A flatband is a state in which energy stops depending on momentum. In an ordinary band, a particle’s energy rises with its momentum, and the slope of that relation is its velocity. Flatten the band and the velocity goes to zero: kinetic energy is quenched, and the density of states piles up at a single energy. That pile-up is what makes flatbands interesting: interactions become the dominant physics. The difficulty has always been making a flatband that is both robust and reachable. A paper published in Nature Materials on October 29, 2026, reports a route that is both: Connor Heimig, Thomas Weber, and colleagues in Hamburg, Munich, and Milan fabricate a metasurface directly from bulk rhenium disulfide, and show that the material’s own in-plane anisotropy does the topological work — splitting a polarization vortex into halves, flattening the photonic dispersion along one direction only, and carrying that flattened dispersion into strongly coupled exciton–polaritons at room temperature.
What a bound state in the continuum is
The resonant object at the centre of the work is the bound state in the continuum, or BIC. A photonic resonator supports modes that leak radiation into the surrounding space. A BIC is a mode that does not leak at all, even though its energy lies inside the continuum of radiating states the structure can support. Symmetry enforces the cancellation: the mode’s field pattern is orthogonal to every outgoing channel. In principle its quality factor is infinite.
A true BIC is invisible by the same token, so the experimental object is the quasi-BIC, or qBIC: a small perturbation breaks the protecting symmetry, giving the mode a finite linewidth and making it radiatively accessible while retaining a very narrow spectral feature. The perturbation here is geometric. An asymmetry parameter α reshapes each square resonator into a rectangular prism with two different side lengths, keeping the footprint constant while enlarging the unit cell by √2. The enlarged cell folds the Brillouin zone, and the dark degenerate modes at the original Γ point become two radiative qBICs, mode 1 and mode 2, with orthogonal polarizations and identical field profiles rotated by 90 degrees. The metasurface’s parameters are a period of 465 nm, a base side length of 220 nm, α = 0.6, and a resonator height of 90 nm, on a silicon dioxide substrate.
In a C₄-symmetric metasurface made of an isotropic material, those two qBICs would be degenerate at the Γ point: same frequency, and the far-field polarization there undefined, since any linear combination of the two modes is an equally valid eigenstate. That degeneracy is the starting point for everything that follows — enforced by the C₄ shape, and broken by something else entirely.
Why ReS₂ breaks the degeneracy
ReS₂ is a group-VII transition metal dichalcogenide, anisotropic in a way the familiar group-VI dichalcogenides — MoS₂, WS₂, WSe₂ — are not. Those materials are optically isotropic in plane. ReS₂ has a distorted triclinic 1T′ structure: its rhenium atoms form chains, and the crystal has a well-defined in-plane axis, the b axis, along which it cleaves. The in-plane permittivity tensor has two unequal components, εₓₓ and εᵧᵧ, aligned with the crystal axes, with multiple linearly polarized excitonic resonances with axis-aligned transition dipoles. Unusually, its excitons and direct bandgap are nearly independent of layer thickness, so the anisotropy persists from monolayer to bulk — which is what makes the device possible: the metasurface is carved from a flake roughly 90 nm thick, not a monolayer.
The experiment aligns the metasurface coordinate system deterministically with the crystal b axis, identified from the elongated edges of exfoliated flakes, which cleave along b. Each qBIC then couples to a distinct component of the anisotropic permittivity tensor. In the paper’s simplified non-dispersive model (ε = 18, Δε = εₓₓ − εᵧᵧ ≈ 1.7, equivalent to Δn ≈ 0.2, ε_zz = 7.25), the degeneracy lifts as soon as Δε is non-zero: mode 1 redshifts, mode 2 blueshifts, and the splitting scales linearly with Δε. Resonant-state expansion reproduces the splitting analytically and matches the full simulations closely. The shape of the metasurface is C₄-symmetric, but once the resonators are made of ReS₂, the anisotropic permittivity tensor reduces the symmetry of the full system to C₂. The geometry promises degeneracy; the material revokes it.
The topological split
The degeneracy at the Γ point is not only a spectral fact. It is a topological one. In a qBIC metasurface, the topology lives in the far-field polarization: moving through momentum space, the radiated polarization rotates, winding by a full turn around certain singular points. The winding number of that vortex is the topological charge. A true BIC sits at a V point, a singularity where the polarization winds and the radiation vanishes entirely. A mode degeneracy of the kind the C₄ metasurface hosts at Γ is a second kind of singularity, where the polarization cannot be uniquely defined at all.
