WorksPhysics and Astronomy
Integer and Fractional Chern Plateaus Shaped by Layer-Dependent Disorder
Abstract
The realization of integer and fractional Chern insulators in moiré systems provides a lattice setting for exploring how the structure of Bloch wave functions influences the interplay between topology and disorder. Here we investigate how layer-dependent disorder shapes Hall plateau widths and transitions. Using a realistic three-orbital model of twisted bilayer MoTe, we compare Hall and bulk transport under onsite-potential, displacement-field, and twist-angle disorder. At equal bare disorder strength, onsite-potential disorder produces broad integer plateaus with nearly direct transitions, whereas displacement-field disorder yields narrow plateaus separated by conducting regions with pronounced Hall-resistance dips. These contrasting behaviors arise from the layer composition of the Bloch states: contributions from different layer under displacement-field disorder partially cancel, weakening the bulk localization. Composite-fermion mean-field theory further yields fractional plateaus at and with the same disorder dependence of plateau widths. These findings identify layer-dependent disorder as an important mechanism shaping Chern plateau stability in moiré systems beyond bare disorder strength alone.
Introduction
The discovery of the integer and fractional quantum Hall effects (IQHE and FQHE) in two-dimensional electron gases established a remarkable interplay between topology, electron correlations, and disorder [1–4]. While topology underlies the quantization of Hall conductivity, disorder localizes bulk electronic states and stabilizes quantized Hall plateaus over finite ranges of carrier density [4–7]. The realization of integer and fractional Chern insulators (CIs and FCIs) extends these phenomena to lattice systems without external magnetic fields [8–14]. Unlike conventional Landau levels, Chern bands possess nontrivial lattice and orbital structures, introducing additional degrees of freedom that can substantially influence electronic localization and transport.
Moiré materials provide a highly tunable platform for exploring these effects, with integer and fractional Chern states observed in twisted transition-metal dichalcogenides [15–18] and rhombohedral graphene superlattices [19, 20]. Transport experiments have revealed strikingly different behaviors of Hall plateaus and their transitions [17–28]. In some systems, broad quantized plateaus are connected by narrow, nearly direct transitions [20, 24], resembling conventional quantum Hall behavior shown in Figure 1(a). In others, narrow plateaus are separated by extended conducting regions where the Hall resistance is strongly suppressed, sometimes approaching zero before recovering toward the next quantized value [17–19, 21–23, 25–28], as shown in Figure 1(b). These contrasting behaviors suggest that the microscopic disorder landscape plays an important role beyond conventional plateau broadening. In layered moiré systems, different disorder sources can act differently on the layer and orbital degrees of freedom [29–33], potentially modifying the localization of bulk states. This raises a central question: How does the layer-dependent nature of disorder govern the stability of integer and fractional Chern plateaus and the transport behavior between them?
Figure 1. Hall-plateau transitions and moiré model. (a),(b) Schematic Hall resistance versus hole filling : broad plateaus separated by a narrow, direct transition in (a), and narrow plateaus separated by a broad transition with a Hall dip toward zero in (b). Yellow and blue shading highlights plateau and transition regions, respectively. (c) Moiré lattice of MoTe with MM (black), MX (blue), and XM (red) stacking regions, primitive vectors and , and the rotation axis. Insets illustrate local potential , perpendicular displacement field , and twist-angle disorder. (d) The three topmost -valley valence bands at from the continuum model (red solid), Wannier TB model (black dashed), and TB model with hoppings retained through the sixth neighbor shell (blue dash-dotted). Band Chern numbers are labeled.
In this work, we identify the layer-dependent disorder sensitivity as a common mechanism in shaping the Chern plateau stability. Using a realistic three-orbital Wannier model of twisted MoTe (MoTe) as an example, we investigate
onsite-potential, displacement-field, and twist-angle disorder, as shown in Figure 1(c). Their distinct orbital structures control bulk localization through different effective disorder strengths, leading to markedly different Hall plateau and transition behaviors sketched in Figure 1(a,Figure 1b). We elucidate this mechanism through orbital form factors and disorder-related charge responses, and extend the analysis to fractional Chern plateaus within composite-fermion (CF) mean-field theory [34, 35].
**Model construction—**The moiré superlattice of is triangular, with MM, MX, and XM stacking regions corresponding to Mo–Mo, Mo–Te, and Te–Mo alignments between the top and bottom layers, respectively, as shown in Figure 1(c). The low-energy valence states originate from the monolayer and valleys, where strong spin–orbit coupling produces spin–valley locking. Opposite spin-polarizations in the two valleys suppress intervalley scattering under smooth moiré potentials [36, 37, 38, 39, 40]. For the valley-polarized Chern states considered here, we focus on the single--valley sector and construct a three-band tight-binding (TB) model. To this end, we start from a continuum model extracted directly from first-principles calculations without parameter fitting [41], whose -valley bands at are shown in red solid lines in Figure 1(d). Throughout the twist angle range –, the three highest valence bands have zero total Chern number, although their individual Chern numbers vary with twist angle. For example, at , they carry Chern numbers , , and , respectively, as labeled in Figure 1(d).
The vanishing total Chern number allows us to construct three maximally localized Wannier functions (MLWFs) using WANNIER90 [42]. These Wannier functions, hereafter referred to as local orbitals, are centered at the MX, XM, and MM stacking regions, respectively. Importantly, they exhibit distinct layer polarizations: the MX and XM orbitals are predominantly localized in opposite layers, whereas the MM orbital has approximately equal weights in both layers [37, 43, 44, 45, 46]. This pattern is consistent with symmetry [40], where denotes time reversal and is a 180° rotation about the in-plane axis that exchanges the layers and MX/XM centers while leaving MM fixed, as shown in Figure 1(c). The resulting three-orbital TB model accurately reproduces the continuum band structure, as shown in black dashed lines Figure 1(d). For real-space transport, we truncate hoppings beyond the sixth-neighbor shell, preserving the overall band dispersion and the individual band Chern numbers, as shown in blue dash-dotted lines in Figure 1(d). Further details of the model construction are provided in Supplementary Material I (SM. I).
We next introduce three types of disorder commonly encountered in moiré systems [47, 48] [29, 32, 33]: local onsite potential variations , perpendicular displacement-field variations , and twist-angle variations , defined on each moiré unit cell , as illustrated in Figure 1(c). These disorder channels act differently on the local Wannier orbitals. Potential disorder shifts all three orbital energies equally, whereas displacement-field disorder produces opposite shifts of the layer-polarized MX and XM orbitals, with a negligible contribution to the layer-balanced MM orbital. Twist-angle disorder primarily changes the relative onsite energies of the MX/XM and MM orbitals, reflecting their different responses to local variations of the moiré structure. Although twist-angle disorder also modifies hopping, we find that its onsite contribution dominates the transport behavior, with comparatively small corrections from hopping variations (see SM. II). We therefore retain the onsite contributions of all three disorder channels and obtain the Hamiltonian,
Here is the clean TB Hamiltonian and contains the annihilation operators of the three local Wannier orbitals. Different disorder channels are labeled by the index and are considered separately. The scalar field specifies the disorder amplitude in cell , while the dimensionless diagonal matrix describes its orbital structure, as illustrated in Figure 3(a). Specifically, , , and . The prefactors ensure a common normalization, , allowing the three disorder channels to be compared at equal bare disorder strength. Their effective coupling to the electronic bands depends on the orbital composition of the Bloch states, as discussed below.
Figure 3. Orbital projections, charge response, and bulk transport. (a) Wannier orbital structure for , , and disorder (left to right), with black, blue, and red dots denoting MM, MX, and XM orbitals, respectively. (b) The two top valence bands colored by the coupling magnitude for the corresponding disorder channels. Each charge response amplitudes per unit disorder strength, , versus hole filling for the top () and second () valence bands. Red, blue, and green curves correspond to , , and disorder, respectively. (d) under (filled squares) and (open circles) disorder. The parameters are in (a-d), and , meV in (c,d).
