WorksPhysics and Astronomy
Charge-transfer excitation drives cooperative phonon-assisted exciton self-trapping in NiPS
Abstract
Carrier self-trapping is a fundamental relaxation pathway in strongly correlated quantum materials, yet the microscopic lattice motions that drive it have remained largely inaccessible to direct observation. Here we investigate the van der Waals antiferromagnet NiPS upon excitation across its ligand-to-metal charge-transfer gap, and reveal a nonequilibrium pathway that initially generates unbound carriers before they evolve into correlated excitonic states. Using ultrabroadband optical-pump terahertz-probe spectroscopy, we simultaneously track carrier localization and the associated lattice dynamics in real time. The transient conductive response quenches with a time constant of 109 fs and evolves into a localized exciton state, accompanied by pronounced renormalization of Ni-S optical phonons and the emergence of a transient 2.7 THz vibrational mode. Temperature-dependent measurements, together with mode-resolved dielectric dynamics and symmetry analysis, identify the 2.7 and 8.0 THz responses as spectroscopic indicators of exciton self-trapping and point to a cooperative two-phonon mechanism. These findings establish the excitation pathway as a key parameter controlling carrier localization and many-body relaxation in strongly coupled quantum materials.
Introduction
In correlated quantum materials, the competition between electronic delocalization and lattice-driven carrier localization plays a central role in determining nonequilibrium dynamics. Optical excitation can drive systems with strongly coupled charge, spin, orbital, and lattice degrees of freedom far from equilibrium, generating transient states inaccessible in equilibrium [1, 2, 3]. The relaxation of photoexcited carriers in strongly coupled materials is governed by the interplay between electronic and lattice degrees of freedom, which drives carrier localization, exciton formation, and emergent many-body behavior [4, 5, 6]. Yet the microscopic lattice coordinates that mediate these transformations are rarely identified directly. Determining such reaction coordinates is therefore a prerequisite for understanding and controlling nonequilibrium phenomena in correlated quantum materials. Two-dimensional van der Waals magnets are particularly well suited to such studies because the reduced dimensionality enhances Coulomb interactions and amplifies the coupling between charge, spin, and lattice degrees of freedom [7, 8, 9, 10]. Within this class, the charge-transfer antiferromagnet offers an ideal setting in which strong S – Ni hybridization makes the elementary excitations highly sensitive to local lattice distortions [11]. Optical and Raman spectroscopies have revealed pronounced exciton–phonon coupling through the observation of phonon sidebands, exciton–phonon bound states, and lattice-induced spectral renormalization [12, 13], highlighting the central role of lattice dynamics in shaping the excited-state landscape of .
Recent ultrafast studies have reported diverse nonequilibrium responses in following resonant or near-resonant excitation of the excitonic manifold (– eV), ranging from transient Drude-like conductivity [14] to population-inverted excitonic states accompanied by stimulated emission [15]. However, these studies primarily probe low-energy excitations that directly address pre-existing excitonic or localized – states. By contrast, excitation across the ligand-to-metal charge-transfer (LMCT) gap generates initially unbound, high-energy carriers and accesses a qualitatively different nonequilibrium relaxation pathway. At the microscopic level, the LMCT transition transfers electrons from sulfur orbitals into nickel states, inducing substantial inter-site charge redistribution and strongly perturbing the local Ni–S bonding environment [11, 16]. Unlike resonant excitation of pre-existing excitons, this process injects substantial excess energy into the coupled charge–lattice system and strongly perturbs the local Ni–S bonding environment, driving pronounced non-adiabatic dynamics. The nonequilibrium trajectory of the photoexcited state is then determined by the competition between electronic delocalization and lattice-assisted localization. Time-resolved optical studies indicate that charge-transfer excitations relax into localized excitonic states within ps [17]. However, purely electronic probes are inherently insensitive to the lattice coordinates involved. Resolving this mechanism requires simultaneous access to both electronic and lattice dynamics.
Here, we employ sub-100-fs ultrabroadband optical-pump terahertz-probe (OPTP) spectroscopy to map the coupled charge and lattice dynamics of following excitation across the LMCT gap. The exceptionally broad probe bandwidth, extending to 15 THz, provides simultaneous access to macroscopic carrier transport and the optical phonons involved in energy dissipation and lattice relaxation. We observe a rapid quenching of the transient conductive response into a localized exciton state, accompanied by pronounced renormalization of Ni–S optical phonons and the transient activation of a folded low-frequency () vibrational mode superimposed on an imaginary dielectric sign-reversing electronic background. Combined with transient absorption spectroscopy and temperature-dependent OPTP measurements, these excited-state dynamic signatures point to a cooperative phonon-assisted pathway toward self-trapped exciton formation.
