WorksPhysics and Astronomy
Moiré trapping of quadrupolar excitons in van der Waals trilayers
Abstract
Quadrupolar excitons in van der Waals heterostructures - quantum superpositions of anti-aligned dipolar excitons - offer a novel platform to explore exotic many-body physics, with applications to quantum sensing and photonic devices. Yet their internal structure, symmetry, and real-space localisation remain largely unknown. Here, we reveal the atomic-scale structure of quadrupolar excitons in twisted WSe2/WS2/WSe2 trilayers by solving the Bethe-Salpeter equation within a large-scale atomistic framework. We discover that large atomic relaxations at small twist angles give rise to two distinct quadrupolar excitons trapped at moiré lattice sites, differing in the in-plane symmetry of the electron density about the hole: one azimuthally symmetric, with the density maximal at the hole, and one threefold symmetric, with a node at the hole. Moiré trapping, neglected in commonly used models of quadrupolar exciton formation, is critical to their many-exciton phases. Without moiré trapping, quadrupolar excitons transition into anti-parallel dipolar excitons on a bipartite square lattice, while with trapping, the same dipoles are confined to a triangular lattice and experience geometric frustration. Our study uncovers the highly non-trivial nature of quadrupolar excitons, with direct implications for simulating frustrated quantum magnetism in a fully tunable excitonic platform.
Introduction
Introduction. Multipole expansions encode the symmetry of a source through its moments, with applications spanning molecular electrostatics[1], gravitational radiation[2], and many-body physics[3, 4]. For example, in electrostatics, dipoles arise from separated positive and negative charges, while quadrupoles appear in more symmetric arrangements where both the total charge and dipole moment vanish. Recently, this framework has found a striking realization in van der Waals heterostructures[5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]. By stacking and twisting ultrathin two-dimensional (2D) materials, experimentalists have demonstrated tunable dipole moments of excitons, which are bound electron–hole pairs. For example, excitons in symmetric trilayer transition metal dichalcogenides such as [5, 6] and [7] show a transition from quadrupolar to dipolar character when an out-of-plane electric field is applied. The absence of a dipole moment leads to qualitatively distinct interaction physics, making these systems an attractive platform for exploring exotic many-body phases[19, 20, 21] and optoelectronic applications[22, 23]. To date, the dipolar and quadrupolar characters have been inferred almost exclusively from the electric-field dependence of photoluminescence (PL), which shifts from linear to non-linear behavior[24]. Direct real-space imaging of their structure, symmetry, and twist-angle dependence remains elusive.
Accurate theory is essential to close this gap, provide atomistic insight, and guide future experiments. Yet, theoretical modeling remains scarce. This scarcity is mainly due to the prohibitive computational cost of state-of-the-art many-body perturbation theory methods, such as the GW-plus-Bethe-Salpeter equation (GW-BSE) approach, for simulating multilayer 2D materials with hundreds to thousands of atoms[25, 26, 27, 28]. As a result, GW–BSE calculations have been limited to lattice-matched trilayer systems containing only a few atoms[13, 29, 30, 31, 32]. However, stacking and twisting 2D materials drive pronounced atomic rearrangements that qualitatively alter exciton behavior. Such effects are entirely absent in the lattice-matched models. For example, in bilayers, atomic rearrangements give rise to localized intralayer excitons in with strong electron-hole overlap and charge-transfer character[33, 34, 35, 36, 37, 38]. Structural relaxations are therefore essential for any predictive description of multipolar excitons and their field-tunable moments. Furthermore, accurate parameters characterizing individual quadrupolar and dipolar excitons are essential input for modeling the many-exciton phase diagram[30, 39]. Their importance is evident from a striking anomaly[5]: in the exciton density regime where experiments observe quadrupolar excitons, models that neglect structural relaxations instead predict a staggered dipolar phase.
Here, we overcome these challenges and provide the first atomistic description of the electric-field-driven quadrupolar-to-dipolar transition in twisted trilayer , fully accounting for structural relaxations. To this end, we develop a transfer-matrix formalism for electron–hole Coulomb interactions in arbitrarily twisted multilayer 2D materials and solve the fully atomistic BSE using ab initio electronic structure as input to determine the exciton spectrum and its real-space structure. We demonstrate that quadrupolar excitons form in trilayers with parallel outer layers, regardless of the twist angle with . At small twist angles, large atomic rearrangements trap quadrupolar excitons at moiré sites, splitting them into two species that differ in the in-plane symmetry of the electron density about the hole: one azimuthally symmetric with maximal density at the hole, and one threefold symmetric with a node at the hole. Moiré trapping resolves the anomaly[5]. By confining excitons to a triangular lattice, it induces geometric frustration that stabilises the quadrupolar phase across the experimentally observed density range.
Bethe-Salpeter equation in multilayers. We construct a fully atomistic BSE Hamiltonian using Wannier functions obtained from large-scale density functional theory (DFT) calculations of arbitrarily twisted multilayers[40].
Here, are the quasiparticle energies of conduction (valence) states, are the Wannier function coefficients, is the position of the -th Wannier function in the home unit cell, are lattice vectors, is the number of -points sampled in the Brillouin zone, and and are the screened and bare Coulomb interactions. Details of the derivation of Eqn. (1) are detailed elsewhere[40]. The key innovation is that is obtained from the multilayer electrostatic problem via a transfer-matrix formalism. This contrasts with previous approaches to excitons in multilayers that also use classical electrostatics for screening but were limited to dipolar interlayer excitons in lattice-matched bilayers and treated within effective-mass frameworks[41, 42, 43, 44, 45].