In the isotropic case, both modes carry an integer winding number of −1 at the Γ point. Introduce the small anisotropy matching ReS₂, and that single integer charge splits into a pair of half-integer charges of −1/2, symmetrically displaced along the kₓ axis. The total charge is conserved — −1 becomes −1/2 plus −1/2 — but it is now carried by two singularities at finite momentum rather than one at the centre. The fractional singularities coincide with points where the two modes touch, forming photonic Dirac cones — two conical bands meeting at a single point, as in graphene. The resonant-state expansion fixes the Dirac point’s location analytically, at kₓ/k₀ = ±√(β/χ), with β proportional to Δε.
Hyperspectral Stokes-parameter measurements in momentum space verified this, mapping the transmitted light’s polarization across the band structure. The measured spectra show two branches intersecting in a Dirac-like crossing at kₓ/k₀ ≈ ±0.35, with singular points in the polarization-angle maps coinciding with the Dirac points. Around a selected Dirac point, the far-field polarization vectors wind by a half-integer, a topological charge of −1/2, as the simulations predict. At these Dirac points only the eigenfrequencies coalesce — unlike exceptional points, where the eigenvectors coalesce too. Rotating the in-plane anisotropy axis revolves the Dirac points through momentum space, with no further geometrical perturbation.
What a directional flatband means, physically
The same anisotropy that splits the charge reshapes the band curvatures. In the isotropic case, mode 1 disperses parabolically from Γ while mode 2 is nearly flat along the high-symmetry directions. With Δn = 0.2, the picture becomes directional: in the k_y = 0 plane, the x-polarized mode 1 develops a flatband while the y-polarized mode 2 becomes strongly curved — and the roles reverse in the k_x = 0 plane. Each mode is nearly dispersionless along one momentum direction and parabolic along the orthogonal one. Rotate the excitation polarization by 90 degrees and the flat and parabolic directions interchange — confirmed in the measured angle-resolved transmittance spectra at room temperature.
Fitting the measured dispersions with a coupled-mode model that accounts for hybridization between the intrinsic qBIC and a substrate-induced grating mode, the team extracted curvature and group velocity. The flat branch’s curvature is more than two orders of magnitude lower than the parabolic branch’s: curvature ratios of 102.6 ± 7.0 for mode 1 and 108.1 ± 8.9 for mode 2, with correspondingly suppressed group velocities near the Γ point. The flatbands extend across 93.7 percent of the accessible k-space region for mode 1 and 91.5 percent for mode 2, limited only by diffraction. In photonic-crystal slabs, group-velocity suppression is typically confined to narrow band-edge regions; here it spans nearly the whole accessible momentum range.
Two qualifications belong in the same paragraph as the numbers. First, these are directional flatbands, not two-dimensional ones: the dispersion is quenched along a single momentum direction only. Second, the flat branches do not have exactly vanishing curvature. Hybridization with the grating channel leaves a residual curvature at the Γ point, so the measured flatbands are strongly suppressed rather than exactly flat. Any account that calls these two-dimensional flatbands or exactly flat bands is overclaiming.
Physically, a directional flatband means light in this mode has almost no kinetic energy along one direction: its group velocity along that axis is near zero, so a wave packet launched in the flat direction barely propagates, while along the perpendicular direction it moves normally. The density of states piles up at the flatband energy, which is what makes the band attractive for nonlinear optics and for condensation. But the directionality is the point: the flatness is a property of a direction, chosen by polarization, not of the whole band.
From photonic to polaritonic
The final step transfers the engineered dispersion into the polaritonic regime. Bulk ReS₂ has two energetically separated excitonic transitions, exciton 1 and exciton 2, polarized along nearly orthogonal in-plane directions. Scaling the metasurface (S = 0.95 for the strongly coupled samples) tuned the qBIC eigenfrequencies into resonance with these excitonic transitions, coupling each qBIC to its corresponding exciton. Because the metasurface is fabricated directly from the excitonic material, the coupling is self-hybridized: no separate emitter layer to align.
At normal incidence, the transmittance spectra for the two excitation polarizations show clear splitting from the hybridization. Coupled-mode fits yield Rabi splittings of ℏΩ_R1 = 110 meV and ℏΩ_R2 = 108 meV, with coupling-strength and linewidth analysis confirming the strong-coupling regime. The Hopfield coefficients give the lower polariton bands as 43 percent exciton 1 and 57 percent mode 1, and 51 percent exciton 2 and 49 percent mode 2, with the ratios reversed for the upper bands. Along the flatband direction, those coefficients remain nearly constant over the accessible momentum range: the polariton’s mixed character does not drift with momentum. The angle-resolved measurements confirm the polaritonic branches inherit the photonic modes’ anisotropic dispersion: topology, flatness, and strong coupling are one dispersion landscape, carried into the polaritons by the hybridization.