We model as a zero-mean Gaussian disorder field with variance , characterizing the bare disorder strength. We consider both spatially uncorrelated disorder () and smoothly correlated disorder with dimensionless correlation length , measured in units of the moiré lattice constant . The latter captures smooth inhomogeneity on moiré length scales [47, 48] [29, 32, 33]. The disorder field is shared by the three orbitals within each unit cell, with their distinct onsite shifts determined by . Details of the disorder construction are provided in SM. II.
**Integer Chern plateau transitions—**To investigate the effects of different disorder channels on Chern plateau stability and transitions, we calculate the six-terminal Hall resistance using the Landauer–Büttiker formalism [49] implemented in KWANT [50] in the Hall-bar geometry with open boundaries, as shown in Figure 2(a), current flows from lead 1 to lead 2, while leads 3–6 serve as voltage probes carrying zero net current. The Hall resistance is defined as
Figure 2. Hall transport under orbital-selective disorder. (a) Six-terminal geometry with open boundary conditions (OBC). (b)–(d) Disorder-averaged Hall resistance versus hole filling for , , and disorder, respectively. Curve colors indicate the system sizes listed in (b). Horizontal dashed lines mark and and vertical lines mark , respectively. Here and count lattice repeats of length along and along , respectively, where is the moiré lattice constant. Parameters: , , and meV, corresponding to for twist-angle disorder (see SM. II). Throughout this work, each transport data point is averaged over 500 disorder realizations.
where is the voltage at lead .
Figure 2(b–Figure 2d) compare the disorder-averaged Hall resistance under the three disorder channels at , with identical bare disorder strength and correlation length . Throughout this study, denotes the hole filling of the three valence bands shown in Figure 1(d), with corresponding to the chemical potential at the top of the highest valence band. Potential disorder produces broad plateaus at
and over – and –, corresponding to widths and , respectively, as shown in Figure 2(b). These quantized values correspond to cumulative Chern numbers and , consistent with the two highest Chern bands in Figure 1(d). The plateaus are connected by a narrow, nearly direct transition, resembling conventional quantum Hall behavior illustrated in Figure 1(a). In contrast, disorder yields much narrower plateaus at the same quantized values, with widths and , respectively, as shown in Figure 2(c). Between these plateaus, exhibits a pronounced dip toward zero before recovering to the second quantized value, characteristic of the broad transition region illustrated in Figure 1(b). Twist-angle disorder produces intermediate behavior, as shown in Figure 2(d): the first plateau has a width , between those of and , while the second remains narrow (). The intervening Hall-resistance dip is also less pronounced than for disorder. Similar trends in plateau widths and transition profiles are found at , with more detailed results presented in SM. III. As discussed below, these contrasting behaviors originate from the interplay between the Bloch wave function structures of the Chern bands and the layer-dependent nature of disorder.
Layer-dependent disorder and localization
The three disorder channels act differently on the local Wannier orbitals through their onsite matrices , as illustrated in Figure 3(a). Their effects on quasiparticles in the Chern bands are determined by the orbital composition of the Bloch states . At momentum transfer , the disorder potential projected onto the -th band is characterized by the orbital form factor , multiplied by the Fourier component of the disorder field. Figure 3(b) shows the magnitude of the zero-momentum form factor for the two highest valence bands. For potential disorder , equal coupling to all three orbitals gives .
For displacement-field disorder , the opposite contributions from MX and XM can largely cancel, while MM does not contribute. For twist-angle disorder , cancellation instead occurs between the MM and MX/XM contributions. Consequently, the diagonal form factors with are substantially suppressed for and compared with . This illustrates how the layer-dependent orbital structure weakens the effective disorder coupling to the Chern bands.
To further characterize the effects of different disorder channels, we calculate the total charge-density response to an orbital-dependent perturbation. The intraband density response is , where
Here is the Fermi–Dirac distribution, and the form factor represents the perturbation vertex. We characterize the typical charge rearrangement by the root-mean-square response amplitude
where incorporates the disorder correlations controlled by (see SM. II). At equal bare disorder strength, quantifies the relative total-charge response to different disorder channels. Figure 3(c) shows that the response is generally strongest for and weakest for , with intermediate, across the two highest valence bands. This hierarchy is consistent with the projected disorder couplings in Figure 3(b), indicating that displacement-field disorder induces substantially weaker total-charge redistribution than potential disorder. Further details of the charge-response analysis are provided in SM. IV.
To examine the corresponding bulk transport behavior, we calculate the two-terminal conductance under and disorder. We retain leads 1 and 2 of the geometry in Figure 2(a) and impose periodic boundary conditions along the direction to eliminate edge conduction. After correcting for the Sharvin contact resistance [51], the conductance reads
where is the total transmission, is the transmission matrix, and denote the numbers of propagating modes in the two leads. Figure 3(d) shows that the disorder-averaged bulk conductance approaches zero within the Hall plateau regions for both disorder channels, consistent with bulk localization. Between plateaus, however, disorder produces a small, nearly size-independent conductance peak of , whereas the peak under disorder increases from to as increases from 16 to 48.
This demonstrates substantially stronger bulk conduction under disorder.
In the isotropic bulk approximation, the Hall resistance satisfies . On quantized plateaus, and . Between plateaus, stronger localization under disorder suppresses bulk conduction, yielding nearly direct Hall transitions resembling those in conventional quantum Hall systems. By contrast, the larger longitudinal conduction under disorder suppresses relative to , producing the pronounced Hall-resistance dip observed in Figure 2(c). Furthermore, in Figure 2 exhibits asymmetric broadening, particularly under and disorder around , as shown in Figure 2(c,Figure 2d). This asymmetry is qualitatively accounted for by the unequal effective disorder coupling at the band edges on either side of , as reflected in the projected disorder matrix elements and charge-response amplitudes in Figure 3(b,Figure 3c). The stronger projected coupling and larger charge response on the side suggest greater sensitivity to disorder, supporting more effective bulk localization and greater plateau broadening. Conversely, the weaker coupling on the side is consistent with the more rapid loss of Hall quantization. Similar bulk transport trends are found at , with further details provided in SM. III.
Fractional Chern plateau transport—Motivated by the fractional Chern plateaus observed in [15–18, 21], we extend the disorder analysis to , a representative twisting angle in the experimentally relevant regime. We retain the full three-band Wannier Hamiltonian with MX, XM, and MM orbitals, as detailed in SM. I, while focusing on hole doping of the top valence band over . Here corresponds to an empty top electron band. This band carries electronic Chern number , corresponding to in the hole representation.
We adopt a CF mean-field description [34, 35] previously employed in studies of moiré FCIs [52, 53, 54, 55, 56, 57, 58, 59]. We implement flux attachment in the electron representation, assigning two statistical flux quanta to each electron of charge , with . The electron filling counts all three bands, so . In the spatially uniform mean-field approximation, the statistical flux per moiré cell is therefore , as illustrated in Figure 4(a). The CF Hamiltonian expressed in hole representation reads
Figure 4. CF spectrum and fractional plateau transport. (a) Schematic of the CF construction and its uniform mean-field representation, with attached to each electron and a flux uniformly distributed within the moiré cell. (b) Upper panel: Clean CF Hofstadter spectrum (blue) and chemical potential (red) versus hole filling . Lower panel: The corresponding spectral gap at . (c),(d) Disorder-averaged physical Hall resistance of disordered CF model with and disorder, respectively. Colored curves correspond to different system sizes. Black curves show experimental data obtained from Ref. [21]. The calculations in (c,d) use , with meV and .
Here annihilates a CF in the hole representation, is the hole TB matrix, and generates the statistical flux defined above. To obtain the physical transport response, we use the Chern–Simons (CS) relation between the physical and CF Hall resistances in the ideal Hall-bar approximation [52, 53],
Here denote disorder averaging.