To illustrate the transfer-matrix method, we consider two 2D layers with arbitrary stacking (Figure 1(a)). We first construct the screened Coulomb potential in momentum space. in layer due to an electron in layer . Along the out-of-plane direction , we express the potential in each region as a sum of exponentials and enforce boundary conditions at each interface by combining propagation (), layer (), and source () matrices. Here, propagates the potential across the spacing between the -th and -th layers, describes the response of the -th layer with polarizability between dielectrics and , and is the electron source (see Figure 1(a)). The transfer matrix de-
Figure 1. Structure and out-of-plane electric-field response of dipolar and quadrupolar excitons. (a), (b): Schematics of dipolar and quadrupolar excitons in WSe/WS and WSe/WS/WSe heterostructures sandwiched between substrates. The top and bottom electrostatic potentials, and , are related by the layer (), propagation (), and electron source matrices. (c),(e): Dipolar excitons exhibit a linear Stark effect, whereas (d),(f) quadrupolar excitons transition from a quadratic to a linear Stark response as the electric field strength increases. The colorbar indicates both optical conductivity and photoluminescence. Optical conductivities are from lattice-matched structures. Experimental data adapted with permission from Yu et al.``{sup}5` (Copyright 2023 Springer Nature).
termines the top potential from the bottom one, . From this we propagate through the layers to find . We repeat this for the second layer, determining and propagating to find . The same approach extends naturally to the three-layer system (Figure 1(b)),
See Sec. A of the Supplementary Information (SI) for details and generalization to arbitrary multilayers. We use Å and Å obtained from DFT[46]. Finally, we apply a Hankel transform to map to the real-space entering the BSE Hamiltonian, Eq. (1).
Quadrupolar excitons in lattice-matched trilayers. WS and WSe have a lattice mismatch of 4%, producing a moiré pattern with about 8 nm lattice constant even in the absence of twist[5, 21, 34, 37]. To disentangle the moiré-related effects, we first construct lattice-matched WSe/WS bilayers and WSe/WS/WSe trilayers with a lattice constant of 0.325 nm. In both systems, WSe and WS are antiparallel (). In the trilayer, we keep the two outer WSe layers parallel (AA stacking, ), as DFT calculations showed this alignment to be necessary for quadrupolar exciton formation[7]. Both systems have type-II band alignment, with the valence band maximum in and the conduction band minimum in , and a direct band gap at the point. Figure 1(c),(Figure 1d) compare the calculated optical conductivity of the two systems under an applied out-of-plane electric field, . To place all results on a common energy reference, the lowest-energy bright excitons are aligned to 1.4 eV at zero field using a rigid scissor shift applied throughout the paper. The bilayer interlayer exciton shows a linear Stark effect over the entire field range, consistent with a permanent dipole. In contrast, the trilayer interlayer exciton shows a quadratic response up to mV/nm, crossing over to a linear regime at higher fields. This quadratic-to-linear crossover is the hallmark of a field-induced transition from quadrupolar to dipolar excitons[24]. Figure 1(c),(Figure 1d) also show higher-energy exciton Rydberg states up to the level. At zero field, the – energy splitting decreases from 82 meV in the bilayer to 68 meV in the trilayer, reflecting enhanced screening in the trilayer. We observe Rydberg excitons beyond the state, but the calculations used here are not sufficiently converged
to accurately determine their energies and oscillator strengths (see SI, Sec. B). Figure 1(e),(Figure 1f) show measured PL spectra for structures where and form a 8 nm moiré pattern and the outer layers are nearly aligned[5]. Our calculations reproduce the experimentally observed lowest bright excitons in PL, validating our approach. We now turn to moiré systems accounting for the lattice mismatch between and .
Twist-angle dependence of quadrupolar excitons. Figure 2 shows the twist-angle dependence of the lowest-energy excitons in under different out-of-plane electric fields. The two layers remain aligned, while the middle layer is twisted by an angle relative to the bottom layer. Regardless of twist angle, the lowest bright exciton exhibits a quadratic-to-linear Stark response as increases from zero. A zoom-in of the lowest excitons near zero field highlighting the quadratic region is also shown in Figure 4(b). At large twist angles (see Figure 2(a)), the peak intensities are weak, but they increase as the twist angle decreases. At small twist angles (see Figure 2(c),(Figure 2d)), we find two quadrupolar excitons, labeled I and II, with exciton II lying within 50 meV of exciton I. These peaks are fundamentally different from the 1s state in the lattice-matched system without a moiré pattern. They correspond to moiré-trapped quadrupolar excitons and are the main result of this work.
Figure 2. Out-of-plane electric-field response of excitons in twisted WSe/WS/WSe. (a–d) As the electric field increases from zero, the lowest bright exciton changes from a quadratic to a linear response, indicating a transition from quadrupolar to dipolar behaviour. Peak I and II in (c) and (d) are due to moiré trapping of quadrupolar excitons. The colorbar indicates the optical conductivities.
Figure 4. Tunable dipole moments and avoided crossings in twisted WSe/WS/WSe. Field-dependent (a) dipole moments of the lower branch and (b) avoided-crossing gaps between the upper and lower branches for several twist angles. The center of the gap is aligned across all twist angles. Hexagon size indicates the oscillator strength of the excitons in (b). Experimental data adapted with permission from Yu et al[5] (Copyright 2023 Springer Nature).