Results
Crystal structure and static properties of NiPS. As illustrated in Figure 1A, NiPS crystallizes in a monoclinic () structure, in which () ions form a honeycomb network coordinated by edge-sharing sulfur octahedra. The local ligand field is trigonally distorted from to symmetry, strongly modifying the crystal-field manifold of the Ni orbitals and thereby shaping the low-energy electronic excitations of the material [18, 19, 20]. The diffuse-reflectance absorbance spectrum (Figure 1B) reveals several characteristic optical transitions. Weak features below 1.2 eV arise from Laporte-forbidden on-site – excitations from the ground state into the split and states derived from the manifold [21]. At higher energies, a broad absorption continuum from 1.2 to 2.1 eV dominates the optical response. Within this continuum, a pronounced resonance centered near 1.4 eV (A) is consistent with the spin-orbit-entangled charge-transfer excitons reported previously in NiPS [14, 22]. Although substantially broadened at room temperature owing to enhanced thermal decoherence and the absence of long-range magnetic stabilization [21], the persistence of this feature indicates that excitonic correlations remain significant in the paramagnetic phase.

Figure 1. Crystal structure, optical transitions, and vibrational properties of . (A) Crystal structure of in top and side views together with the local octahedral coordination (upper right). The schematic illustrates the ligand-to-metal charge-transfer excitation from S to Ni orbitals and the associated coupling to lattice vibrations. (B) Room-temperature optical absorbance spectrum measured in diffuse-reflectance geometry. Gaussian decomposition resolves the excitonic resonance near 1.4 eV (A, red), the LMCT absorption at 1.6 eV (B, green), and higher-energy crystal-field () excitations near 1.78 and 2.0 eV (blue and pink). (C) Time-domain terahertz electric-field waveforms transmitted through free space (cyan) and a crystal (red). Inset: Corresponding frequency-domain spectra, featuring infrared-active optical phonons at 5.7 and 8.0 THz. (D) Room-temperature photoluminescence (PL) spectrum under continuous-wave 637 nm excitation. Shaded areas: Gaussian components of the dominant excitonic emissions (A and A) and a lower-energy sideband (A) that appears at excitation power above 11 . Inset: excitation-power dependence of the integrated intensity ratio A/A.
The strongest absorption feature, centered at 1.6 eV (B), originates from a ligand-to-metal charge-transfer (LMCT) transition [23]. Density-functional-theory (DFT) calculations (Figure S2) show that this excitation transfers electrons from S into antibonding Ni () states, producing inter-site redistribution of electronic charge within the Ni–S network. Moreover, within the strong-coupling framework [24], the narrow bandwidth of the localized Ni states enhances electron-phonon coupling, making LMCT excitation a particularly effective route to carrier localization irrespective of long-range magnetic order [25]. Additional absorption bands near 1.78 and 2.0 eV are assigned to higher-energy crystal-field (–) excitations associated with the split manifold (Supplementary fig. S3). We therefore employ a 1.91 eV (650 nm) pump, which lies above the optical gap to access the LMCT continuum rather than enter the low-energy excitonic manifold or the Laporte-forbidden crystal-field transitions, whose oscillator strengths are orders of magnitude weaker. This excitation pathway generates initially unbound carriers and injects substantial excess energy into the coupled charge-lattice system, thereby accessing a nonequilibrium regime distinct from that explored in previous resonant-excitation studies.

Figure S2. Electronic structure and orbital character of NiPS. Left: Orbital-projected electronic band structure calculated using the strongly constrained and appropriately normed (SCAN) meta-GGA functional. The size and color of the circles indicate the fractional contributions of Ni (red), S (yellow), and P (cyan) orbitals. The valence-band maximum (VBM) is aligned to 0 eV (horizontal dashed line). Right: Corresponding orbital-projected partial density of states (PDOS). The valence-band maximum region is composed predominantly of strongly hybridized S and Ni states, whereas the conduction-band minimum is dominated by localized Ni orbitals. This orbital configuration is consistent with the charge-transfer-insulator character of NiPS and provides the microscopic basis for the ligand-to-metal charge-transfer excitation (S Ni ) employed in the ultrafast measurements. The calculated charge-transfer gap is approximately 1.79 eV.