We first examine the clean CF spectrum. Varying changes both the hole density and the statistical flux, producing the Hofstadter spectrum shown in Figure 4(b). The chemical potential , determined by the hole density, is shown in red, and exhibits jumps at and associated with pronounced spectral gaps , as depicted in the lower panel. These gaps identify incompressible states within the CF description. The occupied CF bands carry total Chern numbers and , respectively, which yield fractionally quantized Hall resistances and through Eq. (7). A gap also appears at with , corresponding to the integer Chern plateau with .
We next include and disorder with the same bare strength meV and correlation length . Using the six-terminal geometry in Figure 2(a), we calculate and convert it to the physical Hall resistance using Eq. (7). As shown in Figure 4(c,Figure 4d), both disorder channels produce fractional plateaus near and , together with an integer plateau near , in agreement with the experimental plateaus [21] shown in black. The fractional plateaus are broader under than under , following the same disorder dependence as the integer plateaus in Figure 2(b,Figure 2c). This trend is consistent with less effective localization of CF bulk states under at equal disorder strength. The narrower experimental fractional plateaus more closely resemble the results.
On the other hand, we note that the calculated transport behavior between fractional plateaus deviates from experiment, highlighting the limitations of the CF mean-field treatment in describing the plateau transitions. Under disorder, the weaker effective disorder coupling allows stronger CF bulk conduction, suppressing and leaving the CS contribution in Eq. (7) dominant. Consequently, approaches , as shown in Figure 4(d), in contrast to the experimentally observed Hall-resistance dips shown by the black solid curve.
**Summary—**Our results show that Chern plateau stability depends not only on the bare disorder strength but also on how disorder acts on the layer components of Bloch states. Taking as an example, we construct three-orbital Wannier-based TB models at different twist angles and compare Hall and bulk transport under onsite-potential, displacement-field, and twist-angle disorder. At equal bare disorder strength, their layer-dependent effects on localization yield contrasting integer plateau profiles: broad plateaus with nearly direct transitions under , versus narrow plateaus separated by Hall dips under , qualitatively consistent with experiment. CF mean-field theory also reproduces the 2/3 and 3/5 plateaus and captures their disorder-dependent widths. These findings suggest a common mechanism, rooted in layer-dependent disorder, for plateau stability in Chern systems and motivate further studies in other moiré platforms, including rhombohedral graphene/hBN superlattices [19, 20].
**Acknowledgments—**This work was supported by the National Natural Science Foundation of China (Nos. 12494594 and 12574150), the Science Center of the NSFC (Grant Nos. 12188101 and 12621008), the National Key R&D Program of China (Grant No. 2022YFA1403800), and the Chinese Academy of Sciences Project for Young Scientists in Basic Research (Grant No. 2022YSBR-048). H.W. also acknowledges support from the New Cornerstone Science Foundation through the XPLORER PRIZE.
Moiré model construction
This section describes the construction of the three-orbital tight-binding (TB) model of used in our transport calculations. We first introduce the single--valley continuum model extracted directly from first-principles calculations without parameter fitting [41]. We then construct maximally localized Wannier functions (MLWFs) for its three highest valence bands using WANNIER90 package [42], whose Chern numbers sum to zero, and obtain the corresponding TB Hamiltonian.
Accurate continuum model without fitting
Unlike earlier continuum models fitted to density-functional-theory (DFT) bands [36, 37, 38, 39, 40], we adopt a model extracted directly from first-principles Hamiltonians without fitting, which accurately reproduces the low-energy DFT bands and wave-functions [41]. The underlying DFT calculations use OpenMX software [60, 61, 62] with the standard pseudo-atomic-orbital basis - and -, with a radial cutoff of 7 Bohr. We use the resulting models at six twist angles, , , , , , and .
The extraction begins by projecting the DFT Hamiltonian onto atomic plane waves near the valley using the truncated atomic plane wave (TAPW) method [63]. Integrating out remote states then leaves one spin-polarized valence state per layer and plane wave. The resulting low-energy Hamiltonian is expanded in symmetry-allowed continuum terms, whose coefficients are determined by projection [41]. Grouping the associated annihilation operators by layer, we write
Here counts the retained plane waves in layer . The index runs over both layers in the order shown in , with each entry specified by its wave vector and layer . The basis state is a plane wave with envelope momentum in layer . The corresponding component of annihilates a particle in this state. The wave vectors satisfy , where is a moiré reciprocal vector and locates the layer’s point relative to a common origin; lies in the moiré Brillouin zone. Here is the single-particle -valley Hamiltonian.
The continuum Hamiltonian combines the valence-band dispersion with moiré couplings within and between the layers:
where , , and describe kinetic energy, intralayer moiré coupling, and interlayer tunneling, respectively. These contributions include both higher moiré harmonics and momentum-dependent couplings. To write their general form, label each basis state by and define and . The matrix elements take the form
The three lines give the kinetic, intralayer, and interlayer terms. The quadratic kinetic coefficients satisfy , with the magnitude of the valence-band effective mass. The momentum transfer labels the moiré harmonics, while label the momentum powers. We retain the terms specified in the source model [41, 64], subject to Hermiticity and crystal symmetries. The resulting bands and Bloch states provide the starting point for the Wannier construction below.
Figure Figure S1 shows the -valley continuum bands (red) at the six twist angles from to . The Chern numbers of the three highest valence bands, ordered from top to bottom, are at , , and , and at , , and , as labeled in Figure S1.
Figure S1. Three highest -valley valence bands of tMoTe at (a) , (b) , (c) , (d) , (e) , and (f) . Red solid, black dashed, and blue dash-dotted lines correspond to the continuum model, the Wannier TB model, and the TB model with hoppings retained through the sixth neighbor shell, respectively. The band Chern numbers are indicated in each panel.
Wannierization and tight-binding model
For real-space transport calculations, we need to extract a lattice TB model from the low-energy bands of the continuum model. The low-energy valence states of MoTe₂ originate from the and valleys, where strong spin–orbit coupling locks opposite spins to the two valleys and suppresses intervalley mixing [36, 37, 38, 39, 40]. We therefore work with the single--valley continuum models constructed above at the six twist angles from to . Their three highest valence bands form an isolated group with Chern numbers that sum to zero, allowing us to Wannierize them by using WANNIER90 package [42] to obtain MLWFs and a three-orbital TB model.
Different from the standard DFT–WANNIER90 workflow, we generate the four input files (.win, .eig, .mmn, .amn) directly from the continuum-model data. This requires particular care with the plane-wave basis at the moiré Brillouin-zone boundary and with the choice of initial trial orbitals. This continuum-model-based Wannierization approach is broadly applicable and has also been used to construct effective low-energy TB models for other moiré systems [46, 65, 66]. The workflow of this approach is summarized in Figure S2, with each step described below:
Figure S2. Workflow for constructing a three-orbital TB model and MLWFs from the three highest -valley continuum bands of MoTe₂. The four Wannier90 inputs contain the moiré cell and momentum mesh (.win), band energies (.eig), neighbor overlaps with Brillouin-zone boundary sewing (.mmn), and LDOS-guided initial projections (.amn). Spread minimization yields the real-space TB Hamiltonian (_hr.dat), unitary rotations for MLWF reconstruction from the continuum Bloch states (_u.mat), and Wannier centers and spreads (.wout).
Lattice and band information (.win, .eig). The .win file specifies the moiré cell, a momentum mesh, and num_bands=num_wann=3. The .eig file stores the corresponding continuum eigenvalues in eV, where is the band index. All input files use the same momentum and band ordering. No disentanglement is required for this isolated three-band group.
Neighbor overlaps (.mmn). The .mmn file stores the overlap matrices between the cell-periodic parts of Bloch states at neighboring momenta and . To evaluate these overlaps, we expand the continuum eigenstates in the layer-resolved plane-wave basis defined above:
Here are the plane-wave coefficients, and is the cell-periodic part.