To demonstrate moiré trapping, we analyze the exciton wavefunctions associated with excitons I and II at zero electric field. Figure 3 shows the electron distributions at a 2.2° twist with the hole fixed at different stacking regions. For both excitons, the electron is located in and the holes are in the layers. As shown in Figure 3(a), exciton I has the electron and hole vertically aligned with the electron showing circular symmetry. The electron density is strongest in the stacking region, corresponding to Bernal stacking with Se atoms of both layers directly above W atoms of . In contrast, Figure 3(b) shows exciton II has the electron distributed across three lobes at the vertices of a triangle, displaced laterally from the hole position. The electron density is strongest in the AAA stacking configuration, where the W and Se atoms in are directly on top of the W and S atoms in , respectively. Crucially, exciton II has reduced rotational symmetry, , generating an in-plane octupolar moment absent in exciton I (see SI, Sec. C for analysis). The exciton trapping in different stacking regions, absent in lattice-matched trilayers and at large twist angles, arises from moiré-induced atomic rearrangements. In the energetically unfavorable AAA stacking region, atoms move anticlockwise in and clockwise in WS to reach a more stable configuration (see SI, Sec. D for more details). The symmetric WSe layers on either side of WS amplify these rearrangements in WS. This produces a much stronger confining potential for electrons than in a twisted bilayer system, consistent with previous experiments[23]. As a result, quadrupolar excitons become trapped at approximately below twist, corresponding to moiré lengths nm.
Figure 3. Moiré quadrupolar exciton wavefunctions in a twisted . (a) Exciton I shows azimuthally symmetric electron density and is trapped at the stacking region. (b) Exciton II shows threefold symmetric electron density, laterally displaced from the hole location and largely trapped at the AAA stacking region. Blue dots indicate the hole positions, while solid lines show the moiré unit cell. The colorbar indicates the electron densities.
Two-dipole avoided crossing model. To quantify the electric field-driven quadrupole-to-dipole transition, we employ an avoided crossing model. In this model, two degenerate dipolar excitons with energy and dipole moments are coupled via a quantum tunneling strength under an applied electric field . The system is described by the Hamiltonian[5, 6, 7, 9, 29]
Eqn. (5) has two eigenvalues, , which correspond to two quadrupolar exciton branches. Since experiments have primarily observed the lower-energy branch[7], we extract and by fitting the field-dependent dipole moment from our BSE calculations. This branch corresponds to the highlighted state at large twist angles and lattice-matched systems, and to peak I at small twist angles (Figure 2). We fit using
At strong fields (), the dipole saturates to , while at weak fields (), it vanishes, recovering the quadrupolar character. Figure 4(a) compares our calculated for all moiré structures and the lattice-matched trilayer with experiment[5], showing qualitative agreement. Table II lists the extracted parameters. The tunneling strength increases monotonically as the twist angle decreases from to . The splitting in Table II, obtained from the BSE calculations, is larger than the model prediction of , pointing to corrections beyond the simple two-level model.
| Lattice constant (nm) | Number of atoms | Number of Wannier functions |
| 0.325 () | 9 | 33 |
| 1.15 () | 111 | 407 |
| 3.49 () | 1029 | 3773 |
| 4.89 () | 2013 | 7381 |
| 5.59 () | 2625 | 9625 |
| 0.325 () | 9 | 33 |
Table II. Summary of lattice constants, number of atoms, and number of Wannier functions used in the electronic structure calculations for the BSE.
Another key distinguishing factor between quadrupolar and dipolar excitons is the field-dependent PL radiative rate, [5, 7]. WWe capture this behavior using Fermi’s golden rule[47] together with the computed oscillator strengths, . Figure 4(b) shows the oscillator strengths of both exciton branches, defined as . At , is highest for the lower branch, while at it is highest for the upper branch. The bright quadrupolar exciton near zero has a radiative rate 1.5–1.8 times higher than the dipolar branch at high , while the dark branch is 4–8 times lower, depending on twist angle. Our results agree well with the experimentally reported 1.7-fold faster radiative decay of the bright quadrupolar exciton branch[7].
Many-exciton phases at zero . To demonstrate the impact of moiré trapping, we map the zero-temperature phase diagram of interacting quadrupolar excitons, both free and trapped. We model excitons as point-like bosons and neglect their kinetic energy[30]. Building on the avoided crossing model, the phases are governed by three competing energy scales: electrostatic interactions, the energy cost of occupying the dipolar instead of the quadrupolar state, and the moiré trapping energy . Each dipolar exciton carries an Ising degree of freedom (up or down), which selects the branch of the pair potential,
where is a prefactor, is the effective dielectric constant of the hexagonal boron nitride substrate, is the planar separation of -th dipoles, and is the interlayer distance. The interaction becomes a symmetric average for the lower quadrupolar exciton branch. The Hamiltonian is
| Lattice constant (nm) | (e.nm) | (meV) | (meV) |
| 0.325 (60°) | 0.61 ± 0.02 | 3.8 ± 0.8 | 14.5 |
| 0.325 (0°) | 0.63 ± 0.02 | 1.3 ± 0.4 | 3.8 |
| 1.15 (16.1°) | 0.59 ± 0.02 | 0.7 ± 0.3 | 1.8 |
| 3.49 (4.7°) | 0.66 ± 0.02 | 0.7 ± 0.4 | 2.0 |
| 4.89 (2.8°) | 0.64 ± 0.02 | 1.7 ± 0.5 | 6.5 |
| 5.59 (2.2°) | 0.64 ± 0.02 | 1.9 ± 0.5 | 6.9 |
Table 1. Parameters of the avoided crossing model. We obtain from the Bethe-Salpeter equation and extract and by fitting to the avoided crossing model. The twist angle is defined between neighboring WSe and WS layers.
where the first term sums electrostatic interactions, counts excitons in the dipolar state, and counts excitons not trapped in a moiré.
