Article Content
Introduction
Moiré heterostructures have recently emerged as versatile quantum simulators of archetypal condensed matter models1,2. When two identical or nearly identical monolayers are twisted, the resulting moiré modulation of the interlayer potential gives rise to an effective moiré discrete translation symmetry. In the moiré BZ, the moiré-modulated interlayer hybridization opens gaps in the folded band structure, quenching the kinetic energy of the monolayer electrons10. The moiré system thus enters an interaction-dominated regime, providing a tunable platform for simulating various prototypical condensed matter systems. A notable example is twisted bilayer graphene5, which hosts unconventional superconductors11 and correlated insulators12 near the magic angle and has recently been shown to simulate topological heavy-fermions13,14. Transition-metal dichalcogenide (TMD) heterobilayers can emulate the Hubbard model on a triangular lattice7,15, whereas twisted WTe2 exhibits signatures of a one-dimensional Luttinger liquid, although its theoretical description remains challenging owing to the complex monolayer band structure16. Beyond these examples, a growing body of theoretical and experimental work has explored other exotic phases in TMDs17,18,19,20,21,22. Furthermore, both integer and fractional Chern insulator states have been reported in moiré TMD23,24,25,26,27,28,29,30, graphene31,32,33 and graphene–boron nitride heterostructures34,35,36,37.
Until now, nearly all moiré heterostructures have been based on twisting monolayers with triangular lattices and low-energy states near the Γ (refs. 3,4) or K (refs. 5,6,7,8,9) points, leading to systems with one or two valleys (in the two-valley case, time-reversal exchanges the valley). This work introduces a new family of moiré materials by twisting monolayers with triangular lattices and low-energy states around the M point of the BZ. These M-point moiré systems feature three time-reversal-preserving valleys related by C3z rotation symmetry. Building on extensive ab initio calculations, we propose (among others38) experimentally exfoliable twisted SnSe2 and ZrS2 as promising platforms for realizing M-point moiré heterostructures. We develop quantitative simplified models for these systems and perform a detailed analysis of the band structure, topology and charge density of the flat bands at the predicted small twist angles. We show analytically that M-point moiré Hamiltonians exhibit a new type of symmetry, termed momentum-space non-symmorphic39,40,41,42,43. In crystallography, space groups are symmorphic or non-symmorphic, depending on whether they include symmetry operations that translate the origin by a fraction of the lattice vectors. Although in real space conventional crystalline groups can feature both symmorphic and non-symmorphic operations, in momentum space, all conventional crystalline groups exhibit only symmorphic operations. M-point moiré systems are the first experimentally realizable non-magnetic systems to exhibit momentum-space non-symmorphic symmetries, all without requiring an applied magnetic field in the range of thousands of Tesla39,40,41. In a single valley, these non-symmorphic symmetries can render the system effectively one-dimensional at the single-particle level, making M-point moiré systems prime candidates for Luttinger-liquid simulators16,44. With all three valleys considered, they can realize a multi-orbital triangular lattice Hubbard model (H. Hu et al., to be published), in which valley-spin local moments couple differently along the three C3z-related directions, in a manner reminiscent of Kitaev’s honeycomb model45.
M-point moiré models
For triangular monolayer lattices, the moiré lattice is also triangular, generated by the reciprocal lattice vectors and (see Supplementary Information Section IV). These vectors span the moiré reciprocal lattice , as depicted in Fig. 1a. In general, the single-particle Hamiltonians of moiré systems take the form of a hopping model in momentum space. This arises because the moiré potential breaks the monolayer translation symmetry and couples momentum states that are connected by reciprocal moiré vectors. The single-particle moiré Hamiltonian can be written , in which denotes the moiré plane-wave operators at moiré momentum k, and i denotes a combined index comprising orbital, spin, valley, layer or other further degrees of freedom.

a–c, The three panels correspond to the cases in which the low-energy degrees of freedom are located at the Γ (a), K (b) and M (c) points. In each panel, the sublattices are coloured according to the legend above each plot. The moiré BZ is shown by the grey hexagon, whereas the reciprocal moiré vectors as well as the auxiliary vectors and are shown by the black arrows.
When the low-energy fermions of the monolayer are located at the Γ point3,4, the operators carry total momentum k − Q and the Q-vectors lie on the triangular lattice shown in Fig. 1a. In the case of a monolayer with low-energy states located at the K point5,6,7, the moiré fermions carry an extra valley index η = ±, in which the moiré Hamiltonian is diagonal. The Q-vectors form a honeycomb lattice, as illustrated in Fig. 1b. The moiré and monolayer operators are related by for , in which represents the monolayer operators from layer l = ± at momentum p and is the K-point momentum of layer l.