The neighbor pairs and reciprocal translations are specified in the .nnkp file generated by Wannier90 preprocessing. When a neighbor crosses the moiré Brillouin-zone boundary, we write , with on the stored mesh and a moiré reciprocal vector. We introduce the sewing matrix to express the eigenvectors stored at in the plane-wave basis at . It shifts the labels by without mixing layers, giving
Each nonzero element connects states in the same layer with the same plane-wave momentum, since . We use and retain only matches within the plane-wave cutoff.
Initial projections (.amn)
The .amn file stores the initial projection matrices for MLWF construction, defined by the overlaps between continuum Bloch states and localized trial orbitals. Guided by the layer-resolved local density of states (LDOS) calculated from the continuum model, we choose Gaussian trial orbitals centered at MX, XM, and MM, which take the form
Here is the real-space width and is a normalized layer spinor. For moiré primitive vectors , the centers are , , and (see Figure 1). The MX and XM trials occupy opposite layers, while the MM trial has equal layer weights. These layer weights respect the . The operation is a 180° rotation about the in-plane axis that flips the layers and exchanges MX/XM centers while leaving MM fixed [40], and is time reversal symmetry. The layer spinor also specifies the relative phase, which is not determined by the density alone. Fourier transformation gives the trial coefficients in the continuum basis and hence the projection matrix:
Here is a momentum-independent normalization constant. The matrix must have full rank throughout the mesh to initialize all three Wannier orbitals. Their final layer weights are determined by the subsequent localization.
With these inputs, Wannier90 optimizes a unitary matrix to minimize the total quadratic spread . The resulting Wannier functions are
where is the number of momentum points and is a moiré lattice vector. The expectation values in refer to the home-cell states . The matrices are saved in .mat, while .wout records the Wannier centers, spreads, and minimization history.
The Wannier rotation and Fourier transformation give the clean TB Hamiltonian in Eq. (1) of the main text:
Here and label moiré unit cells, and is the hopping matrix in the (MX, XM, MM) orbital basis, with as in the main text. The band energies form the diagonal matrix . The matrix elements are stored in hr.dat. For the finite momentum mesh, the standard Wannier90 Wigner–Seitz weights and equivalent-cell images are retained in the real-space interpolation. The Wannier TB bands (black dashed lines) accurately reproduce the continuum bands (red solid lines) at all six twist angles [Figure S1]. For real-space transport calculations, we truncate hoppings beyond the sixth neighbor shell. The truncated bands (blue dash-dotted lines) remain in good agreement with the continuum dispersions and preserve the Chern number of each band.
To visualize each Wannier orbital in real space, we Fourier transform the rotated plane-wave coefficients to obtain the layer-resolved Wannier function distribution. For the home cell , the Wannier function density in layer are
Here is the moiré-cell area and is the total number of momentum points. Integrating gives the weight in layer , with the two layer weights summing to unity.
Figures S3 and S4 show these Wannier function densities at and , respectively. Although the individual band Chern numbers differ between the two angles (see Figure S1), the Wannier centers remain at MX, XM, and MM. MX and XM exhibit strong, opposite layer polarizations, with approximately 93–95% of their weight in the bottom and top layers, respectively, whereas MM has nearly balanced layer components. Moreover, the densities exhibit the expected symmetry about each center and the layer-exchange relations of . All three orbitals are well localized, with spatial widths .
In this localized Wannier basis, we examine how the TB Hamiltonian parameters vary with twist angle. We focus on three aspects: onsite energies, hopping amplitudes and hopping phases, as shown in Figure S5. The MX and XM onsite energies, defined as and , coincide within numerical precision, as required by the symmetry that exchanges these two orbitals [40]. Their common offset from MM rises from nearly zero at to meV at , followed by a slight decrease at [Figure S5]. For the hopping terms, we group them into shells according to the distance between the connected Wannier centers: denotes nearest neighbors, next-nearest neighbors, and so on. The root-mean-square hopping amplitude in shell is
Figure S5. Twist-angle dependence of the Wannier TB parameters. (a) Onsite energy offsets and . (b) Root-mean-square hopping amplitudes (see Eq. (18)) for the first six neighbor shells, ordered by the distance between Wannier centers. (c) Phases of intercell NN hoppings between the indicated orbitals. Results are shown for six twist angles from to .
where contains the bonds in that shell, with each Hermitian pair counted once. Figure S5(b) shows that nearest-neighbor (shell 1) hopping terms remain dominant throughout the studied range, increasing from approximately 3.1 to 6.5 meV, while-vesm more distant shells remain much weaker, consistent with the well-localized character of the MLWFs. Since first-shell hopping dominates, we focus on the corresponding nearest-neighbor hopping phases. To compare them across angles, we fix the Wannier gauge by making the two intracell reference hoppings and real and positive. The phases in Figure S5 instead refer to selected intercell nearest-neighbor bonds, , with ; these hoppings can therefore remain complex in the same gauge. Over , the phase pattern changes little: the MM–MX and MM–XM phases stay nearly constant at and , respectively, while the MX–XM phase shows a modest shift from approximately to . The systematic angle dependence of the onsite energy difference between MM and MX/XM and of the nearest-neighbor hopping amplitudes are used to construct the twist-angle disorder Hamiltonian below.
Disorder Hamiltonian construction
In this section, we introduce the construction of the disorder Hamiltonians for three types of disorder discussed in the main text: potential disorder , displacement-field disorder , and twist-angle disorder . We first describe the construction of spatially uncorrelated disorder fields and then introduce how to generate spatially correlated disorder from them.
Spatially uncorrelated disorder
We consider the three disorder channels separately and first construct spatially uncorrelated disorder by assigning an independent random value to each moiré unit cell . For each disorder realization, we first generate , , or and then construct the Hamiltonian as described below, reflecting how local inhomogeneities modify the electronic structure of real moiré systems. To compare their orbital effects, we first retain the onsite terms, as in Eq. (1) of the main text:
where . In the common Gaussian convention, the spatial uncorrelated onsite disorder term are drawn independently for different cells, with the follow distribution:
All three orbitals in a cell share the same random variable, with their orbital components set by ; independence refers to different cells, not to different orbitals within a cell. We impose so that provides a common measure of disorder strength for all three channels. This allows us to compare the effects of orbital selectivity at equal .
A local potential disorder shifts all three onsite energies equally, giving and . A perpendicular displacement field instead produces opposite potential shifts in the two layers. Since MX and XM are predominantly localized in opposite layers (see Figure S3 and Figure S4), their onsite shifts have opposite signs and nearly equal magnitudes, whereas the shift of the layer-balanced MM orbital vanishes. produces equal and opposite onsite shifts of MX and XM, while the small shift of MM is neglected. We therefore use , with setting the energy scale. The factor ensures , so that and measure disorder strength on the same scale.
Figure S3. Layer-resolved Wannier function densities of tMoTe at . (a)–(c) Top-layer densities of the MX, XM, and MM orbitals; (d)–(f) the corresponding bottom-layer densities. Squares, triangles, and circles mark MM, XM, and MX sites, respectively; arrows indicate the moiré lattice vectors and . The two layers of each orbital share a color scale in Å.
Figure S4. Layer-resolved Wannier function densities of tMoTe at . (a)–(c) Top-layer densities of the MX, XM, and MM orbitals; (d)–(f) the corresponding bottom-layer densities. Squares, triangles, and circles mark MM, XM, and MX sites, respectively; arrows indicate the moiré lattice vectors and . The two layers of each orbital share a color scale in Å.