Figure 5(a) shows the resulting phase diagram for free quadrupolar excitons, computed with , meV, nm, and . We use the dimensionless parameters , and to define the phase diagram, where is the exciton density. Three phases emerge: a triangular lattice of quadrupolar excitons [Figure 5(b)], a square lattice of staggered dipolar excitons [Figure 5(c)], and an inhomogeneous “droplet” phase of dense dipolar clusters without long-range order [Figure 5(d)][30]. Quadrupolar excitons repel via , while antiparallel dipoles attract via . At low , the electrostatic gain from conversion to staggered dipoles is outweighed by , stabilising the triangular quadrupolar lattice. At high , the gain exceeds and the system transitions to the staggered dipolar phase, selecting a square bipartite geometry to avoid the frustration inherent to the triangular lattice [Figure 5(e),(Figure 5g)]. Moiré trapping fundamentally alters this competition by confining excitons to the triangular lattice, expanding the quadrupolar phase boundaries Figure 5(e) with meV]. This resolves an anomaly reported in Ref. 5, where quadrupolar excitons were observed at densities below ().
Figure 5. Many-exciton phase diagram for free and trapped quadrupolar excitons. (a)–(d): Free quadrupolar exciton phases include (b) a triangular lattice of quadrupoles, (c) a square lattice of antiparallel dipoles, and (d) droplet phases lacking long-range order. (e)–(g): Trapped quadrupolar exciton phases include (e) a triangular lattice of quadrupoles, (g) frustrated triangular lattice of anti-parallel dipoles, (f) droplet.
Below such with , untrapped excitons are expected to favour the dipolar droplet phase (Figure 5(a)), making the observation of quadrupolar excitons surprising. Moiré trapping stabilises the quadrupolar phase (Figure 5(e)). While evidence for staggered dipolar excitons has been found at higher densities[5], the in-plane lattice ordering (for instance, triangular versus square) has yet to be resolved experimentally.
In summary, we develop a transfer-matrix formalism coupled with a fully atomistic BSE framework, revealing the non-trivial real-space structure of quadrupolar excitons in van der Waals trilayers. Moiré trapping further establishes these excitons as a platform for exotic many-body physics[48], from frustrated quantum magnetism to spin liquids.
Author contributions
I. M. and J. L. discussed the initial project ideas. I. M. led the project, developed the algorithms, carried out the simulations, and wrote the first draft of the manuscript. A. A. M. and J. L. provided computing resources. All authors discussed the results and contributed to the manuscript.
Ethics declaration
The authors declare no competing interests.
Data availability
All data used to produce the plots in the main text of the paper will be made publicly available at https://github.com/imaitygit/quadrupolar-excitons.
Code availability
Structure construction, atomic relaxations, electronic structure, and exciton calculations presented in this work were performed using publicly available codes. An updated version of the PyMEX code, including the new capabilities described here, is planned for release by the end of 2026. Code for computing many-exciton phases will be made publicly available at https://github.com/imaitygit/quadrupolar-excitons.
Acknowledgements
This work used the ARCHER2 UK National Supercomputing Service (https://www.archer2.ac.uk)[49] through our membership of the UKCP consortium (EP/X035891/1) and the HEC Materials Chemistry Consortium (EP/X035859), both funded by EPSRC. We thank Sufei Shi for valuable comments on the manuscript.
Methods
Atomic structure. Twisted bilayer WSe/WS structures were generated using the Twister package[50]. In the trilayer system, the top and bottom WSe layers are aligned in parallel to form a symmetric moiré superlattice. Atomic relaxations were performed using classical interatomic potentials in Lammps[51, 52]. Interactions within each layer were modeled using the Stillinger–Weber potential[53]. Interactions between neighboring layers were modeled using the Kolmogorov–Crespi potential[54], as these dominate over contributions from next-nearest-neighbor layers[55].
Electronic structure. The Siesta package was employed for electronic structure calculations using localized atomic orbitals as basis functions[56]. Norm-conserving Troullier–Martins pseudopotentials[57] and the local density approximation were used for the exchange-correlation energy[58]. Wavefunctions were expanded in a double- plus polarization basis. All moiré calculations used -point sampling, while lattice-matched system calculations employed a -point grid for the charge density, a plane-wave energy cutoff of 100 Ry, an energy shift of 0.02 Ry, and a vacuum spacing of 20 Å in the out-of-plane direction. Spin-orbit coupling[59] was excluded during Wannier function generation but was included to extract the – ordering of bands.
Wannier functions were generated via the one-shot projection method[60] as implemented in the Wannier90 code[61], using and orbitals as initial projections. Specifically, Kohn-Sham wavefunctions were projected onto atom-centered (, , , , ) orbitals on every W atom and (, , ) orbitals on every S and Se atom, followed by orthogonalization and disentanglement[62]. We used minimal-distance replica selection for Fourier interpolation of the Wannier Hamiltonian[35, 61].
Spin-orbit coupling (SOC) was excluded from Wannier function generation, which caused -derived bands of the pristine unit cell to fold above -derived bands in the moiré Brillouin zone, incorrectly ordering the and transition energies. We corrected this by applying a rigid shift to -derived valence bands identified at each -point via the orbital-weight score
where are the Wannier projection coefficients for orbital on atom , summed over all atoms per layer. Bands with were shifted by a rigid determined by explicit comparison to SOC-included calculations, restoring the correct – ordering. This is physically motivated by the well-known orbital composition of 2D transition metal dichalcogenides valence bands: states are dominated by and , while states are dominated by [63]. We implemented this procedure in PyMEX.