Distinctly, in M-point moiré materials, the Q-vectors form a kagome lattice, as shown in Fig. 1c. To be specific, the moiré operators in layer l—which, for the present case, include only an extra spin s = ↑,↓ index—are related to the monolayer ones according to , for , in which is the momentum of the monolayer M point. The three C3z-related valleys indexed by η = 0, 1 and 2 are implicitly encoded by the kagome sublattice to which Q belongs: the valley-η fermions are supported on the sublattices (in which η + l is taken modulo 3), as derived in Supplementary Information Section IV. As we will show, the kagome Q-lattice leads to substantially different properties of M-point moiré materials.
Materials realizations
We now turn to 1T-SnSe2 and 1T-ZrS2 as experimentally exfoliable monolayers for realizing M-point moiré heterostructures (see Supplementary Information Section II). The monolayer crystal structure of both materials is shown in Fig. 2a,b and belongs to the group, which is generated by translations, C3z rotations, in-plane twofold rotations C2x, inversion and time-reversal symmetries. The Sn (Zr) atoms form a triangular lattice, with the Se (S) atoms being located at the other C3z-invariant Wyckoff positions above and below the Sn (Zr) plane. The ab initio band structures of monolayer SnSe2 and ZrS2 shown in Fig. 2c,d reveal two insulators for which the conduction band minimum is located at the M point. The first isolated Kramers-degenerate conduction band of SnSe2 is atomic, being spanned by an effective s-like molecular orbital centred on the Sn atom. For ZrS2, the low-energy M-point states are contributed primarily by the orbitals of Zr.

a,b, Side and top views of the crystal structures of 1T-SnSe2 and 1T-ZrS2. c,d, The ab initio band structures for SnSe2 and ZrS2, respectively. The lowest spinful conduction band, with minima at the M points, is highlighted in red, whereas the Wannier orbitals contributing to the low-energy states are shown as insets. The yellow (blue) colours correspond to the positive (negative) sign of the orbitals.
Moiré Hamiltonians
Because the SnSe2 and ZrS2 monolayers lack twofold out-of-plane rotation symmetry (C2z), there are two distinct ways to stack and subsequently twist them by an angle θ to achieve a large-scale moiré periodicity. In the so-called AA-stacking configuration, the top (l = +1) and bottom (l = −1) layers are stacked directly on top of each other and then twisted by the layer-dependent angle . By contrast, for AB-stacking, the bottom layer is first rotated by 180° around the axis, before applying the twist. As discussed in Supplementary Information Section III, the two configurations have different crystalline symmetries. Although both stackings feature C3z and symmetries, they differ in the direction of the in-plane twofold rotation symmetry: the AA (AB)-stacking arrangement has C2x (C2y) symmetry.
We perform large-scale ab initio calculations (which include relaxation effects) at commensurate twist angles 13.17° ≥ θ ≥ 3.89° (see Methods and Supplementary Information Sections III and VIII) and construct two types of moiré Hamiltonian model for each angle and stacking configuration according to the method outlined in Supplementary Information Section IX. The first is a numerically exact model, which accurately reproduces a large set of spinful bands (at least the first five in each valley) in both energy and wavefunction. The second is an analytical approximate continuum model capturing the dispersion and wavefunction of the first or first two (depending on the angle) lowest-energy spinful gapped bands (and, qualitatively, the higher-energy spectrum) in each valley. The comprehensive results at all angles are presented in Supplementary Information Section XI. Unlike the case of Γ-point or K-point twisting, ab initio simulations are crucial for obtaining even the correct qualitative moiré Hamiltonian. The two-centred first-monolayer harmonic approximation incorrectly predicts continuous translation symmetry along one direction (for example, along the direction in valley η) and an overall gapless spectrum, as shown in Supplementary Information Section VI.