The lattice degrees of freedom are characterized by the THz transmission spectrum shown in Figure 1C. Two prominent infrared-active optical phonons are observed at 5.7 and 8.0 THz. In particular, the high energy mode corresponds to an odd-parity vibration involving in-plane sulfur motions that modulate the Ni–S octahedral environment and the crystal-field splitting of the Ni -levels, whereas the 5.7 THz mode is a vibration dominated by Ni displacements within the octahedra (Supplementary fig. S4A) [26, 27, 28]. Exciton–phonon coupling is independently evidenced in the room-temperature photoluminescence (PL) spectrum (Figure 1D), which shows two partially resolved excitonic emission features, A and A, separated by meV. Their relative spectral weights shift systematically with excitation power, consistent with previous reports of pronounced exciton-phonon coupling in NiPS [12]. At higher excitation densities, both excitonic branches exhibit sub-linear saturation together with the appearance of a lower-energy sideband (A) near 1.295 eV (Supplementary fig. S5), pointing to additional many-body relaxation channels such as exciton-exciton annihilation. Together, these measurements establish NiPS as a strongly coupled charge-lattice system and define the spectroscopic reference frame for the ultrafast experiments.
Characterization of ultrafast excitonic dynamics. To investigate the nonequilibrium evolution of charge-transfer excitations in NiPS, we employ ultrabroadband optical-pump terahertz-probe (OPTP) spectroscopy in the reflection geometry shown in Figure 2A. This configuration avoids strong attenuation of the probe field by the intrinsic infrared-active phonons (Figure 1C), thereby preserving the ultrabroad bandwidth and enabling direct monitoring of the transient electrodynamic response. Following 650 nm (1.91 eV) excitation at a fluence of 1.3 mJ/cm, the transient reflectivity change () rises rapidly with a characteristic time () of 90 fs (Figure 2B), marking the prompt generation of itinerant carriers in the LMCT continuum. This initial response then decays with fs, far faster than typical carrier recombination or lattice thermalization processes [14, 29, 30]. Such rapid suppression of the conductive response indicates efficient conversion of initially photoexcited electron-hole pairs into strongly coupled excitons. Following this stage, the THz response undergoes a sign inversion and evolves into a pronounced negative signal over ps. This behavior contrasts sharply with the monotonic recovery commonly observed in conventional semiconductors and indicates a qualitative change in the nature of the photoexcited state. The negative response signifies that the transient electrodynamics are no longer governed by itinerant carriers but instead by localized photoexcited states with a predominantly dielectric character. On longer timescales, the signal gradually recovers with a characteristic lifetime of ps.

Figure 2. Ultrafast carrier dynamics following charge-transfer excitation in . (A) Schematic of the ultrabroadband optical-pump terahertz-probe (OPTP) setup in reflection (RL) geometry. (B) Room-temperature transient response measured by OPTP (red circles) and transient absorption spectroscopy (TAS, black circles). Dashed curves represent convolved multi-exponential fits. Regions I–IV denote the successive dynamical stages associated with carrier generation, ultrafast exciton formation, exciton localization, and subsequent relaxation. (C and D) Pump-fluence dependence of the OPTP signal extrema corresponding to the initial positive response (C) and the subsequent negative response (D). Dashed lines are linear fits. Error bars represent standard deviations from repeated measurements. (E) OPTP dynamics measured at 75, 188, and 300 K. Shaded bands denote the data with their noise envelope, colored dashed lines are exponential fits, and grey dashed lines mark the zero level of each trace. Traces are vertically offset by the values indicated for clarity. Inset: Arrhenius analysis (dashed line) of the temperature-dependent absolute amplitude of the negative response.
Complementary transient absorption spectroscopy (TAS) measurements (Figure 2B, black symbols) provide an independent probe of the excited-state population dynamics underlying the transient electrodynamic response. Following an initial ps build-up of the band-edge ground-state bleach, the TAS signal exhibits a fast decay with a characteristic time of ps, closely matching the timescale over which the negative THz response develops. The bleach build-up, slower than the prompt THz rise and steepening as it proceeds, suggests transient occupation of a charge-neutral intermediate state such as an on-site – excitation. Analysis of the TAS amplitudes indicates that approximately 60% of the photoexcited population is transferred into a long-lived localized excitonic state. Notably, this fraction is comparable to the relative spectral weight () observed in the steady-state PL spectrum (Figure 1D). This development, synchronized with the negative THz trajectory, indicates their direct connection and is consistent with the conversion of a large fraction of the initially photoexcited carriers into localized excitonic configurations. The remaining excitonic population relaxes on a nanosecond timescale ( ns).
