tschirhart, M. Serlin, H. Polshyn, A. Shragai, Z. Xia, J. Zhu, Y. Zhang, K. Watanabe, T. Taniguchi, M. E. Huber, and A. F. Young, Imaging orbital ferromagnetism in a moiré Chern insulator, Science 372, 1323 (2021).
  8. [8]S. Grover, M. Bocarsly, A. Uri, P. Stepanov, G. Di Battista, I. Roy, J. Xiao, A. Y. Meltzer, Y. Myasoedov, K. Pareek, K. Watanabe, T. Taniguchi, B. Yan, A. Stern, E. Berg, D. K. Efetov, and E. Zeldov, Chern mosaic and Berry-curvature magnetism in magic-angle graphene, Nat. Phys. 18, 885 (2022).DOI
  9. [9]V. Morrisseau, L. C. Rhodes, P. Wahl, and C. A. Marques, How to measure loop currents in scanning tunneling microscopy (2026), arXiv:2607.20030 [cond-mat.str-el].arxiv.org/abs/2607.20030
  10. [10]A. H. MacDonald and P. Středa, Quantized Hall effect and edge currents, Phys. Rev. B 29, 1616 (1984).DOI
  11. [11]T. Thonhauser, D. Ceresoli, D. Vanderbilt, and R. Resta, Orbital magnetization in periodic insulators, Phys. Rev. Lett. 95, 137205 (2005).
  12. [12]D. Ceresoli, T. Thonhauser, D. Vanderbilt, and R. Resta, Orbital magnetization in crystalline solids: Multi-band insulators, Chern insulators, and metals, Phys. Rev. B 74, 024408 (2006).
  13. [13]L. R. F. Lima and C. Lewenkopf, Local equilibrium charge and spin currents in two-dimensional topological systems, Phys. Rev. B 105, 085420 (2022).DOI
  14. [14]P. Středa, Theory of quantised Hall conductivity in two dimensions, J. Phys. C: Solid State Phys. 15, L717 (1982).
  15. [15]A. Kitaev, Anyons in an exactly solved model and beyond, Ann. Phys. 321, 2 (2006).DOI
  16. [16]R. Bianco and R. Resta, Mapping topological order in coordinate space, Phys. Rev. B 84, 241106(R) (2011).
  17. [17]D. Varjas, M. Fruchart, A. R. Akhmerov, and P. M. Perez-Piskunow, Computation of topological phase diagram of disordered Pb₁₋ₓSnₓTe using the kernel polynomial method, Phys. Rev. Research 2, 013229 (2020).DOI
  18. [18]T. V. C. Antão, Y. Sun, A. O. Fumega, and J. L. Lado, Tensor network method for real-space topology in quasicrystal Chern mosaics, Phys. Rev. Lett. 136, 156601 (2026).DOI
  19. [19]R. Favata, N. Baù, and A. Marrazzo, Fluctuations and correlations of local topological order parameters in disordered two-dimensional topological insulators, Phys. Rev. Lett. 135, 026603 (2025).DOI
  20. [20]B. D. Assunção, G. J. Ferreira, and C. H. Lewenkopf, Phase transitions and scale invariance in topological Anderson insulators, Phys. Rev. B 109, L201102 (2024).DOI
  21. [21]K. Kovalenka, A. Ranjbar, S. Azadi, R. V. Belosludov, T. D. Kühne, and M. S. Bahramy, Universal multifractality at the topological Anderson insulator transition (2026), arXiv:2601.22065 [cond-mat.mtrl-sci].arxiv.org/abs/2601.22065
  22. [22]P. Ding and R. Queiroz, Vacancy-driven topological transition of a Chern insulator (2026), arXiv:2610.30241 [cond-mat.dis-nn].arxiv.org/abs/2610.30241
  23. [23]P. d’Ornellas, R. Barnett, and D. K. K. Lee, Quantized bulk conductivity as a local Chern marker, Phys. Rev. B 106, 155124 (2022).
  24. [24]A. Markov, D. Golovanova, A. Yavorsky, and A. Rubtsov, Locality of topological dynamics in Chern insulators, SciPost Phys. 17, 152 (2024).arxiv.org/abs/2304.13708
  25. [25]C. S. Setescak, C. Lewenkopf, and M. Ludewig, Coarse geometric approach to topological phases: Invariants from real-space representations, Phys. Rev. B 111, 104207 (2025).DOI
  26. [26]M. Repše, T. Rejec, and J. Mravlje, Local topological markers for Chern insulators in ribbon geometry (2026), arXiv:2604.10190 [cond-mat.mes-hall].