Excitonic structure. We implemented the transfer matrix approach described above in PyMEX[40] for arbitrarily stacked multi-layer structures. We employed PyMEX to construct and diagonalise the BSE Hamiltonian and compute absorption spectra using Wannier functions as basis, with diagonalisation accelerated by the Eigenvalue soLvers for Petaflop Applications (ELPA) library[64, 65]. We applied a vertical electric field by adding the term to the electronic Hamiltonian, where is the -coordinate of the -th Wannier function and is the electron charge. At each twist angle, we performed BSE calculations at 25 electric-field values and interpolated the results to construct a heatmap of the field-dependent absorption spectra. We used -point grids of , , , , and , and included 3, 6, 20, 24, and 40 valence bands and 3, 6, 10, 8, and 16 conduction bands in the BSE Hamiltonian for the lattice-matched, , , , and WSe/WS/WSe trilayers, respectively. A full description of PyMEX, including implementation details, documentation, and an open-source release, is beyond the scope of this work and will appear in a subsequent paper. Spin-orbit coupling is omitted from our BSE Hamiltonian but can be included perturbatively[40]. It splits the quadrupolar 1s-like exciton into singlet and triplet states separated by meV, with the singlet being optically bright[29]. All results presented in the main text apply to both branches independently.
Exciton phase diagram. Lattice sums were evaluated over all integer pairs with , excluding the origin, on both square and triangular lattices, with lattice constants set by the site density via and respectively. See SI, Sec. E for additional details. The dipolar phases carry an additional energy cost per exciton (the quadrupole-to-dipole gap), and the dipolar square lattice incurs a further moiré penalty per site for being incommensurate with the triangular moiré lattice. Below their respective optimal densities , each dipolar phase undergoes a droplet construction: excitons phase-separate into dense droplets at surrounded by empty regions, with the interaction energy fixed to its minimum value. Since moiré potentials can exceed 100 meV[66, 67], the value meV used here is conservative, and our conclusions remain robust.
Supplementary Information to “Moiré trapping of quadrupolar excitons in van der Waals trilayers”
Methods
Atomic structure and Wannier functions
Multilayer screening using the transfer matrix method
Two stacked 2D materials
Outside the stacked 2D materials, there are three regions: I (), II (), and III (). We take . The electrostatic potential admits the following solutions in each regions
We define a coefficient vector \psi_z = \begin{pmatrix} \text{coeff. of }e^{-qz}\text{ evaluated at }z \ \text{coeff. of }e^{qz}\text{ evaluated at }z \end{pmatrix}.
Propagation matrix: Inside a homogeneous dielectric, the coefficients of the exponential basis evolve independently. Starting from
an out-of-plane shift by a distance gives
This evolution can be captured compactly by a propagation matrix,
where
Layer matrix: The layer matrix is 2D polarizable, and the boundary conditions on and need to be adjusted suitably. The setup consists of a 2D material with polarizability , and dielectric constants and above and below it, respectively. Potenital continuity condition (i.e., ) leads to
On the other hand, the discontinuous jump in the normal component of the electric field (i.e., ) leads to
In the above, we did not include source charge (i.e., ). After a bit of algebra, this simplifies to
Both Equations 13 and 15 must be satisfied simultaneously. Using the matrix formulation, both equations can be solved efficiently.
We can extract , where
Source matrix: Note that the source charge is not included in the above. The inclusion of source charge modifies Eqn. (17)
This leads to different set of equations to solve
The source matrix is given below, applies only at the source layer, and determines the propagated solutions thereafter.
We use the transfer matrix method to calculate the response of a multilayer system. Consider the bilayer 2D material setup shown in Figure 6. In the bottom region (), there is a dielectric with constant . Layer 1 is modeled as an ideal 2D layer at with polarizability . Between Layers 1 and 2 (), there is vacuum with . Layer 2 is an ideal 2D layer at with polarizability . In the top region (), a dielectric with constant is present.
Figure 6. Schematics of dipolar and quadrupolar excitons.
Case I: Source charge in layer 1
The transfer matrix for the electrostatic potential, expressed in terms of the state vector , can be written as
We obtain and propagate the solutions to extract , , .
Case II: Source charge in layer 2
Again, is computed first, and the solutions are then propagated to determine the potentials in the other layer.
Sanity check and comparison with other works:
In the limit where both sides are vacuum (i.e., ), our results are consistent with those of Danovich and co-workers[42]. We have used .
Three stacked 2D materials
There are four regions of interest: region I (), region II (), region III (), and region IV (). Figure 6 shows the setup. The electrostatic potential in these regions admits solutions of the form
Three 2D materials have polarizabilities , , and . Using the same approach as for the bilayer, we construct the electrostatic potential with the transfer matrix method. The source can be located in any of the three layers.
Case I: Electron in layer 1
We extract . As a sanity check, we verify that the left-hand side and right-hand side match after obtaining the solution. We propagate the solution, to obtain , , .
Case II: Electron in layer 2
We extract . As a sanity check, we verify that the left-hand side and right-hand side match after obtaining the solution. Very similar to above, we propagate the solutions.
Case III: Electron in layer 3
We extract . Very similar to above, we propagate the solutions.
Explicit expressions are given below. Even with three layers the equations quickly become very complicated.
with
Sanity checks:
Symmtery: . This is consistent with Green’s reciprocity theorem.
Symmetric trilayer limit: When , , , we find , , with , , unchanged.
Exponential decay of Off-diagonal elements: , , , , and , .
Bilayer limit: i.e.,, and examine the potentials to check consistency with bilayer calculations.
These are consistent with the bilayer results.
n stacked 2D materials
We follow the same strategy to solve the n-layer system. For example, when the source charge is in the bottom layer, we solve
We extract . As a sanity check, we verify that the left-hand side and right-hand side match after obtaining the solution. We propagate the solution, to obtain . And very similar to bilayer and trilayer system, we compute where .
Monolayer 2D material
A special case of the transfer matrix method is the monolayer Keldysh potential. Using the same strategy as above, we set . The potential becomes
This is the well-known Rytova-Keldysh potential in momentum space[68, 69, 70].