The origin of the THz response is further elucidated by pump-fluence-dependent measurements. As shown in Figure 2C,D, both the initial positive peak and the subsequent negative minimum scale linearly with excitation fluence across the investigated range. The absence of saturation or threshold behavior indicates that the dynamics are not dominated by high-density many-body effects. In particular, the linearity rules out strong exciton–exciton annihilation and is inconsistent with the threshold-like behavior expected for stimulated emission or population inversion scenarios proposed under resonant excitation conditions [14, 15]. Instead, the results point to an intrinsic localization process that remains operative over a broad range of excitation densities. Temperature-dependent results provide further insight, as shown in Figure 2E. Lowering the temperature progressively suppresses the negative response and slows the localization kinetics, a signature of thermal activation.
Arrhenius analysis of the negative amplitude (Figure 2E, inset) and the corresponding build-up rate (Figure S6B) yields two numerically close yet physically distinct energies, and , respectively. , consistent with from the PL spectrum, characterizes the thermally activated growth of the localized population, whereas is the activation barrier between the free-exciton and localized states. of this magnitude places in the shallow self-trapping regime, in which free and localized excitons remain simultaneously accessible at room temperature rather than the photoexcited population collapsing irreversibly into a single deep minimum (Supplementary Note 6.) It thus sets a quantitative benchmark for the microscopic origin of the localized state.

Figure S6. Excitation-energy dependence and temperature dependence of the localization dynamics. (A) Time-resolved terahertz reflectivity dynamics measured at 300 K following excitation at 650 nm (1.91 eV, red) and 800 nm (1.55 eV, blue). The 800 nm trace is multiplied by a factor of three for clarity. Both measurements were performed at the same incident pump fluence of . (B) Temperature dependence of the localization rate (upper panel) and the relaxation time constant extracted from multi-exponential fits to the OPTP dynamics (lower panel) at 75, 188, and 300 K. Dashed lines represent Arrhenius analysis employed to extract the localization barrier (upper panel) and effective activation energy (lower panel) of the subsequent relaxation process.
Spectro-temporal mapping and phonon-resolved dynamics. To directly probe the lattice dynamics accompanying exciton localization, we extracted the transient complex dielectric function () from the two-dimensional spectro-temporal OPTP measurements (Figure 3A, B, details in Supplementary Note 7.). The real component () reflects changes in dipole polarizability, whereas the imaginary component () tracks energy dissipation in the coupled charge–lattice system. Immediately after photoexcited maximum (), shows a short-lived positive response during the first fs window, corresponding to the itinerant-carrier-dominated nonequilibrium regime. Subsequently, the macroscopic low-frequency ( THz) rapidly drops to negative and reaches the minimum at ps, while the real dielectric response undergoes a concurrent rise. This delayed buildup, coinciding with the characteristic localization timescale identified from the OPTP kinetics signals the prompt quenching of the itinerant carrier response and formation of a localized dielectric state.

Figure 3. Spectro-temporal mapping of transient dielectric and vibrational dynamics in . (A and B) Two-dimensional spectro-temporal maps of the pump-induced changes in the real (, A) and imaginary (, B) components of the dielectric function at 300 K following 1.91 eV excitation. (C) Temporal evolution of the imaginary dielectric response () at 2.7 THz (blue circles), 5.7 THz (green triangles), 8.0 THz (red squares). Dashed curves are exponential fits, and error bars denote standard deviations. (D) Calculated equilibrium phonon dispersion of (solid lines) overlaid with the emergent transient branches (bubbles) derived from the charge-transfer excited state. Dark red denotes the two intrinsically THz-active modes; grey, the remainder. Symbol colors indicate the fractional atomic contributions from Ni (red), P (cyan), and S (yellow). Insets show the corresponding real-space vibrational eigenvectors of the 8.0 THz and 2.7 THz modes.
The intrinsic infrared-active phonons at 5.7 THz () and 8.0 THz () exhibit distinct nonequilibrium dynamics (Figure S8 and Figure 3C). Unlike the low-frequency modes that undergo an electrodynamic inversion, the responses of both intrinsic modes remain positive. However, the 5.7 THz mode sustains a relative steady amplitude, whereas the 8.0 THz mode undergoes a 70% suppression during the precursor-exciton formation. Following this window, the dissipative imaginary response () of both modes undergoes a prompt frequency blueshift (Figure 3B). This dynamic stiffening indicates the subsequent spatial contraction of the excitonic wavepacket, exerting localized electrostrictive stress on the octahedra and transiently enhancing the effective Ni–S restoring forces. Over the subsequent ps trajectory, the blueshift persists without progressive lattice softening, signaling a stabilized structural reorganization that accommodates the localized excitonic state [4, 31]. Concurrently, two transiently brightened modes emerge at 2.7 and 4.5 THz.