For twist-angle disorder , we use the angle dependence of the Wannier parameters in Figure S5. At each clean angle, symmetry relates MX and XM and enforces equal onsite energies [40]. Their energy relative to MM changes with angle Figure S5. We describe this change using , the negative of the offset plotted in that panel. For a local angle , let . To isolate the layer-dependent onsite contribution of the twist angle disorder, we omit a shift common to all three orbitals, which would enter the scalar-potential channel. The remaining onsite correction is
Thus, the signs in describe relative onsite shifts after separating the common shift. We obtain by piecewise-linear interpolation of the six clean Wannier models, continuing the endpoint slopes outside the tabulated angle range.
The relation to a twist angle disorder strength follows by drawing independently from . Linearizing the onsite offset gives
This connects Gaussian angle fluctuations to the common Gaussian onsite disorder model. For the full interpolated angle dependence, the disorder strength is instead evaluated directly from the onsite changes:
The angle-width label used in the transport figures is . At , , , and correspond to , 1.83, and 3.03 meV, respectively. This mapping relates the twist-angle width to the onsite disorder strength , enabling comparison with the other two channels.
Twist-angle fluctuations also modify the hopping amplitudes. We restrict hopping disorder to the dominant nearest-neighbor (NN) amplitudes, since longer-range hoppings are much weaker and the phases vary little with angle Figure S5,Figure S5). Let denote the mean NN hopping magnitude for an orbital-pair type: MM–MX and MM–XM share one response function since (protected by ), and MX–XM has a second, which reads . As for , we interpolate linearly between the tabulated angles and extrapolate linearly beyond them. For each NN bond, we use the average twist angle of its two endpoint cells to determine the hopping change:
Here is the clean hopping. Only the real amplitude ratio is varied; if it becomes negative under extrapolation, the hopping changes sign. The lattice geometry and longer-range hoppings remain unchanged.
Combining the two contributions (onsite and NN hopping amplitude) gives the extended twist-disorder Hamiltonian
where each NN bond is counted once. Both contributions are generated from the same cell-angle field, so no independent hopping noise is added. The model in the main text (see Eq. (1)) isolates the layer-dependent onsite contribution of twist-angle disorder and adopts a unified strength convention for all three disorder types. For twist angle disorder, Sec. III compares transport with only the onsite correction, only the NN correction, and both contributions retained.
Spatially correlated disorder
To construct spatially correlated disorder, we now introduce spatial correlations into the independent disorder fields generated above. We generate spatially correlated disorder with correlation length (), while keeping the Gaussian distribution at each cell unchanged as Eq. (20). We use the notation , , and , with standard deviations (), to describe disorder field in the real space. The corresponding onsite disorder strengths are , , and in the linearized limit, as discussed in Sec. II A.
The workflow of the construction is summarized in Figure S6, with each step described below:
Figure S6. Workflow of the construction of spatially correlated disorder. Independent Gaussian cell values are extended with random padding on an auxiliary periodic grid. The field is Fourier transformed, multiplied by , and inverse transformed, with . Normalization preserves the original ensemble variance . Selecting the original OBC cells gives the correlated field , representing , , or . Dashed outlines mark the original sample.
Start from the uncorrelated field. We use the independent Gaussian field generated in the previous subsection, denoted here by . The superscript (0) labels the input field, which has zero mean and variance .
Add random padding. We surround the sample with a buffer at least () thick, filled with additional independent values from the same distribution. The enlarged grid is treated as periodic for the fast Fourier transformation (FFT). The buffer keeps the original sample far from these boundaries, reducing artificial correlations caused by periodic repetition. The sample itself retains open boundaries.
Transform to momentum space. The FFT gives the Fourier components
where the sum runs over the enlarged grid. The wave vectors are obtained from the moiré reciprocal vectors.
Smooth the field. We multiply each Fourier component by
This suppresses short-wavelength variations and averages nearby cell values in real space. The filter smoothly suppresses large- components, smoothing the disorder field in real space. Its squared amplitude, , gives a Gaussian spatial correlation of width in the smooth-field limit.
Transform back and preserve the variance. The inverse FFT gives
where is the number of cells in the enlarged grid. Dividing by restores the original variance, . We keep the component and apply no further mean subtraction or rescaling to individual samples.
Select the sample and construct its Hamiltonian. We keep only the values on the original OBC cells and discard the buffer. The resulting , , or is then used to construct the sample Hamiltonian through the mappings in the previous subsection.
The filtered fields remain Gaussian with zero ensemble mean and variance . Their spatial correlation on the finite grid is
The last expression applies when the field varies smoothly over cells and periodic copies have negligible influence. For , , and the linearized twist-angle model, the mappings to preserve this correlation, with replaced by as in the main text. For twist-angle disorder, we first generate the correlated and then map to the onsite energies and NN hoppings.
Using the construction above, Figure S7 illustrates the onsite energy shifts generated from one random Gaussian disorder realization at meV. Increasing produces broader, smoother spatial patterns on each orbital sublattice. Within each cell, shifts all three orbitals equally, whereas gives opposite MX/XM shifts and leaves MM unchanged. For , the construction of angle-dependent onsite terms gives equal MX/XM shifts and an opposite shift on MM.
Figure S7. Orbital-resolved onsite disorder in real space. Rows show (a–c), (d–f), and (g–i); columns correspond to , , and . Each dot represents a Wannier orbital, colored by its onsite energy shift . All panels use the same underlying Gaussian realization and OBC sample with , at the same disorder strength meV. The twist-angle panels use the full onsite-parameter mapping at , with . The common color scale saturates beyond meV, as indicated by the colorbar extensions.
Transport calculation results
In this section, we present additional and results over a wider parameter range, calculated using the six-terminal geometry and the two-terminal geometry, respectively, to examine general trends in plateau behavior. The results are organized into three subsections for potential disorder , displacement-field disorder and twist-angle disorder . For , we compare onsite disorder alone, nearest-neighbor (NN) hopping disorder alone, and both contributions together to distinguish their effects on transport.
Potential disorder
Figures S8 and S9 show the Hall resistance and bulk conductance under disorder at . We consider , 2, 3 meV and , 1, 2, with four sample sizes , 24, 36, 48. Increasing markedly broadens the plateaus at and , reflecting the strong sensitivity to potential disorder. For example, at in the largest sample, the first plateau width increases from approximately to 0.5 as increases from 1 to 3 meV. The intervening Hall dip weakens, and the transition between plateaus becomes narrower and nearly direct. Correspondingly, tends to zero in the plateau regions, while its maximum between them falls from approximately to . At meV, this peak is nearly size independent, consistent with critical bulk transport at a Hall plateau transition.
Figure S10 and Figure S11 show the corresponding results at for the same disorder parameters and sample sizes. Here, the two highest valence bands carry and , with zero total Chern number. Consistent with this clean-band topology, the Hall response shows a single plateau at near . As at , increasing broadens the plateau and suppresses bulk conduction. The interval with nearly vanishing around widens, while the conductance on either side decreases substantially. Outside this interval, the conductance still increases with sample size, although this size dependence becomes weaker at stronger disorder.
Figure S10. Disorder-averaged Hall resistance versus hole filling for potential disorder at , using the six-terminal geometry with open boundaries Figure 2(a). Panels (a)–(c), (d)–(f), and (g)–(i) have , 2, and 3 meV, respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , , , and , respectively. Horizontal dashed lines mark and . The vertical dash-dotted line marks .
Figure S11. Disorder-averaged two-terminal conductance versus hole filling for potential disorder at . The geometry has periodic boundaries along , and the conductance is corrected for the Sharvin contact resistance. Panels (a)–(c), (d)–(f), and (g)–(i) have , 2, and 3 meV, respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , (24, 40), (36, 60), and (48, 80), respectively. The vertical dash-dotted line marks .
At fixed , increasing from 0 to 2 makes the potential smoother and broadens the Hall plateaus at both twist angles [Figure S8–Figure S8) and S10(g–i)]. The corresponding intervals with also widen, consistent with the Hall results [Figure S9–Figure S9) and S11(g–i)].