Figure 3 summarizes the ab initio results for twisted AA-stacked and AB-stacked SnSe2 and ZrS2 at low twist angle. Both stacking configurations exhibit approximate spin SU(2) symmetry (see Supplementary Information Section IX) and feature two sets of spinful gapped bands in each of the three C3z-related valleys, as shown in Fig. 3a–d. The lowest-energy set of bands has a narrow bandwidth of around 10 meV. The charge density distribution (CDD) for the lowest two bands in valley η = 0, shown in Fig. 3e–h, reveals that these moiré systems have approximate spatial symmetries beyond the exact valley-preserving C2x and C2y symmetries expected in the AA-stacked and AB-stacked configurations, respectively. For instance, the CDD of the first set of spinful bands in AA-stacked SnSe2, as well as the first two sets of bands in twisted ZrS2, feature an approximate twofold rotation symmetry (the second set of spinful bands in AA-stacked SnSe2 exhibits this symmetry to a lesser extent). In the AB-stacked configuration, the centre of the approximate C2z symmetry aligns with the unit cell origin, whereas in the AA-stacked case, the effective rotation centre is shifted away from the unit cell origin and will be specified below. Moreover, the CDD suggests the presence of an approximate in-plane mirror symmetry, . These effective symmetries (whose origin is explained below and in Supplementary Information Section VB) prompt us to construct simplified analytical continuum models that can capture and explain these features.

a–d, Band structures for AA-stacked (a,c) and AB-stacked (b,d) twisted SnSe2 and ZrS2 at the commensurate angle θ = 3.89°. Both the ab initio and valley-resolved continuum model band structures are shown. e–h, The layer-resolved CDD corresponding to the first and second sets of spinful bands in valley η = 0. The Wigner–Seitz unit cell is indicated by the dashed hexagon.
In valley η = 0, the simplified M-point moiré Hamiltonian can be expressed as
in which mx and my are the anisotropic effective masses of SnSe2 and ZrS2 (see Methods). As shown in Fig. 4a,b, the moiré potential takes the form of a hopping model on two of the three sublattices of the kagome M-point Q-lattice. Explicitly, the simplified Hermitian moiré potential tensor exhibits spin SU(2) symmetry and includes only interlayer terms, given by and . The interlayer hopping parameters, obtained by fitting to the ab initio band structure, are listed in Methods. The band structure of the simplified model for AA-stacked SnSe2 is shown in Fig. 4c, indicating excellent qualitative agreement with the ab initio results for such a small number of parameters. In the simplified models, for both SnSe2 and ZrS2, the overlap between the fitted and ab initio bands is larger than 95% (85%) with the first (second) set of spinful bands, as we show in Supplementary Information Section XI.

a, Relationship between the monolayer and moiré BZs, with the coloured and grey hexagons representing the respective BZs. b, Generation of the -lattice in the η = 0 valley, showing the hopping terms of the moiré potential matrix . c, The band structure of the simplified moiré model for AA-stacked SnSe2 at θ = 3.89°. The colour scheme matches that of Fig. 3.
Momentum-space non-symmorphic symmetries
The approximate symmetries inferred from the layer-resolved CDD of the M-point moiré Hamiltonian are exact symmetries in the simplified moiré models from equation (1) (see detailed discussion in Supplementary Information Section VI). Specifically, the centre of the effective twofold rotation symmetry for the AA-stacked Hamiltonian is located at in valley η. By contrast, the simplified AB-stacked moiré Hamiltonian exhibits C2z symmetry, with its rotation centre aligned with the origin of the moiré unit cell. Because both models are effectively spinless (owing to atomistic arguments presented in Supplementary Information Section XI) and exhibit either or symmetry in each valley, the Berry curvature of any gapped set of bands is exactly zero. Consequently, the first two sets of bands of both the AA-stacked and AB-stacked moiré Hamiltonians are topological trivial and, hence, Wannierizable. This is also consistent (and the result of) the bands being flat and exhibiting a large (40 meV) gap from one another. However, the physics of these Hubbard (with interaction) bands is far from trivial in this system, as shown below.
Unlike the C2z and symmetries, the effective mirror symmetry has an unconventional action on the momentum-space moiré fermions. Specifically, acts non-symmorphically in momentum space, with for . Because , the action of can only be made conventional by folding the moiré BZ along , which would break the moiré translation symmetry. The non-symmorphic action of the symmetry originates from the moiré fermions realizing a projective representation of the symmetry group of the system. Letting denote the two moiré translation operators for valley η = 0 along the direct moiré lattice vectors (with ), we find that (contrasting with a conventional mirror Mz symmetry, which would commute with both and ).
