For six years, one spectral line has refused to explain itself. Nickel phosphorus trisulfide, NiPS₃, is a layered antiferromagnet: below its magnetic ordering temperature, its photoluminescence collapses into an emission line at 1.475 electron-volts so narrow — linewidths below 500 micro-electron-volts — that it looks less like the glow of a semiconductor and more like the fingerprint of a single, well-defined quantum state. The line appears only once the zigzag antiferromagnetic order sets in, and its polarization tracks the magnetic lattice. Everything about it says the exciton and the magnetism are the same story. But what kind of exciton, exactly, has been argued over since the line’s discovery: a spin-orbit-entangled excitation built from Zhang–Rice states, as the team that found it proposed in 2020, or a Hund’s exciton — a triplet-to-singlet flip on a single nickel site, its energy set by Hund’s exchange — as resonant X-ray scattering experiments later suggested?
A new paper in Nature Communications answers by stretching the material and watching which lines move and by how much. Joanna L. P. Wolff, Loïc Moczko, Jérémy Thoraval, Michelangelo Romeo, Benjamin Bacq-Labreuil, Stéphane Berciaud, and Arnaud Gloppe, all of the Institut de Physique et Chimie des Matériaux de Strasbourg (IPCMS), with Bacq-Labreuil also affiliated with the Institut Quantique at the Université de Sherbrooke, suspended a ten-layer NiPS₃ membrane over a five-micrometre hole, pulled it electrostatically like a drumhead, and measured how each emission line shifted under strain. The main exciton barely moved. Its satellites moved twice as much, and then four times as much. That ladder — one, two, four — is the signature of an excitation that starts localized on a single site and delocalizes step by step as it hops through the magnetic lattice, picking up a magnon at each hop. It is exactly what the Hund’s picture, formalized as a low-energy t–J model, predicts, and it is difficult to reconcile with a delocalized Zhang–Rice exciton. The paper appeared in Nature Communications on October 27, 2026; a preprint has been public since December 2025.
A line that refused to behave
NiPS₃ is a transition-metal phosphorus trichalcogenide: magnetic nickel ions on a honeycomb lattice, separated by phosphorus and sulfur. In the bulk, Ni²⁺ moments order below about 155 kelvin into zigzag chains — ferromagnetic along each chain, antiferromagnetic between chains — with moments pointing mostly along the crystal’s a-axis. The luminescence is that order’s shadow: the 1.475 eV line exists only in the ordered phase, is polarized orthogonally to the zigzag direction in the bulk, and narrows dramatically at low temperature.
The 2020 discovery, by Soonmin Kang, Je-Geun Park, and colleagues, interpreted the line as a coherent many-body exciton arising from Zhang–Rice states — the entangled nickel–sulfur singlets familiar from cuprate physics — with the excitation described as a transition from a Zhang–Rice triplet to a Zhang–Rice singlet. The evidence was strong: resonant inelastic X-ray scattering, photoluminescence, and optical absorption, tied together by configuration-interaction theory. But a competing reading soon gained ground. In 2024, Wei He and colleagues at Brookhaven National Laboratory, using resonant inelastic X-ray scattering, reported a magnetically propagating Hund’s exciton in NiPS₃ — an excitation whose energy is set by Hund’s exchange coupling, dispersing through the crystal in a way reminiscent of a double magnon. The two pictures differ in something measurable: how spread out the excitation is. A Zhang–Rice exciton is an extended, coherent object; a Hund’s exciton is, at birth, a local spin flip on one nickel site. The Strasbourg experiment turns that difference into a measurement.
The drum
The device is a study in controlled deformation. A ten-layer NiPS₃ flake, about nine nanometres thick, is suspended over a five-micrometre-diameter hole etched into a silicon/silicon-dioxide substrate and contacted by titanium/gold electrodes. Ten layers is a deliberate compromise: thick enough to preserve a clean photoluminescence spectrum — the emission degrades as the monolayer limit is approached — and thin enough that the stacking-induced anisotropy, which pins the zigzag direction in the bulk, is weakened, while the membrane stays mechanically compliant. A static gate voltage between the gold electrode and the silicon substrate pulls the membrane downward; at 75 volts the center deflects by roughly 80 nanometres, producing an integrated in-plane strain of about 0.07 percent. An alternating gate component drives the membrane’s mechanical resonance, read out by interferometry: the fundamental mode sits at 48.6 megahertz, with quality factors of 10³ to 10⁴. The same 633-nanometre laser that probes the mechanics also excites the photoluminescence, all in vacuum at 5.5 kelvin, far below the ordering temperature.
That combination — a light-emitting membrane whose strain can be dialed continuously and read back optically — is what makes the experiment possible. Strain is the probe. In a conventional semiconductor, stretching the lattice shifts exciton energies by tens of milli-electron-volts per percent of strain, because the exciton is a large, Coulomb-bound object whose energy tracks the band edges. A localized excitation, confined to a single site and a few neighbors, barely notices. So the question the experiment asks is simple: when you stretch NiPS₃, how much does each emission line move?
Two zigzag directions in one laser spot
Before straining anything, the team mapped what the membrane actually emits. Polarization-resolved photoluminescence at the center of the drum reveals not one main peak but two, and that doubling is itself a result. The principal peak, labeled X_A, sits at 1.4758 eV. About 0.3 milli-electron-volts higher, with roughly half the intensity and a comparable linewidth, sits a second peak, X_B, linearly polarized at about 72 degrees to X_A. In the bulk, the monoclinic stacking forces a single zigzag direction; in a few-layer flake, the threefold hexagonal symmetry is nearly restored, and the magnetic moments can settle along a discrete set of directions — nematic states. The two peaks are two such states, two zigzag domains, coexisting within the sub-micrometre laser spot. The nematic families are robust under strain: polarization-resolved measurements as the gate voltage rises show no significant change in the polarization distribution, and no domain switching.
Above the main peaks sit the satellites that carry the paper’s argument. A dimmer peak S_A appears 1.8 milli-electron-volts above X_A, sharing X_A’s polarization. Its partner S_B sits about 1.9 milli-electron-volts above X_B with X_B’s polarization. Higher still, around 3.5 milli-electron-volts above the main line, stands a broader, fainter peak S′ with no clear polarization dependence at zero gate voltage. The energies are suggestive: a zone-center magnon in few-layer NiPS₃ sits near 1.5 milli-electron-volts, hinted at by Raman and inelastic neutron scattering. S is one magnon above X; S′ is roughly two. The paper’s reading, drawn from ab initio calculations combined with a low-energy effective t–J model, is that S and S′ are the same excitation after one and two magnon-mediated hops: the exciton moves from site to site through the magnetic lattice, each hop flipping spins along the way and leaving magnons that dress the moving excitation. The dominant hopping term in the model couples third-nearest neighbors, across zigzag chains — the same logic as hole motion in the copper-oxide superconductors, where a moving charge leaves a string of disturbed spins behind it.
Strain as a ruler
The strain measurements give that picture its quantitative teeth. At the center of the drum, the main peak X_A red-shifts by 0.18 milli-electron-volts as the gate voltage rises from 0 to 80 volts — about half its initial apparent linewidth of 0.4 milli-electron-volts — following the fourth power of the gate voltage, as the integrated radial strain does. Converted to strain, the shift is in the range of −1 milli-electron-volt per percent: roughly two orders of magnitude smaller than the strain response of Coulomb-bound Wannier–Mott excitons in transition-metal dichalcogenides, and in the territory of quantum dots and defect-like emitters. The shift is reversible, and it is absent on a nearby supported region of the same flake under identical gate voltages, which rules out doping or field effects. X_B’s strain sensitivity matches X_A’s, and the two peaks’ shifts track each other position by position: the strain-perturbed environment does not care which zigzag direction the exciton belongs to, itself evidence that the bare exciton is very localized, at the scale of a lattice parameter.
The satellites tell a different story, and it is the contrast that matters. The one-hop satellite S_A shifts about twice as much as X_A under the same strain — its gate-voltage fit coefficient is −10.3 pico-electron-volts per volt⁴ against −4.5 for X_A, and the two drums measured agree, at −9 ± 1.5 and −4.2 ± 0.3 respectively. The two-hop feature S′ shifts up to −1.2 milli-electron-volts, about twice the sensitivity of S and four times that of X. The ladder is clean: one, two, four. Each magnon-mediated hop makes the exciton less localized, more extended, more sensitive to the lattice being stretched. S′ adds one wrinkle: its shift is maximized for excitons recombining with polarization along the B-family direction, a hint of two-hop selection rules the authors flag as not yet established.
That is the sense in which the experiment weighs in on the exciton’s identity. The main line’s near-immobility under strain matches a triplet-to-singlet transition between two wavefunctions centered on a single nickel site and spreading only onto neighboring phosphorus and sulfur atoms, with an energy set by Hund’s exchange, 2J ≈ 1.47 eV. A delocalized Zhang–Rice exciton, built from extended states, would be expected to move with the lattice the way band-edge excitons do. It does not. The dressed satellites confirm the mechanism rather than merely the starting point: the excitation does move, but by hopping through the magnetic order, and every hop costs it localization. The paper is careful about the limits — the polarization rules for two-hop recombination remain open, and no signature of the second zone-center antiferromagnetic magnon, expected near 5 milli-electron-volts, is observed — but the direction of the evidence points away from the Zhang–Rice reading that launched the field.
What it is not
Three distinctions keep this result in its proper place. First, it is not the Northwestern electroluminescence result from 2025, and the two should not be merged in memory just because they share a material and a linewidth. That work, by Dmitry Lebedev, J. Tyler Gish, Nathaniel P. Stern, Mark C. Hersam, and colleagues, drove NiPS₃ electrically: gate-tunable transistors that stay conductive below the ordering temperature, with capacitively induced electroluminescence from the magnetic exciton at about 1 milli-electron-volt linewidth and a degree of linear polarization of 0.78. It was a device demonstration — that the magnetic exciton can be excited with current rather than light. The Strasbourg work is optical spectroscopy on a suspended membrane, and its contribution is a microscopic identification: what the exciton is, and how it moves. The two results are complementary, not competing.
Second, this is not a room-temperature result, and nothing in it should be read as a path to a room-temperature device. The measurements are at 5.5 kelvin, in vacuum, on a membrane that only luminesces this way because the antiferromagnetic order is intact. That order vanishes above roughly 155 kelvin in the bulk, and with it the narrow line. The authors’ suggested futures — optomechanical coupling between the membrane’s vibrations and its excitons, or strain-engineered control of magnon-dressed excitons — live at cryogenic temperatures unless some other material carries the same physics upward.
Third, this is one membrane, ten layers thick, measured in detail, with a second drum confirming the strain coefficients. The nematic coexistence it reports is a property of few-layer flakes where the bulk stacking anisotropy has weakened; it does not rewrite the bulk picture, where a single zigzag direction dominates. And the t–J description of magnon-dressed hopping is a model fit to this system, not a measurement of a hopping rate in real time — time-resolved measurements, which the authors propose as next steps, would be needed to watch the hops happen rather than infer them from energy shifts.
What to watch
The most immediate consequence is for the NiPS₃ literature itself. The field now has a strain-based fingerprint for distinguishing bare and dressed magnetic excitons: a line that barely moves is local; a satellite that moves twice as much has hopped once; one that moves four times as much has hopped twice. Groups working on related trichalcogenides — MnPS₃, FePS₃, CoPS₃, and the alloys the authors suggest, in which chemical substitution tunes the magnetic configuration — can apply the same ruler.
The longer consequence is for the hybrid device the membrane embodies. A suspended antiferromagnetic membrane is simultaneously a mechanical resonator, an optical emitter, and a magnetic system, and this paper shows the three degrees of freedom can be addressed together: the gate voltage tunes the strain, the strain tunes the emission, and the emission reports on the magnetic excitation’s motion. The authors raise dynamical hybrid optomechanical coupling — the tripartite interaction of phonons, magnons, and excitons — as the direction that would make the drum more than a measurement platform. That is a research program, not a result, and it inherits the cryogenic caveat. But the measurement that motivates it is now on the record: an exciton that sits still, and then, hop by hop, learns to move.
Sources and further reading
Wolff, J. L. P.; Moczko, L.; Thoraval, J.; Romeo, M.; Bacq-Labreuil, B.; Berciaud, S.; Gloppe, A. “Magnetic excitons in a suspended 2D antiferromagnetic membrane.” Nature Communications (published October 27, 2026); preprint arXiv:2512.20528 (December 2025). Device and measurement details (ten-layer membrane, five-micrometre hole, 5.5 K, 633-nm excitation, 48.6 MHz fundamental mode), nematic families X_A/X_B (1.4758 eV, 0.3 meV splitting, 72° polarization angle), satellites S (1.8–1.9 meV) and S′ (~3.5 meV), and strain sensitivities (X ≈ −1 meV/%; S ≈ 2×; S′ ≈ 4×) are from the paper. For the Zhang–Rice reading of the NiPS₃ exciton, see S. Kang et al., “Coherent many-body exciton in van der Waals antiferromagnet NiPS₃,” Nature 583, 785–789 (2020), DOI: 10.1038/s41586-020-2520-5. For the Hund’s exciton evidence, see W. He et al., “Magnetically propagating Hund’s exciton in van der Waals antiferromagnet NiPS₃,” Nature Communications (2024), DOI: 10.1038/s41467-024-47852-x. For the electrically driven counterpart, see D. Lebedev et al., “Ultranarrow electroluminescence from magnetic excitons in the van der Waals antiferromagnetic semiconductor NiPS₃,” Nature Communications 16, 10550 (2025), DOI: 10.1038/s41467-025-65576-4. This article is analysis of published research, not investment advice.