Figure S8. Disorder-averaged Hall resistance versus hole filling for potential disorder at , using the six-terminal geometry with open boundaries [Figure 2(a)]. Panels (a)–(c), (d)–(f), and (g)–(i) have , 2, and 3 meV, respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , (24, 40), (36, 60), and (48, 80), respectively. Horizontal dashed lines mark and . Vertical dash-dotted lines mark , 2.
Figure S9. Disorder-averaged two-terminal conductance versus hole filling for potential disorder at . The geometry has periodic boundaries along , and the conductance is corrected for the Sharvin contact resistance. Panels (a)–(c), (d)–(f), and (g)–(i) have , 2, and 3 meV, respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , , , and , respectively. Vertical dash-dotted lines mark , 2.
Displacement field disorder
Figures S12 and S13 show the Hall resistance and bulk conductance under disorder at . The disorder strengths, correlation lengths, and sample sizes are the same as for . The plateaus at and remain much narrower and broaden only modestly with increasing , consistent with the weaker sensitivity to disorder. For example, at in the largest sample, the first plateau width increases only from approximately to as increases from to meV. The plateaus remain separated by a broad transition region where dips toward zero. Correspondingly, tends to zero in the plateau regions, while its maximum between them decreases from approximately to , remaining much larger than those of at the same disorder strength and correlation length. At meV and , this maximum still increases with sample size, indicating strong bulk conduction on the simulated length scales.
Similar behavior is found at [Figure S14 and Figure S15]. The single plateau at near broadens only slightly with increasing and remains narrower than under disorder. The bulk conductance is nearly zero in the plateau region and decreases on either side as increases. Nevertheless, it retains a clear increase with sample size outside the plateau, as at .
Figure S14. Disorder-averaged Hall resistance versus hole filling for displacement-field disorder at , using the six-terminal geometry with open boundaries [Fig. 2(a)]. Panels (a)–(c), (d)–(f), and (g)–(i) have , 2, and 3 meV, respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , , , and , respectively. Horizontal dashed lines mark and . The vertical dash-dotted line marks .
Figure S15. Disorder-averaged two-terminal conductance versus hole filling for displacement-field disorder at . The geometry has periodic boundaries along , and the conductance is corrected for the Sharvin contact resistance. Panels (a)–(c), (d)–(f), and (g)–(i) have , 2, and 3 meV, respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , , , and , respectively. The vertical dash-dotted line marks .
Figure S12. Disorder-averaged Hall resistance versus hole filling for displacement-field disorder at , using the six-terminal geometry with open boundaries [Figure 2(a)]. Panels (a)–(c), (d)–(f), and (g)–(i) have , , and meV, respectively. Columns from left to right correspond to , , . Blue, orange, green, and red curves correspond to , , , and , respectively. Horizontal dashed lines mark and . Vertical dash-dotted lines mark , .
At fixed , different from the behavior under disorder, increasing changes the plateau widths only slightly but enhances bulk conduction outside the plateaus. At , this is accompanied by a deeper Hall dip between the two plateaus. For example, at meV in the largest sample, increasing from 0 to 2 leaves the first plateau width near , while the maximum between plateaus rises from approximately 1.7 to 4.3 [Figs. S12(g–i) and S13(g–i)]. At and fixed , the Hall response outside the single plateau is likewise further suppressed, while increases on both sides with increasing [Figs. S14 and S15]. The reason for enhanced Hall dip and bulk conduction under disorder is further discussed in SM. IV.
Figure S13. Disorder-averaged two-terminal conductance versus hole filling for displacement-field disorder at . The geometry has periodic boundaries along , and the conductance is corrected for the Sharvin contact resistance. Panels (a)–(c), (d)–(f), and (g)–(i) have , 2, and 3 meV, respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , , , and , respectively. Vertical dash-dotted lines mark , 2.
Twist angle disorder
To distinguish the onsite and hopping contributions (see Eq. (25)) to twist-angle disorder, we compare three versions of Eq. (25): onsite disorder only, NN hopping disorder only, and both contributions together. The two contributions (onsite and NN hopping) are generated from the same local angle field. All results use , , , and the same sample sizes as above. When the onsite term is included, these angle widths correspond to , respectively, as defined in Eq. (22).
The onsite-only and full twist-angle models show the same qualitative evolution of the Hall plateaus [Figure S17 and Figure S16]. Increasing broadens the plateaus, narrows the intervening transition region, and weakens the Hall dip. The maximum between the plateaus decreases markedly in both models [Figure S20 and Figure S19] with increasing . For , the conductance maximum increases with system size at weak disorder but is nearly size independent at stronger disorder over the system sizes studied. At , this maximum is approximately in both models.
Figure S16. Disorder-averaged Hall resistance versus hole filling for twist-angle disorder at mean twist angle , using the six-terminal geometry with open boundaries [Figure 2(a)]. Both onsite and nearest-neighbor hopping variations are included. Panels (a)–(c), (d)–(f), and (g)–(i) have twist-angle disorder strengths , , and , respectively. Columns from left to right correspond to , , . Blue, orange, green, and red curves correspond to , , , and , respectively. Horizontal dashed lines mark and . Vertical dash-dotted lines mark , .
Figure S17. Disorder-averaged Hall resistance versus hole filling for twist-angle disorder at mean twist angle , using the six-terminal geometry with open boundaries [Fig. 2(a)]. Only the twist-induced onsite variations are retained. Panels (a)–(c), (d)–(f), and (g)–(i) have twist-angle disorder strengths , , and , respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , , , and , respectively. Horizontal dashed lines mark and . Vertical dash-dotted lines mark .
Figure S19. Disorder-averaged two-terminal conductance versus hole filling for twist-angle disorder at mean twist angle . The geometry has periodic boundaries along , and the conductance is corrected for the Sharvin contact resistance. Both onsite and nearest-neighbor hopping variations are included. Panels (a)–(c), (d)–(f), and (g)–(i) have twist-angle disorder strengths , , and , respectively. Columns from left to right correspond to , , . Blue, orange, green, and red curves correspond to , , , and , respectively. Vertical dash-dotted lines mark , .
Figure S20. Disorder-averaged two-terminal conductance versus hole filling for twist-angle disorder at mean twist angle . The geometry has periodic boundaries along , and the conductance is corrected for the Sharvin contact resistance. Only the twist-induced onsite variations are retained. Panels (a)–(c), (d)–(f), and (g)–(i) have twist-angle disorder strengths , , and , respectively. Columns from left to right correspond to , 1, 2. Blue, orange, green, and red curves correspond to , (24, 40), (36, 60), and (48, 80), respectively. Vertical dash-dotted lines mark , 2.
At fixed disorder strength, increasing changes the plateau widths only modestly but substantially enhances conduction between them. For example, at in the largest sample, increasing from 0 to 2 raises the maximum from approximately in the onsite-only model and from to in the combined model. The Hall dip deepens, and the conductance develops a clearer increase with sample size over the simulated length scales. These shared trends show that onsite part of the twist angle disorder dominates the suppression of bulk conduction when both onsite and NN-hopping contributions are present. With NN hopping disorder alone, the Hall response remains strongly suppressed in the transition region between the plateaus, and the conductance maximum increases with sample size throughout the parameter range shown Figure S18 and Figure S21. Nevertheless, increasing the disorder strength still substantially broadens the plateaus. For example, at in the largest sample, the first plateau width increases from approximately to 0.27 as increases from to . Even at , the maximum grows from approximately 4.9 to as increases from 16 to 48. This behavior suggests that hopping disorder tends to localize states near the band edges, stabilizing Hall plateaus while preserving strong bulk conduction in the band interior over the system sizes studied.