K-point convergence of dipolar and quadrupolar excitons
Figure 7. Convergence of the lowest-energy excitons in lattice-matched WSe/WS and WSe/WS/WSe van der Waals heterostructures with a lattice constant of 0.325 nm. We find that the 1s dipolar and quadrupolar excitons converge with a 45 45 -point grid, while the 2s excitons require a 78 78 grid for convergence. We do not apply any rigid energy shifts. In constructing the BSE Hamiltonian, we include 2 valence and 2 conduction bands for the bilayer, and 3 valence and 3 conduction bands for the trilayer. We find that the oscillator strength of both dipolar and quadrupolar excitons is times weaker than that of intralayer excitons in WSe or WS. The inset shows a plot zoomed in by a factor of 500.
Electrostatic potential for excitons I and II
Exciton I is the collinear arrangement , , along . Exciton II replaces the central by three charges on an equilateral triangle of side in the plane. For exciton I, with and cylindrical coordinates , superposing the Coulomb potentials gives
We characterize both distributions through their multipole moments. The Cartesian quadrupole tensor, with labeling the point charges, is defined as
and is traceless by construction (). Equivalently, the moments are organized by the spherical multipole expansion,
with the multipole order and the azimuthal index (the quadrupole is , the octupole ), and , where are the associated Legendre polynomials.
Exciton I. Measuring positions from the central at (), the lower moments vanish:
The charges lie on the axis, so , and tracelessness yields
so . Thus, , and . Also, note that .[71]
Exciton II. Centering at the origin (the midpoint of the hole charges and the centroid of the triangle), the in-plane charges sit at circumradius ,
The monopole and dipole again vanish:
The positive charges each contribute to , while each in-plane charge contributes , so
giving and again . As a result, remains unchanged.
Octupole
The two distributions first differ at the octupole (). For exciton I, all charges lie on the axis (), where for ; the only possible component again vanishes,
as enforced by the inversion center, which forbids all odd- moments. For exciton II, the positive charges again give no octupole, while the triangle (, , ) gives
The azimuthal sum imposes a threefold selection rule, , leaving ; since and , only
remains. Exciton II thus carries a threefold in-plane octupole, absent in exciton I.
Structural relaxations in a twisted trilayer
Figure 8. Large atomic relaxations in a twisted trilayer. Arrows show the displacement of W atoms in each layer relative to the rigid, unrelaxed structure. Displacements of W atoms in the middle layer are amplified.
Electrostatic energies of dipolar and quadrupolar excitons on different lattice geometries

Figure 9. We consider three electrostatic energies: , quadrupolar excitons on a triangular lattice; (square), staggered dipolar excitons on a square lattice without frustration; and (triangular), staggered dipolar excitons on a triangular lattice with minimal frustration.
References
References
- [1]A. D. Buckingham, “Permanent and induced molecular moments and long-range intermolecular forces,” in Advances in Chemical Physics (John Wiley & Sons, Ltd, 1967) pp. 107–142.
- [2]K. S. Thorne, Rev. Mod. Phys. 52, 299 (1980).
- [3]M.-T. Suzuki, H. Ikeda, and P. M. Oppeneer, Journal of the Physical Society of Japan 87, 041008 (2018).
- [4]P. Santini, S. Carretta, G. Amoretti, R. Caciuffo, N. Magnani, and G. H. Lander, Rev. Mod. Phys. 81, 807 (2009).
- [5]L. Yu, K. Pistunova, J. Hu, K. Watanabe, T. Taniguchi, and T. F. Heinz, Nature Materials 22, 1485 (2023).
- [6]Z. Lian, D. Chen, L. Ma, Y. Meng, Y. Su, L. Yan, X. Huang, Q. Wu, X. Chen, M. Blei, T. Taniguchi, K. Watanabe, S. Tongay, C. Zhang, Y.-T. Cui, and S.-F. Shi, Nature Communications 14, 4604 (2023).
- [7]W. Li, Z. Hadjri, L. M. Devenica, J. Zhang, S. Liu, J. Hone, K. Watanabe, T. Taniguchi, A. Rubio, and A. Srivastava, Nature Materials 22, 1478 (2023).
- [8]Y. Bai, Y. Li, S. Liu, Y. Guo, J. Pack, J. Wang, C. R. Dean, J. Hone, and X. Zhu, Nano Letters 23, 11621 (2023).
- [9]Y. Xie, Y. Gao, F. Chen, Y. Wang, J. Mao, Q. Liu, S. Chu, H. Yang, Y. Ye, Q. Gong, J. Feng, and Y. Gao, Phys. Rev. Lett. 131, 186901 (2023).
- [10]M. Chen, R. Li, H. Wang, Y. Yang, Y. Lai, C. Hu, T. Taniguchi, K. Watanabe, J. Yan, J.-H. Chu, E. Henriksen, C. Zhang, L. Yang, and X. Wang, Nature Communications 16, 10359 (2025).
- [11]J. Jasiński, J. Hagel, S. Brem, E. Wietek, T. Taniguchi, K. Watanabe, A. Chernikov, N. Bruyant, M. Dyksik, A. Surrente, M. Baranowski, D. K. Maude, E. Malic, and P. Plochocka, Nature Communications 16, 1382 (2025).
- [12]Y. Xie, F. Chen, Y. Gao, Y. Wang, J. Mao, Q. Liu, S. Chu, H. Yang, Y. Ye, Q. Gong, J. Feng, and Y. Gao, Phys. Rev. B 110, L201402 (2024).
- [13]S. Feng, A. J. Campbell, B. M. Francis, H. Baek, T. Taniguchi, K. Watanabe, I. C. Gerber, B. D. Gerardot, and M. Brotons-Gisbert, Phys. Rev. Lett. 134, 166901 (2025).
- [14]Z. Zhang and S. Wang, The Journal of Physical Chemistry Letters 15, 10104 (2024).