  27. [27]F. Cioni, L. Ciavacchi, N. Africiani, G. Mazza, F. Taddei, A. Yacoby, and M. Polini, Persistent currents, whirlpools, and local Chern markers in twisted TMD Chern insulators (2026), arXiv:2606.09812 [cond-mat.mes-hall].
  28. [28]R. Bianco and R. Resta, Orbital magnetization as a local property, Phys. Rev. Lett. 110, 087202 (2013).DOI
  29. [29]A. Daido, Quantum theory of current-generating local orbital magnetization (2026), arXiv:2606.29330 [cond-mat.mes-hall].arxiv.org/abs/2606.29330
  30. [30]S. Al Saati, K. Le Hur, and F. Piéchon, Theory of local orbital magnetization: local Berry curvature (2025), arXiv:2512.12343 [cond-mat.mes-hall].
  31. [31]F. Nathan, M. S. Rudner, N. H. Lindner, E. Berg, and G. Refael, Quantized magnetization density in periodically driven systems, Phys. Rev. Lett. 119, 168801 (2017).DOI
  32. [32]E. O. Lachman, A. F. Young, A. Richardella, J. Cuppens, H. R. Naren, Y. Anahory, A. Y. Meltzer, A. Kandala, S. Kempinger, Y. Myasoedov, M. E. Huber, N. Samarth, and E. Zeldov, Visualization of superparamagnetic dynamics in magnetic topological insulators, Sci. Adv. 1, e1500740 (2015).DOI
  33. [33]A. Marguerite, J. Birkbeck, A. Aharon-Steinberg, D. Hallal, K. Bagani, I. Marcus, Y. Myasoedov, A. K. Geim, D. J. Perello, and E. Zeldov, Imaging work and dissipation in the quantum Hall state in graphene, Nature 575, 628 (2019).
  34. [34]A. Uri, S. Grover, Y. Cao, J. A. Crosse, K. Bagani, D. Rodan-Legrain, Y. Myasoedov, K. Watanabe, T. Taniguchi, P. Moon, M. Koshino, P. Jarillo-Herrero, and E. Zeldov, Mapping the twist-angle disorder and Landau levels in magic-angle graphene, Nature 581, 47 (2020).
  35. [35]G. M. Ferguson, R. Xiao, A. R. Richardella, D. Low, N. Samarth, and K. C. Nowack, Direct visualization of electronic transport in a quantum anomalous Hall insulator, Nat. Mater. 22, 1100 (2023).arxiv.org/abs/2112.13122
  36. [36]A. Weiße, G. Wellein, A. Alvermann, and H. Fehske, The kernel polynomial method, Rev. Mod. Phys. 78, 275 (2006).
  37. [37]A. Ferreira and E. R. Mucciolo, Critical delocalization of chiral zero energy modes in graphene, Phys. Rev. Lett. 115, 106601 (2015).DOI
  38. [38]J. H. García, L. Covaci, and T. G. Rappoport, Real-space calculation of the conductivity tensor for disordered topological matter, Phys. Rev. Lett. 114, 116602 (2015).pubmed.ncbi.nlm.nih.gov/25839298
  39. [39]Z. Fan, J. H. Garcia, A. W. Cummings, J. E. Barrios-Vargas, M. Panhans, A. Harju, F. Ortmann, and S. Roche, Linear scaling quantum transport methodologies, Phys. Rep. 903, 1 (2021).DOI
  40. [40]S. M. João, M. Andjelković, L. Covaci, T. G. Rappoport, J. M. V. P. Lopes, and A. Ferreira, KITE: high-performance accurate modelling of electronic structure and response functions of large molecules, disordered crystals and heterostructures, R. Soc. Open Sci. 7, 191809 (2020).DOI
  41. [41]S. M. João and J. M. Viana Parente Lopes, Basis-independent spectral methods for non-linear optical response in arbitrary tight-binding models, J. Phys.: Condens. Matter 32, 125901 (2020).DOI
  42. [42]J. P. Santos Pires, B. Amorim, and J. M. Viana Parente Lopes, Landauer transport as a quasisteady state on finite chains under unitary quantum dynamics, Phys. Rev. B 101, 104203 (2020).DOI
  43. [43]H. Veiga, S. M. João, J. M. A. Pinho, J. P. Santos Pires, and J. M. Viana Parente Lopes, Unambiguous simulation of diffusive charge transport in disordered nanoribbons, SciPost Phys. 17, 149 (2024).DOI
  44. [44]J. P. Santos Pires, S. M. João, A. Ferreira, B. Amorim, and J. M. Viana Parente Lopes, Anomalous transport signatures in weyl semimetals with point defects, Phys. Rev. Lett. 129, 196601 (2022).