Figure S18. Disorder-averaged Hall resistance versus hole filling for twist-angle disorder at mean twist angle , using the six-terminal geometry with open boundaries [Figure 2(a)]. Only the twist-induced variations of nearest-neighbor hoppings are retained. Panels (a)–(c), (d)–(f), and (g)–(i) have twist-angle disorder strengths , , and , respectively. Columns from left to right correspond to . Blue, orange, green, and red curves correspond to , , , and , respectively. Horizontal dashed lines mark and . Vertical dash-dotted lines mark .
Figure S21. Disorder-averaged two-terminal conductance versus hole filling for twist-angle disorder at mean twist angle . The geometry has periodic boundaries along , and the conductance is corrected for the Sharvin contact resistance. Only the twist-induced variations of nearest-neighbor hoppings are retained. Panels (a)–(c), (d)–(f), and (g)–(i) have twist-angle disorder strengths , , and , respectively. Columns from left to right correspond to . Blue, orange, green, and red curves correspond to , , , and , respectively. Vertical dash-dotted lines mark .
Finite-size scaling
In this part, we use finite-size scaling [67, 68, 69, 70] to examine the critical behavior of the bulk conductance in Figure 3(d) of the main text for and disorder and Figure S20 for disorder. Our noninteracting single-valley model belongs to the unitary class (class A), which in two dimensions supports delocalized states in the Hall or Chern plateau transition regions [71] [2]. The strong sensitivity to favors bulk localization, leaving a nearly size-independent conductance peak. We therefore fit the region around this peak [Fig. S22(a)]. For and , weaker effective disorder leaves a broad conducting region on the simulated length scales. We instead analyze the crossings near its lower-filling edge [Fig. S22(b,c)].
We fit the four sample sizes using the single-parameter scaling form
where and are the fitted critical filling and localization-length exponent, respectively. The reference size 24 only sets the scale of . The fitting windows are shaded in the upper panels. We represent by a quadratic polynomial in and and by cubic polynomials, using weighted least squares with the standard errors of the mean.
The rescaled data collapse onto the fitted curves [Fig. S22(d–f)], with , 2.69, and 2.64 for , , and , respectively. For these fixed windows and fitting forms, the corresponding 90% bootstrap intervals are [2.32, 2.78], [2.41, 3.04], and [2.39, 2.91]. All three fitted exponents lie close to , the characteristic value of the integer quantum Hall universality class [72, 73], supporting compatibility with this critical scaling despite the different finite-size conductance profiles.
Figure S22. Finite-size scaling of the disorder-averaged bulk conductance. (a–c) versus for , , and disorder, respectively, corresponding to Figure 3(d) of the main text and Figure S20. The strengths are meV and (). Colors denote , , , and , as indicated in (a). Yellow shading marks the fitting windows. (d–f) The selected data plotted against the rescaled variable defined in Eq. (30); black dashed curves show the fits. Labels give the fitted and its 90% bootstrap confidence interval, evaluated from 400 resamples with the fitting window and functional form held fixed. They measure window sensitivity, not statistical confidence intervals. The other parameters are , and .
Charge response
In this section, we derive the intraband charge density susceptibility (Eq. (3)) and root-mean-square charge response amplitude (Eq. (4)) used in the main text. The clean-band charge response provides a measure of the system’s sensitivity to each disorder channel within the perturbation theory and provides a qualitative perspective on the transport results in Sec. III.
Derivation of the charge density response
The charge density is the total occupation of a moiré cell, , whereas the disorder couples to . Here . We consider periodic cells and use the unitary Fourier convention , with . The annihilation operator for a clean Bloch state in band is
where is the component of the Bloch eigenvector on Wannier orbital . Projecting both the density operator and the disorder coupling operator onto the target band gives
where . The disorder-induced perturbation within the target band is . At fixed chemical potential , static linear response gives
Here is the change in the equilibrium density, , denotes imaginary-time ordering, and is the connected thermal average in the clean system. Both operators are fermion bilinears. Evaluating their four-fermion correlation using Wick’s theorem gives
Here . The Bloch-state overlap comes from summing the density over all three orbitals, while describes the applied disorder channel. Equation (S27) reproduces Eq. (3) of the main text.
For zero-mean disorder, the induced density change also has zero ensemble mean. We therefore characterize its typical magnitude by the root-mean-square charge response. To perform the disorder average, we write
Fourier transformation gives
Thus, describes how the fixed disorder variance is distributed among momentum transfers. For onsite fields generated by the Gaussian filter in Eq. (27),
This is the spectrum corresponding to the finite-grid covariance in Eq. (29). For , ; increasing shifts weight toward small while preserving the local variance in real space. For a nonlinear twist-angle mapping, is obtained from the covariance of the mapped onsite fluctuations.
Using Eq. (33) and Parseval’s identity, the charge response amplitude becomes
This gives Eq. (4) of the main text for both uncorrelated and spatially correlated disorder.
Charge density response in tMoTe2 and connection to transport
Figure S23 shows calculated from Eq. (38) for the highest two valence bands at and . The chemical potential is expressed through the clean-band hole filling . For every shown, the main response maxima follow the same order: is strongest, is weakest, and lies between them. This reflects the partial cancellation of opposite orbital contributions for and compared with in the total-charge response.
Figure S23. Charge response versus hole filling at . (a–c) top and second highest -valley valence band with twist-angle . Columns correspond to from left to right, with correlation length . Red, blue, and green curves denote , , and the onsite component of disorder, respectively. The amplitudes are normalized by the standard deviation of the onsite fluctuations and have units . All results use and a Fermi-function smoothing scale of .
Increasing enhances the main response maxima from approximately 0.82 to 1.59 in the top band and from 0.53 to 1.13 in the second band [Fig. S23(a–c)]. The response remains much smaller, changes little between . The main maxima also increase, while remaining below those of . These trends follow from the weighting in Eq. (38): a smoother field emphasizes the small-momentum charge response, which depends on the orbital channel.
Spatial smoothing changes the effective disorder strength, modifying the charge response and disorder sensitivity through the orbital-dependent kernel . It also suppresses large-momentum-transfer scattering, which can reduce backscattering and enhance bulk conduction.
For , the charge response increases substantially with , indicating greater sensitivity to smooth potential disorder. The Hall plateaus also show clear broadening as increases from 0 to 2 at and meV [Fig. S8(g–i)]. The plateau broadening suggests that enhanced disorder sensitivity plays the dominant role, despite the reduction in large-momentum scattering.
For , the charge response remains weak and shows no comparable enhancement with . At the same angle, taking , meV as an example, increasing from 0 to 2 raises the conductance maximum from approximately 1.7 to 4.3 [Fig. S13(g–i)]. The Hall dip deepens while the plateau widths change only slightly [Fig. S12(g–i)]. These trends are consistent with reduced backscattering enhancing bulk conduction while the disorder sensitivity remains weak.
For disorder, the main charge-response maxima lie between those of and , with a modest increase from to 2 [Fig. S23(b,c)]. In the onsite-only model at , the conductance maximum between plateaus increases as grows from 0 to 2 [Fig. S20(g–i)]. This suggests that reduced backscattering plays the dominant role in enhancing bulk conduction, despite the increased disorder sensitivity.
References
- [34]J. K. Jain, Composite-fermion approach for the fractional quantum Hall effect, Phys. Rev. Lett. 63, 199 (1989).pubmed.ncbi.nlm.nih.gov/10040805
- [35]A. Lopez and E. Fradkin, Fractional quantum Hall effect and Chern-Simons gauge theories, Phys. Rev. B 44, 5246 (1991).pubmed.ncbi.nlm.nih.gov/9998334
- [36]Wu, F., Law, K. T., Ivanov, E., Martin, T., & MacDonald, A. H. (2019). Topological insulators in twisted transition metal dichalcogenide homobilayers. Phys. Rev. Lett., 122, 086402.