- [15]J. Sung, Y. Zhou, G. Scuri, V. Zólyomi, T. I. Andersen, H. Yoo, D. S. Wild, A. Y. Joe, R. J. Gelly, H. Heo, S. J. Magorrian, D. Bérubé, A. M. M. Valdivia, T. Taniguchi, K. Watanabe, M. D. Lukin, P. Kim, V. I. Fal’ko, and H. Park, Nature Nanotechnology 15, 750 (2020).
- [16]Z. Sun, H. Sun, X. Jia, A. Li, N. J. Zhang, K. Seungmin Hong, J. DePinho, C. Y. Long, K. Watanabe, T. Taniguchi, O. Chen, J. Wang, J. Li, B. Rubenstein, and Y. Bai, arXiv e-prints , arXiv:2602.00419 (2026), arXiv:2602.00419 [cond-mat.mtrl-sci].arxiv.org/abs/2602.00419
- [17]S. Gim, H. Jo, S. Kim, S. M. Yang, K. Watanabe, T. Taniguchi, and H. Baek, ACS Photonics 12, 6957 (2025).
- [18]J. Zhu, T. Liang, F. Shen, Z. Chen, and J. Xu, Nature Communications 16, 10661 (2025).
- [19]B. Xu, J. Fu, L. Huang, C. Ge, Z. Xu, W. Zheng, Q. Deng, S. Xie, Q. Tong, D. Li, Y. Jiang, H. Yuan, and A. Pan, Nature Communications 16, 3938 (2025).
- [20]Z.-H. Guo, T. Yan, J.-Z. Zhao, Y.-J. Jin, and Q. Zhu, arXiv preprint arXiv:2504.10189 (2025).arxiv.org/abs/2504.10189
- [21]Y. Meng, L. Ma, L. Yan, A. Khalifa, D. Chen, S. Zhang, R. Banerjee, T. Taniguchi, K. Watanabe, S. A. Tongay, B. Hunt, S.-Z. Lin, W. Yao, Y.-T. Cui, S. Chatterjee, and S.-F. Shi, Nature Photonics 19, 1219 (2025).
- [22]H. Liu, S. Chen, H. Xu, R. Han, Z. Sun, S. Huang, X. Zhang, J. Luo, and D. Liu, Phys. Rev. Lett. 135, 206302 (2025).
- [23]H. Zheng, B. Wu, S. Li, J. Ding, J. He, Z. Liu, C.-T. Wang, J.-T. Wang, A. Pan, and Y. Liu, Light: Science & Applications 12, 117 (2023).
- [24]E. Barré, M. Dandu, and A. Raja, Nature Materials 22, 1437 (2023).
- [25]X. Blase, I. Duchemin, D. Jacquemin, and P.-F. Loos, The Journal of Physical Chemistry Letters 11, 7371 (2020).
- [26]M. Rohlfing and S. G. Louie, Phys. Rev. B 62, 4927 (2000).
- [27]G. Onida, L. Reining, and A. Rubio, Rev. Mod. Phys. 74, 601 (2002).
- [28]S. G. Louie, Y.-H. Chan, F. H. da Jornada, Z. Li, and D. Y. Qiu, Nature Materials 20, 728 (2021).
- [29]T. Deilmann and K. Sommer Thygesen, 2D Materials 11, 035032 (2024).
- [30]Y. Slobodkin, Y. Mazuz-Harpaz, S. Refaely-Abramson, S. Gazit, H. Steinberg, and R. Rapaport, Phys. Rev. Lett. 125, 255301 (2020).
- [31]Z. Zhang and S. Wang, The Journal of Physical Chemistry C 129, 15843 (2025).
- [32]L.-A. Qin, Y. Wang, Y. Xu, D. Sun, Y. Dai, B. Huang, and W. Wei, Nano Letters 25, 16123 (2025).
- [33]M. H. Naik, E. C. Regan, Z. Zhang, Y.-H. Chan, Z. Li, D. Wang, Y. Yoon, C. S. Ong, W. Zhao, S. Zhao, M. I. B. Utama, B. Gao, X. Wei, M. Sayyad, K. Yumigeta, K. Watanabe, T. Taniguchi, S. Tongay, F. H. da Jornada, F. Wang, and S. G. Louie, Nature 609, 52 (2022).
- [34]S. Susarla, M. H. Naik, D. D. Blach, J. Zipfel, T. Taniguchi, K. Watanabe, L. Huang, R. Ramesh, F. H. da Jornada, S. G. Louie, P. Ercius, and A. Raja, Science 378, 1235 (2022).
- [35]I. Maity, J. Lischner, A. A. Mostofi, and A. Rubio, Nano Letters 26, 1349 (2026).
- [36]R. Dalal, H. Singh, R. Dutta, H. Swaminathan, K. Watanabe, T. Taniguchi, M. H. Naik, M. Jain, and A. Singh, arXiv preprint arXiv:2601.03045 (2026).arxiv.org/abs/2601.03045
- [37]C. Jin, E. C. Regan, A. Yan, M. Iqbal Bakti Utama, D. Wang, S. Zhao, Y. Qin, S. Yang, Z. Zheng, S. Shi, K. Watanabe, T. Taniguchi, S. Tongay, A. Zettl, and F. Wang, Nature 567, 76 (2019).
- [38]R. Wang, K. Chang, W. Duan, Y. Xu, and P. Tang, Phys. Rev. Lett. 134, 026904 (2025).
- [39]G. E. Astrakharchik, I. L. Kurbakov, D. V. Sychev, A. K. Fedorov, and Y. E. Lozovik, Phys. Rev. B 103, L140101 (2021).
- [40]I. Maity, A. A. Mostofi, and J. Lischner, npj 2D Materials and Applications 9, 20 (2025).