Why this route is different
Frustrated lattices — kagome, Lieb — and strained honeycomb polariton lattices produce flatbands through the interference of hopping between lattice sites, which requires the lattice geometry to stay right under fabrication. Superlattice and BIC-merging approaches need finely tuned interference between resonances. All of them are sensitive to disorder: small errors accumulate, and the flatband degrades. The anisotropy mechanism uses a single resonance with no auxiliary quasiparticles. The flatband follows from the topological rearrangement the anisotropy forces, which the paper claims makes it robust and spectrally scalable — the same mechanism should work at other wavelengths by scaling the structure. The authors name black phosphorus and two-dimensional magnets as candidates. Those are proposals, not demonstrations.
What it is not
Three cautions, stated plainly because the framing invites overreading. First, this is not a polariton laser, a quantum logic gate, or a nonlinear device. The paper points toward polarization-resolved quantum emitters, flatband-enhanced nonlinear optics, and low-threshold polariton devices as target functionalities. Those are directions the platform could enable. Nothing in the paper demonstrates lasing, condensation, logic, or a nonlinear threshold — the demonstration is the dispersion landscape itself, measured at room temperature.
Second, this is not a demonstration of a topological device in the sense of protected transport. The topology here is the topology of the far-field polarization — winding numbers of polarization vortices in momentum space — not the edge-state topology of a topological insulator: the paper claims no protected edge mode and no nonreciprocity. The directional flatbands are direction-selective, not one-way: the flat direction is chosen by polarization. The title of this article says the flatbands “only run one way”; precisely, they only run flat in one direction.
Third, the strong coupling is demonstrated at the level of the dispersion and the normal-incidence splitting. The Rabi splittings of 110 and 108 meV are large, and the Hopfield analysis shows genuinely mixed branches. But strong coupling is a regime, not a device metric: the paper reports no polariton lifetime, coherence time, or condensation threshold. Whether these flatband polaritons condense is the open question the platform is built to ask — not a result of this paper.
What to watch
The most immediate follow-up is the one the authors point to: polarization-selective polariton lasing. A flatband polariton combines quenched kinetic energy, qBIC-enhanced photon lifetimes, and room-temperature excitons — what a low-threshold laser wants. The test: does a lasing threshold appear, does it beat unstructured ReS₂, and does the emission inherit the flatband’s polarization selectivity and directionality?
A second follow-up is nonlinear optics. Slow light enhances nonlinear interactions, and here the group-velocity suppression extends over more than 90 percent of the far-field accessible momentum range. The experiment to watch is a power-dependent measurement: does the nonlinear response grow relative to the parabolic direction, over the wide angular range the flatband spans?
A third is the generalization claim: black phosphorus is the obvious next material, two-dimensional magnets the more ambitious one.
The deeper reason this result matters: anisotropy, long treated as a complication to be managed, turns out to be a topological control knob. A C₄-symmetric shape plus an anisotropic material gives you a split charge, a Dirac point, and a flatband — from a single resonance, in the far field, at room temperature. The material was doing the work all along.
Sources and further reading
Heimig, C.; Weber, T.; Cruciano, C.; Genco, A.; Possmayer, T.; Sortino, L.; Valentini, G.; Manzoni, C.; Gorkunov, M. V.; Cerullo, G.; Antonov, A. A.; Tittl, A. “Topological control of polaritonic flatbands in anisotropic van der Waals metasurfaces.” Nature Materials (2026), published October 29, 2026. DOI: 10.1038/s41563-026-02776-3. Received September 16, 2025; accepted September 18, 2026. Equal contribution: Heimig and Weber. Corresponding authors: Genco, Antonov, Tittl. Affiliations: Institute of Photonics, Hamburg University of Technology; Chair in Hybrid Nanosystems, Nanoinstitute Munich, LMU Munich; Dipartimento di Fisica, Politecnico di Milano; Università di Pisa; IFN-CNR Milano; MEPhI; MISIS. Data: Zenodo, DOI: 10.5281/zenodo.22256282. Preprint: arXiv:2509.01258.
For the qBIC context, the paper’s own framing cites symmetry-protected BICs with theoretically infinite quality factors and the Brillouin-zone-folding qBIC construction. For the ReS₂ excitonic background, see Coriolano, A. et al. “Rydberg polaritons in ReS₂ crystals.” Science Advances 8, eadd8857 (2022). For flatband polariton condensation in lattice geometries, the context the paper contrasts itself against, see the kagome and Lieb polariton-lattice literature cited therein. This article is analysis of published research, not investment advice.