It is important to note that the effective symmetry is not accidental. In the AA-stacked case, it can be shown to hold exactly for arbitrary moiré harmonics within the local-stacking approximation46. In the limit of vanishing twist angle (θ → 0), the moiré Hamiltonian can be constrained by the exact symmetries of the untwisted bilayer configuration. The inversion symmetry of the untwisted AA-stacked bilayer gives rise to the symmetry of the moiré Hamiltonian, as shown in Supplementary Information Sections VB and VI. In the AB-stacked case, the true in-plane mirror symmetry of the untwisted bilayer leads to an effective inversion symmetry of the corresponding moiré Hamiltonian, which also acts non-symmorphically in momentum space. In the simplified AB-stacked model, the approximate C2z symmetry, combined with the symmetry, leads to an symmetry of the system.
Projective fermion representations that realize momentum-space non-symmorphic symmetries have previously been proposed in magnetic systems42 or systems subjected to a large magnetic field (on the order of thousands of Tesla)40,41,47. M-point moiré materials provide the first experimentally viable realization of these symmetries in any (that is, magnetic or non-magnetic) system. To better understand the origin of the momentum-space non-symmorphic action of the symmetry, we construct a simple one-dimensional tight-binding model that incorporates it. The resulting ladder model, shown in Fig. 5a, mimics the dispersion of an atomic band in the M-point moiré Hamiltonian for valley η = 0 along the direction (see Supplementary Information Section VI). Each unit cell is threaded by a uniform perpendicular magnetic field, enclosing a π-flux. Because π-flux and (−π)-flux are equivalent, the model also respects time-reversal and symmetry. In the Fourier-transformed basis , the symmetry acts non-symmorphically as , ensuring that the spectra of the Hamiltonian at k and k + π are identical, as shown in Fig. 5b.

a, A ladder tight-binding model with magnetic flux that realizes the symmetry. Fermion operators and hopping amplitudes are indicated above each site (black dots). b, Dispersion relation. c, Energy dispersion of the first band of AA-stacked SnSe2 at θ = 3.89° in the first moiré BZ. d, Schematic illustration of the corresponding quasi-one-dimensional character of atomic bands in M-point moiré systems for valley η = 0. Each Wannier orbital (dots) is coloured according to its eigenvalue. The grey rectangle represents the rectangular unit cell of each symmetry sector.
Hubbard and Luttinger simulators
Within each valley, the first two sets of spinful bands in SnSe2 and ZrS2 bilayers are individually Wannierizable, with their bandwidths tunable by adjusting the twist angle. Given the excellent SU(2) symmetry, these M-point moiré systems become effective simulators of the Hubbard model when Coulomb interactions are included (H. Hu et al., to be published). However, owing to the extra valley degree of freedom, these systems go beyond the single-band U(2) Hubbard model, instead realizing a six-flavour U(2) × U(2) × U(2) Hubbard model.
Another key distinction from the standard Hubbard model can arise from the symmetry. In real space, does not change the position along the moiré heterostructure. As a result, the continuum moiré Hamiltonian can be made diagonal in the basis. Because and both commute with , each mirror sector of valley η = 0 will feature reduced translation symmetry specified by the rectangular lattice vectors and . The operator anticommutes with , exchanging the two mirror sectors. The Wannier orbitals of any atomic band—such as the first conduction band of AA-stacked SnSe2 from Fig. 5c—can therefore be split by their eigenvalues: the orbitals of each mirror sector are displaced by and form two interpenetrating rectangular lattices shown in Fig. 5d. Within each mirror sector and in valley η = 0, the interorbital separation is larger by a factor of along the direction compared with the one. Provided that the Wannier orbital spread is approximately isotropic (as it happens for the first band of AA-stacked SnSe2 but not in the first band of twisted ZrS2), this will lead to reduced hopping along compared with (see Supplementary Information Section VI). As the tunnelling between sectors is forbidden, the system in each valley will behave quasi-one-dimensionally, with flatter dispersion along the direction, effectively emulating a Luttinger model. In the three-valley system, this quasi-one-dimensional behaviour causes the U(2) × U(2) × U(2) local moments to couple differently along three C3z-related directions, similar (but not identical) to the couplings of the Kitaev model45.
We note, however, that quasi-one-dimensionality along the direction (that is, flatter dispersion along the direction) is not an inherent or universal feature of M-point moiré materials. Instead, it is the presence of the effective symmetry, not previously identified, that plays a more general role. Together with approximately isotropic Wannier orbitals for the bands, the effective symmetry can enforce one-dimensional behaviour in the single-particle valley-projected moiré Hamiltonian. However, this symmetry is also compatible with two-dimensional physics in general (see Methods). For instance, because of the elongated Wannier orbitals, twisted ZrS2 exhibits excellent effective symmetry, but its first set of conduction bands is not quasi-one-dimensional along the direction.