- [41]L. S. R. Cavalcante, A. Chaves, B. Van Duppen, F. M. Peeters, and D. R. Reichman, Phys. Rev. B 97, 125427 (2018).
- [42]M. Danovich, D. A. Ruiz-Tijerina, R. J. Hunt, M. Szyniszewski, N. D. Drummond, and V. I. Fal’ko, Phys. Rev. B 97, 195452 (2018).
- [43]S. Ovesen, S. Brem, C. Linderälv, M. Kuisma, T. Korn, P. Erhart, M. Selig, and E. Malic, Communications Physics 2, 23 (2019).
- [44]M. N. Brunetti, O. L. Berman, and R. Y. Kezerashvili, Phys. Rev. B 98, 125406 (2018).
- [45]M. Van der Donck and F. M. Peeters, Phys. Rev. B 98, 115104 (2018).
- [46]T. C. Berkelbach, M. S. Hybertsen, and D. R. Reichman, Phys. Rev. B 88, 045318 (2013).
- [47]M. Palummo, M. Bernardi, and J. C. Grossman, Nano Letters 15, 2794 (2015).
- [48]R. Moessner and A. P. Ramirez, Physics Today 59, 24 (2006).
- [49]G. Beckett, J. Beech-Brandt, K. Leach, Z. Payne, A. Simpson, L. Smith, A. Turner, and A. Whiting, “ARCHER2 service description,” (2024).DOI
- [50]S. Naik, M. H. Naik, I. Maity, and M. Jain, Computer Physics Communications 271, 108184 (2022).
- [51]“Lammps,” Accessed: December 16, 2025.
- [52]A. P. Thompson, H. M. Aktulga, R. Berger, D. S. Bolintineanu, W. M. Brown, P. S. Crozier, P. J. in ’t Veld, A. Kohlmeyer, S. G. Moore, T. D. Nguyen, R. Shan, M. J. Stevens, J. Tranchida, C. Trott, and S. J. Plimpton, Comp. Phys. Comm. 271, 108171 (2022).
- [53]J.-W. Jiang and Y.-P. Zhou, Handbook of Stillinger-Weber Potential Parameters for Two-Dimensional Atomic Crystals (IntechOpen, Rijeka, 2017).DOI
- [54]M. H. Naik, I. Maity, P. K. Maiti, and M. Jain, The Journal of Physical Chemistry C 123, 9770 (2019).
- [55]I. Maity, P. K. Maiti, and M. Jain, Phys. Rev. B 97, 161406 (2018).
- [56]J. M. Soler, E. Artacho, J. D. Gale, A. García, J. Junquera, P. Ordejón, and D. Sánchez-Portal, Journal of Physics: Condensed Matter 14, 2745 (2002).
- [57]N. Troullier and J. L. Martins, Phys. Rev. B 43, 1993 (1991).
- [58]J. P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981).
- [59]Fernandez-Seivane, L. and Oliveira, M. A. and Sanvito, S. and Ferrer, J., Journal of Physics: Condensed Matter 18 (2006), 10.1088/0953-8984/19/19/489001.
- [60]N. Marzari, A. A. Mostofi, J. R. Yates, I. Souza, and D. Vanderbilt, Rev. Mod. Phys. 84, 1419 (2012).
- [61]G. Pizzi, V. Vitale, R. Arita, S. Blügel, F. Freimuth, G. Géranton, M. Gibertini, D. Gresch, C. Johnson, T. Koretsune, J. Ibañez-Azpiroz, H. Lee, J.-M. Lihm, D. Marchand, A. Marrazzo, Y. Mokrousov, J. I. Mustafa, Y. Nohara, Y. Nomura, L. Paulatto, S. Poncé, T. Ponweiser, J. Qiao, F. Thöle, S. S. Tsirkin, M. Wierzbowska, N. Marzari, D. Vanderbilt, I. Souza, A. A. Mostofi, and J. R. Yates, Journal of Physics: Condensed Matter 32, 165902 (2020).
- [62]I. Souza, N. Marzari, and D. Vanderbilt, Phys. Rev. B 65, 035109 (2001).
- [63]A. Kormányos, G. Burkard, M. Gmitra, J. Fabian, V. Zólyomi, N. D. Drummond, and V. Fal'ko, 2D Materials 2, 022001 (2015).
- [64]T. Auckenthaler, V. Blum, H.-J. Bungartz, T. Huckle, R. Johanni, L. Krämer, B. Lang, H. Lederer, and P. Willems, Parallel Computing 37, 783 (2011), 6th International Workshop on Parallel Matrix Algorithms and Applications (PMAA’10).
- [65]A. Marek, V. Blum, R. Johanni, V. Havu, B. Lang, T. Auckenthaler, A. Heinecke, H.-J. Bungartz, and H. Lederer, Journal of Physics: Condensed Matter 26, 213201 (2014).
- [66]I. Maity, A. A. Mostofi, and J. Lischner, Nano Letters 23, 4870 (2023).
- [67]A. Rossi, J. Zipfel, I. Maity, M. Lorenzon, M. Dandu, E. Barré, L. Francaviglia, E. C. Regan, Z. Zhang, J. H. Nie, E. S. Barnard, K. Watanabe, T. Taniguchi, E. Rotenberg, F. Wang, J. Lischner, A. Raja, and A. Weber-Bargioni, ACS Nano 18, 18202 (2024).
- [68]N. S. Rytova, arXiv preprint arXiv:1806.00976 (2018).arxiv.org/abs/1806.00976
- [69]L. V. Keldysh, Soviet Journal of Experimental and Theoretical Physics Letters 29, 658 (1979).
- [70]P. Cudazzo, I. V. Tokatly, and A. Rubio, Phys. Rev. B 84, 085406 (2011).
- [71]J. D. Jackson, Classical Electrodynamics (Wiley, 1998).