  45. [45]S. G. de Castro, J. M. V. P. Lopes, A. Ferreira, and D. A. Bahamon, Fast fourier-chebyshev approach to real-space simulations of the kubo formula, Phys. Rev. Lett. 132, 076302 (2024).DOI
  46. [46]K. J. U. Vidarte, H. P. Veiga, J. M. Viana Parente Lopes, R. Cardias, A. Ferreira, T. P. Cysne, and T. G. Rappoport, Real-space spectral approach to orbital magnetization, Phys. Rev. B 113, 224438 (2026).DOI
  47. [47]F. D. M. Haldane, Model for a quantum Hall effect without Landau levels: Condensed-matter realization of the “parity anomaly”, Phys. Rev. Lett. 61, 2015 (1988).pubmed.ncbi.nlm.nih.gov/10038961
  48. [48]M. Walz, J. Wilhelm, and F. Evers, Current patterns and orbital magnetism in mesoscopic dc transport, Phys. Rev. Lett. 113, 136602 (2014).pubmed.ncbi.nlm.nih.gov/25302913
  49. [49]D. A. Bahamon, G. Gómez-Santos, and T. Stauber, Emergent magnetic texture in driven twisted bilayer graphene, Nanoscale 12, 15383 (2020).DOI
  50. [50]This C is minus the Berry-curvature Chern number C_B = (2π)−1 ∫ Ω d²k of [11, 12]. For electrons, e = −|e|, Eq. (23) takes the familiar form ∂M_z/∂μ = |e|/h C_B.
  51. [51]M. R. Geller and G. Vignale, Currents in the compressible and incompressible regions of the two-dimensional electron gas, Phys. Rev. B 50, 11714 (1994).pubmed.ncbi.nlm.nih.gov/9975307
  52. [52]D. R. Hofstadter, Energy levels and wave functions of Bloch electrons in rational and irrational magnetic fields, Phys. Rev. B 14, 2239 (1976).
  53. [53]Y. Hatsugai, Chern number and edge states in the integer quantum Hall effect, Phys. Rev. Lett. 71, 3697 (1993).pubmed.ncbi.nlm.nih.gov/10055049
  54. [54]J. Kellendonk, T. Richter, and H. Schulz-Baldes, Edge current channels and Chern numbers in the integer quantum Hall effect, Rev. Math. Phys. 14, 87 (2002).
  55. [55]A. Elgart, G. M. Graf, and J. H. Schenker, Equality of the bulk and edge Hall conductances in a mobility gap, Commun. Math. Phys. 259, 185 (2005).
  56. [56]E. Prodan and H. Schulz-Baldes, Bulk and Boundary Invariants for Complex Topological Insulators: From K-Theory to Physics, Mathematical Physics Studies (Springer, Cham, 2016).
  57. [57]W. Kohn, Density functional and density matrix method scaling linearly with the number of atoms, Phys. Rev. Lett. 76, 3168 (1996).pubmed.ncbi.nlm.nih.gov/10060892
  58. [58]B. Huckestein, Scaling theory of the integer quantum Hall effect, Rev. Mod. Phys. 67, 357 (1995).
  59. [59]F. Evers and A. D. Mirlin, Anderson transitions, Rev. Mod. Phys. 80, 1355 (2008).
  60. [60]K. Yasuda, M. Mogi, R. Yoshimi, A. Tsukazaki, K. S. Takahashi, M. Kawasaki, F. Kagawa, and Y. Tokura, Quantized chiral edge conduction on domain walls of a magnetic topological insulator, Science 358, 1311 (2017).
  61. [61]H. Polshyn, J. Zhu, M. A. Kumar, Y. Zhang, F. Yang, C. L. Tshchirhart, M. Serlin, K. Watanabe, T. Taniguchi, A. H. MacDonald, and A. F. Young, Electrical switching of magnetic order in an orbital Chern insulator, Nature 588, 66 (2020).
  62. [62]G. Jotzu, M. Messer, R. Desbuquois, M. Lebrat, T. Uehlinger, D. Greif, and T. Esslinger, Experimental realization of the topological Haldane model with ultracold fermions, Nature 515, 237 (2014).DOI
  63. [63]S. Mu, A. A. Saberi, and M. R. Zirnbauer, Multifractality at the integer quantum Hall transition: acquittal of the parabolic law (2026), arXiv:2606.21679 [cond-mat.mes-hall].arxiv.org/abs/2606.21679

Paper details

vlnc:W_8wkcbp9q
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 6

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents