Ta₂NiSe₅ is the excitonic insulator field’s cleanest test case, and its dirtiest secret is that the lattice keeps getting in the way. The material is a layered chalcogenide in which electrons and holes are thought to bind into excitons and condense into an ordered ground state, opening an insulating gap. But the condensation arrives hand in hand with a structural distortion — the crystal goes from orthorhombic to monoclinic around 328 kelvin, shortening some tantalum–nickel bonds as the excitonic order hybridises nickel 3d and tantalum 5d orbitals. Every measurement of the electronic order therefore carries a lattice shadow, and the field has spent years trying to separate the two.

A paper published in Discover Physics on February 4, 2026, offers the sharpest separation yet — not by removing the lattice, but by weakening it chemically. Masaki Hattori, Mario Okawa, and colleagues at Waseda University and the University of Tokyo’s Institute for Solid State Physics, led by corresponding authors Takashi Mizokawa and Kozo Okazaki, replaced 40 percent of the selenium in Ta₂NiSe₅ with sulfur and watched what happened when an intense laser pulse drove the crystal from its excitonic insulating state into a semimetal. The gap collapsed by 80 femtoseconds — much sooner than in the cobalt-substituted relative, where it survives to roughly 170 femtoseconds. It also leaves a puzzle the authors do not hide: the semimetal that emerges looks gapless, when the simple theory says a finite bare gap should have survived.

The cleanest candidate, with a lattice attached

The excitonic insulator was proposed in 1967 by Jérôme, Rice, and Kohn: in a narrow-gap semiconductor or a semimetal, the Coulomb attraction between electrons and holes can be strong enough that the binding energy of an electron–hole pair exceeds the band gap, so excitons form spontaneously and condense. Confirmation has been slow, because the most obvious signature of a condensate — a periodic modulation of the charge density — is exactly what a conventional lattice-driven charge-density wave produces by another route. TiSe₂, the longest-running candidate, is entangled with a charge-density wave; Tm(Se,Te) is entangled with magnetism. Ta₂NiSe₅, with neither, has been the comparatively ideal platform.

Chemical substitution lets the field change their balance. The phase diagram mapped by Lu and colleagues has two directions. Applying physical pressure shrinks the bare band gap and pushes the system toward the BCS side, where electron–hole pairs are large and loosely bound; above a critical pressure around 8 GPa the excitonic state is suppressed and superconductivity appears. Substituting sulfur for selenium does the opposite: it increases the bare band gap and drives the system toward the BEC side, where pairs are tightly bound. It also does something pressure does not. Raman spectroscopy by Volkov and colleagues showed that sulfur doping suppresses the ferroelastic fluctuations tied to the monoclinic distortion, making the distortion negligibly small at high sulfur content. The lattice steps partly out of the picture — while the excitonic gap remains.

The experiment

The measurements were made at 100 kelvin, well inside the excitonic insulating phase. Static photoemission with a helium discharge lamp established the starting point: the valence band top sits about 0.3 electronvolts below the Fermi level, shifted upward relative to pure Ta₂NiSe₅, where it sits around 0.2 electronvolts — and a clear excitonic gap is visible.

The time-resolved measurements used the high-harmonic tr-ARPES setup at the LASOR facility of the University of Tokyo’s Institute for Solid State Physics. The pump was a titanium–sapphire amplifier at 800 nanometres, 35-femtosecond pulses, at a fluence of 3.00 millijoules per square centimetre. The probe was the seventh harmonic of the frequency-doubled pump, at 21.7 electronvolts, with an evaluated temporal resolution of about 80 femtoseconds. That resolution matters for reading the results: the headline collapse time and the instrument’s own time resolution are the same number, so “by 80 femtoseconds” is the measurement’s statement, and the true collapse could in principle be faster still.

The staggered collapse

The central finding is a sequence, not a single event. At a delay of 27 femtoseconds, the conduction band above the Fermi level is already populated by photoexcited electrons, and the valence band peak around −0.3 electronvolts has broadened considerably. But the gap has not closed: the valence band top has not shifted toward the Fermi level, and the spectral weight from −0.1 to +0.1 electronvolts is not fully enhanced. The carriers arrive first; the gap follows.

By 80 femtoseconds, the gap is completely closed. The conduction and valence bands cross the Fermi level, creating what the authors describe as a nodal semimetal state. The gap remains absent at 133 femtoseconds. By 400 femtoseconds, the photoexcited electrons in the conduction band have mostly relaxed away, leaving the broadened valence band behind.

Fitting the temporal evolution of the photoinduced spectral weight above the Fermi level gives a rise time of about 74 femtoseconds and a decay time of roughly 307 femtoseconds — the decay longer than the values reported for pure Ta₂NiSe₅ and for the cobalt-substituted relative. Dividing the measured spectra by Fermi–Dirac functions, the authors estimate that the thermalised photoexcited electrons reached roughly 1300 to 1500 kelvin, with the spectral weight at the Fermi level reaching about 40 percent of the valence band peak. The upward shift of the valence band top follows its immediate broadening and peaks around 130 to 240 femtoseconds.

The delay itself is informative. At 27 femtoseconds, the estimated remaining gap is about 0.2 electronvolts — larger than the lattice thermal fluctuation energy. That is the paper’s argument that the delayed collapse reflects persistent electron–hole interaction rather than lattice dynamics alone: the excitonic binding is still holding the gap open after the carriers have arrived, and it takes tens of femtoseconds more for the screening to finish the job.

Faster than cobalt

The comparison with the cobalt-substituted relative is where the sulfur substitution makes its mark. In Ta₂Ni₀.₉Co₀.₁Se₅, cobalt substitution for nickel drives the system toward the BCS side, and previous work by Takahashi and colleagues found that the gap did not collapse even at 70 femtoseconds, disappearing only by 170 femtoseconds. In the sulfur-doped crystal, the gap is gone by 80 femtoseconds — less than half the delay.

The authors attribute the difference to the electron–lattice coupling. In the cobalt system, the monoclinic distortion — and the ferroelastic fluctuations that sustain it — remain strong, and the delayed optical response of the valence band has been interpreted as the signature of that lattice involvement: the lattice has to move before the gap can fully collapse. In the sulfur-doped system, the Raman evidence says those ferroelastic fluctuations are suppressed and the monoclinic distortion is weakened. With less lattice involvement to slow it down, the excitonic gap collapses more promptly once the photoexcited carriers screen the electron–hole attraction.

This is the sense in which the delay time of the gap collapse becomes a probe. Across the Ta₂NiSe₅ family, the same experiment returns different collapse times depending on where the material sits in the phase diagram and how strongly the lattice is coupled to the electronic order. Pressure and cobalt substitution approach from the BCS side; sulfur substitution approaches from the BEC side.

There is a wrinkle the authors flag themselves. It is, on the face of it, puzzling that the gap collapses faster in the sulfur-doped system, which is supposed to sit in the BEC regime with strong electron–hole interaction. Stronger binding should, naively, mean a more robust gap. The resolution they offer is that the collapse time measures the balance between the excitonic interaction and the electron–lattice interaction, not the binding strength alone — and in the sulfur system the lattice term has fallen further than the excitonic term has risen. That is a reasonable reading of the combined evidence, but it is an inference across two techniques, and it should be read as one.

The gap that should have survived

The deeper puzzle is what the photoinduced state actually is. In the simple excitonic picture, screening the Coulomb attraction should reveal the bare band structure — the bands as they would be without the excitonic coupling. For pure Ta₂NiSe₅, the bare structure is that of a semimetal or a zero-gap semiconductor, so a gapless photoinduced state is expected. But for Ta₂Ni(Se₀.₆S₀.₄)₅, the bare structure may retain a finite gap of order 0.1 electronvolts, extrapolating from the known gap of the fully sulfur-substituted compound Ta₂NiS₅. Screening the excitons should have left a semiconductor, not a semimetal.

It did not. The photoinduced semimetal phase of the doped crystal looks remarkably similar to that of pure Ta₂NiSe₅, with substantial overlap between the conduction and valence bands. Something else is closing the gap — or, more precisely, something else is responsible for part of the gap in the sulfur-rich compounds, and the photoexcitation removes that something too.

The authors’ proposed answer reaches into the chemistry of nickel. In the negative charge-transfer picture, the electronic configuration of nickel is not the simple d¹⁰ state of a band insulator, but closer to d⁹L or even d¹⁰L², where L denotes a ligand hole in the sulfur 3p orbitals. In that valence-bond state, the band gap can be larger than the excitonic interaction alone would produce, and its magnitude can suppress the monoclinic distortion — consistent with the Raman observation that the distortion fades with sulfur content. In the photoinduced phase, the ligand holes could dissociate from the nickel sites, and the gap could collapse for a reason that has nothing to do with excitons.

This is, by the authors’ own framing, a proposal, not a finding. It comes with a test attached: time-resolved resonant inelastic X-ray scattering, which can watch the ligand-hole states directly, could settle whether the charge-transfer picture is right. Until that experiment is done, the gapless photoinduced state is a fact in search of a complete explanation — and the excitonic interpretation of the collapse, while supported by the timing evidence, shares the stage with a chemical mechanism the same data cannot rule out.

A second observation adds to the caution. At 27 femtoseconds, the conduction band dispersion already resembles that of the semimetal phase, even though the valence band top has not yet shifted — suggesting the semimetal and gapped phases may briefly coexist, with the gap collapsing inhomogeneously in time. But the paper notes that the energy shift of the valence band top is not compatible with a simple change in the volume fraction of the two phases, so the coexistence picture is itself incomplete. The transition is not a clean switch from one uniform state to another; it is a messy, staggered process, and the mess is part of the result.

What it is not

Three cautions, stated plainly because the framing invites overreading. First, this is not a proof that Ta₂Ni(Se₀.₆S₀.₄)₅ is a “purer” excitonic insulator in any absolute sense. The paper’s claim is comparative and specific: the photoinduced transition in this composition shows less lattice involvement than in the pure or cobalt-substituted systems, as measured by the collapse delay. The underlying ground state remains a coupled electron–lattice system.

Second, this is not a device result. The photoinduced semimetal exists for a few hundred femtoseconds at 100 kelvin under a pump fluence of 3.00 millijoules per square centimetre, with the electron system heated to well over a thousand kelvin. Nothing in the paper points toward an application.

Third, the February publication date matters for how this story should be read. The paper appeared in Discover Physics on February 4, 2026, and reached wider trade coverage in late October. This is an analysis of published research, not breaking news, and the resonant inelastic X-ray scattering experiment the authors propose has not yet been reported.

What to watch

The most immediate follow-up is the experiment the authors themselves propose: time-resolved resonant inelastic X-ray scattering on the sulfur-doped system, to test whether ligand holes in the sulfur 3p orbitals dissociate under photoexcitation. If the charge-transfer picture is confirmed, the sulfur-doped phase diagram will need to be read as a three-way competition — excitonic, lattice, and charge-transfer — rather than a two-way one.

A second follow-up is systematic: map the collapse delay across the full sulfur series, from x = 0 to x = 0.6 and beyond, where the zero-gap condition is reached and the transition temperature falls sharply. If the delay tracks the weakening of the monoclinic distortion smoothly, the case for the collapse time as a probe of electron–lattice coupling strengthens. If it jumps, or reverses, that would point to a change in mechanism.

A third is to repeat the measurement with better time resolution. The 80-femtosecond collapse time sits at the edge of this experiment’s temporal resolution, so the true collapse could be faster. A shorter probe pulse would sharpen the comparison with the cobalt system’s 170 femtoseconds.

Ta₂NiSe₅ has been the field’s cleanest candidate for a decade, and the cleanest thing about it may be how clearly it shows that “clean” is relative. Sulfur doping takes part of the lattice out of the picture, and the gap collapses faster — which is what the excitonic picture predicts. But the state it collapses into is gapless when it should not be, which is what the excitonic picture does not predict. Both halves of that sentence are now measured facts, and the field’s job is to hold them together.

Sources and further reading

Hattori, M.; Okawa, M.; Suzuki, T.; Takahashi, Y.; Lu, Y.; Takagi, H.; Zhong, Y.; Liu, K.; Kanai, T.; Itatani, J.; Shin, S.; Okazaki, K.; Mizokawa, T. “Impact of S substitution for Se on the photoinduced insulator-to-metal transition in an excitonic insulator Ta₂NiSe₅.” Discover Physics 2, 3 (2026), published February 4, 2026. DOI: 10.1007/s44418-026-00004-0. Received March 30, 2025; accepted January 13, 2026. Open access (CC BY-NC-ND 4.0). For a plain-language account, see “Sulfur Doping Speeds the Light-Driven Metal Transition in an Excitonic Insulator,” scienmag.com, October 29, 2026.

For the cobalt-substituted comparison, see Mitsuoka, T.; et al. “Quasi One-Dimensional Band Structure of Photoinduced Semimetal Phase of Ta₂Ni₁₋ₓCoₓSe₅ (x = 0.0 and 0.1).” Journal of the Physical Society of Japan 92, 023703 (2023), and Takahashi, Y.; et al. on the delayed optical response of the valence band in the cobalt system, as cited in the Hattori paper. For the sulfur phase diagram, see Lu, Y. F.; et al. “Zero-gap semiconductor to excitonic insulator transition in Ta₂NiSe₅.” Nature Communications 8, 14408 (2017). For the Raman evidence that sulfur doping suppresses ferroelastic fluctuations, see Volkov, P. A.; et al., as cited in the Hattori paper. For the robustness of excitonic coupling against sulfur-induced inhomogeneity, see Hattori, M.; et al. “Robustness of excitonic coupling in Ta₂NiSe₅ against electronic inhomogeneity introduced by S substitution for Se.” Advanced Quantum Technologies 6, 2300034 (2023). For the excitonic-insulator proposal, see Jérôme, D.; Rice, T. M.; Kohn, W. “Excitonic insulator.” Physical Review 158, 462–475 (1967). This article is analysis of published research, not investment advice.