- [37]T. Devakul, V. Crépel, Y. Zhang, and L. Fu, Magic in twisted transition metal dichalcogenide bilayers, Nat. Commun. 12, 6730 (2021).DOI
- [38]A. P. Reddy, F. Alsallom, Y. Zhang, T. Devakul, and L. Fu, Fractional quantum anomalous Hall states in twisted bilayer MoTe2 and WSe2, Phys. Rev. B 108, 085117 (2023).DOI
- [39]C. Wang, X.-W. Zhang, X. Liu, Y. He, X. Xu, Y. Ran, T. Cao, and D. Xiao, Fractional Chern insulator in twisted bilayer MoTe2, Phys. Rev. Lett. 132, 036501 (2024).DOI
- [40]Y. Jia, J. Yu, J. Liu, J. Herzog-Arbeitman, Z. Qi, H. Pi, N. Regnault, H. Weng, B. A. Bernevig, and Q. Wu, Moiré fractional Chern insulators. I. first-principles calculations and continuum models of twisted bilayer MoTe2, Phys. Rev. B 109, 205121 (2024).arxiv.org/abs/2311.04958
- [41]Zhang, Y., Pi, H., Liu, J., Miao, W., Qi, Z., Regnault, N., Weng, H., Dai, X., Bernevig, B. A., Wu, Q., & Yu, J. (2024). Universal moiré-model-building method without fitting: Application to twisted MoTe2 and WSe2. https://arxiv.org/abs/2411.08108
- [42]A. A. Mostofi, J. R. Yates, Y.-S. Lee, I. Souza, D. Vanderbilt, and N. Marzari, wannier90: A tool for obtaining maximally-localised wannier functions, Computer Physics Communications 178, 685 (2008).DOI
- [43]W.-X. Qiu, B. Li, X.-J. Luo, and F. Wu, Interaction-driven topological phase diagram of twisted bilayer MoTe2, Phys. Rev. X 13, 041026 (2023).DOI
- [44]C. Xu, J. Li, Y. Xu, Z. Bi, and Y. Zhang, Maximally localized Wannier functions, interaction models, and fractional quantum anomalous Hall effect in twisted bilayer MoTe2, Proc. Natl. Acad. Sci. U.S.A. 121, e2316749121 (2024).DOI
- [45]V. Crépel and A. Millis, Bridging the small and large in twisted transition metal dichalcogenide homobilayers: A tight binding model capturing orbital interference and topology across a wide range of twist angles, Phys. Rev. Res. 6, 033127 (2024).DOI
- [46]Z. Qi, H. Pi, Y. Zhang, J. Liu, N. Regnault, H. Weng, B. A. Bernevig, J. Yu, and Q. Wu, Chern-selective multivalley flat bands in twisted mono-bilayer and mono-trilayer MoTe2, Phys. Rev. B 113, 125116 (2026).DOI
- [47]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).DOI
- [48]B. A. Foutty, C. R. Kometter, T. Devakul, A. P. Reddy, K. Watanabe, T. Taniguchi, L. Fu, and B. E. Feldman, Mapping twist-tuned multiband topology in bilayer WSe2, Science 384, 343 (2024).DOI
- [49]M. Büttiker, Four-terminal phase-coherent conductance, Phys. Rev. Lett. 57, 1761 (1986).pubmed.ncbi.nlm.nih.gov/10033538
- [50]C. W. Groth, M. Wimmer, A. R. Akhmerov, and X. Waintal, Kwant: a software package for quantum transport, New J. Phys. 16, 063065 (2014).DOI
- [51]K. M. Schep, J. B. A. N. van Hoof, P. J. Kelly, G. E. W. Bauer, and J. E. Inglesfield, Interface resistances of magnetic multilayers, Phys. Rev. B 56, 10805 (1997).DOI
- [52]A. Stern and L. Fu, Transport properties of a half-filled Chern band at the electron and composite fermion phases, Phys. Rev. Lett. 133, 246602 (2024).DOI
- [53]T. Lu and L. H. Santos, Fractional Chern insulators in twisted bilayer MoTe2: A composite fermion perspective, Phys. Rev. Lett. 133, 186602 (2024).DOI
- [54]X. Hu, D. Xiao, and Y. Ran, Hyperdeterminants and composite fermion states in fractional Chern insulators, Phys. Rev. B 109, 245125 (2024).arxiv.org/abs/2312.00636
- [55]T. Lu, Y.-M. Wu, and L. H. Santos, Electromagnetic response and emergent topological orders in transition metal dichalcogenide MoTe2 bilayers, Phys. Rev. B 112, 085138 (2025).DOI
- [56]X. Hu, Y. Ran, and D. Xiao, Composite fermion theory of fractional Chern insulator stability, Phys. Rev. Lett. 136, 066504 (2026).DOI
- [57]H. Jin and J. Shi, Generalizing the composite fermion theory for fractional Chern insulators, Phys. Rev. B 114, L111106 (2026).DOI
- [58]A. S. Patri, Z. Dong, and T. Senthil, Extended quantum anomalous Hall effect in moiré structures: Phase transitions and transport, Phys. Rev. B 110, 245115 (2024).DOI
- [59]Ding, P., Millis, A. J., & Zeng, Y. (2026). Composite fermion theory of fractional Chern insulators in rhombohedral graphene. https://arxiv.org/abs/2601.07174
- [60]T. Ozaki, Variationally optimized atomic orbitals for large-scale electronic structures, Physical Review B 67, 155108 (2003).DOI
- [61]T. Ozaki and H. Kino, Numerical atomic basis orbitals from H to Kr, Physical Review B 69, 195113 (2004).DOI
- [62]T. Ozaki and H. Kino, Efficient projector expansion for the ab initio LCAO method, Physical Review B 72, 045121 (2005).DOI
- [63]W. Miao, C. Li, X. Han, D. Pan, and X. Dai, Truncated atomic plane wave method for subband structure calculations of moiré systems, Phys. Rev. B 107, 125112 (2023).DOI
- [64]Zhang, Y. (2024). Continuum model for twist tmds. <a href="https://doi.org/https://github.com/zhangyan31415/continuum-model-twist-tmds" rel="noopener noreferrer" target="_blank">https://doi.org/https://github.com/zhangyan31415/continuum-model-twist-tmds</a>
- [65]Z.-D. Song and B. A. Bernevig, Magic-angle twisted bilayer graphene as a topological heavy fermion problem, Phys. Rev. Lett. 129, 047601 (2022).DOI
- [66]J. Yu, M. Xie, B. A. Bernevig, and S. Das Sarma, Magic-angle twisted symmetric trilayer graphene as a topological heavy-fermion problem, Phys. Rev. B 108, 035129 (2023).DOI
- [67]E. Abrahams, P. W. Anderson, D. C. Licciardello, and T. V. Ramakrishnan, Scaling theory of localization: Absence of quantum diffusion in two dimensions, Phys. Rev. Lett. 42, 673 (1979).DOI
- [68]J. L. Pichard and G. Sarma, Finite size scaling approach to Anderson localisation, Journal of Physics C: Solid State Physics 14, L127 (1981).DOI
- [69]A. MacKinnon and B. Kramer, One-parameter scaling of localization length and conductance in disordered systems, Phys. Rev. Lett. 47, 1546 (1981).DOI
- [70]Z. Qi, Y. Zhang, M. Qin, H. Weng, and K. Jiang, Anderson localization: A density matrix approach, Phys. Rev. X 16, 011043 (2026).arxiv.org/abs/2509.26206
- [71]F. Evers and A. D. Mirlin, Anderson transitions, Rev. Mod. Phys. 80, 1355 (2008).DOI
- [72]E. J. Dresselhaus, B. Sbierski, and I. A. Gruzberg, Scaling collapse of longitudinal conductance near the integer quantum Hall transition, Phys. Rev. Lett. 129, 026801 (2022).DOI
- [73]S. Bera, J. Dieplinger, and N. P. Nayak, Quantum Hall criticality in an amorphous Chern insulator, Phys. Rev. B 109, 174213 (2024).DOI


