Figure S8. Spectro-temporal mapping and supplementary phonon dynamics. (A and B) Two-dimensional maps of the pump-induced change in the reflected terahertz electric field () measured at 300 K and 75 K following 1.91 eV charge-transfer excitation. The low-temperature data reveal an additional long-lived low-frequency oscillatory feature near THz, consistent with the antiferromagnetic magnon response reported previously. (C) Temporal evolution of the transient real dielectric response () at 2.7 THz (blue circles), 5.7 THz (green triangles) and 8.0 THz (red squares). Error bars represent standard deviations and dashed curves are exponential fits. (D) Real and imaginary components of the emergent 4.5 THz mode. Insets illustrate the calculated atomic displacement patterns of the vibrational eigenmodes. Error bars represent standard deviations and dashed lines are exponential fits.
Localization of the excitation removes the C-centering translation of the lattice at its site, which amounts to a local doubling of the primitive cell. This folds the zone-boundary dark branches, both transforming as odd-parity modes, to the () point and relaxes selection constraints, rendering them optically active [31]. Among all these modes (Figure 3D), the 2.7 THz branch is calculated to be one of the most efficiently activated by this symmetry breaking. Their early appearance and the ps dielectric extrema provide experimental signatures of the structural reorganization, indicating a modified lattice potential in the photoexcited state (Supplementary Note 4). Beyond the electrodynamic inversion, the negative amplitude of the 2.7 THz mode is 3 times that of the 4.5 THz mode. Although the 2.7 THz mode has a lower static density of states (Figure S4B), its threefold larger dielectric response indicates that the nonequilibrium dynamics are governed by mode-specific dipole coupling rather than available phase space.

Figure S4. Phonon dispersion and projected density of states of NiPS. (A) Phonon dispersion of the primitive cell calculated along the high-symmetry directions of the Brillouin zone. Marker sizes and color indicate the fractional atomic contributions from Ni (red), P (cyan), and S (yellow). The right panel shows the corresponding atom-projected phonon density of states (PDOS). (B) Unfolded phonon dispersion and corresponding PDOS calculated for the charge-transfer excited state. The vibrational spectrum of the photoinduced symmetry-broken state is projected back onto the equilibrium Brillouin zone, enabling identification of the origin of the emergent transient modes. The calculations reveal that the weakly dispersive 2.7 THz branch, dominated by coupled Ni–S vibrations, originates from the zone-boundary phonons that become folded toward the zone center following the photoinduced structural distortion. The low group velocity of this branch gives rise to an enhanced PDOS, consistent with the transiently activated mode observed experimentally.
Of the resolved phonon responses, the 2.7 and 8.0 THz modes constitute the most direct spectroscopic indicators of exciton localization. Because the optical activity of the 2.7 THz mode requires the C-centering translation to be broken at the excitation site, its emergence is intrinsically linked to exciton self-trapping. Its follows the sign inversion of the low-frequency background and builds up over the same ps window as the localized dielectric state (Figure 3C). Moreover, in contrast to the 5.7 THz mode, the 8.0 THz Ni–S stretching mode is suppressed during exciton formation and remains renormalized thereafter, reflecting a persistent local distortion of NiS octahedra. Furthermore, the combined quantum energy of the two modes (44 meV) lies within the uncertainty of ( meV), consistent with a multi-phonon contribution to self-trapping, as discussed below.
Discussion
The present measurements reveal a nonequilibrium pathway through which photoexcited charge-transfer carriers in NiPS evolve from an initially itinerant state into a localized self-trapped exciton. As illustrated by the configuration coordinate picture in Figure 4, excitation across the ligand-to-metal charge-transfer (LMCT) gap generates high-energy carriers that rapidly bind into excitons and subsequently localize, accompanied by substantial lattice renormalization. The localization barrier meV is consistent with an elastic barrier separating delocalized and self-trapped excitonic configurations within the Rashba–Toyozawa framework of exciton self-trapping [31]. Taken together, the OPTP, TAS, and PL measurements indicate that charge-transfer excitations do not simply relax through a conventional energy dissipation pathway, but instead funnel into a self-trapped excitonic configuration stabilized by strong exciton-phonon coupling.

Figure 4. Configuration-coordinate model for exciton self-trapping in . Schematic potential-energy surfaces along the generalized Ni–S lattice displacement coordinate , showing the correlated ground state (GS, dark grey), charge-transfer hot carriers (HC, red), free exciton (FE, orange), and self-trapped exciton (STE, dark red). Vertical arrows denote the initial 1.91 eV excitation and the eventual radiative recombination. The extracted activation barrier meV sets the kinetic threshold for spatial localization. The wavy red arrows indicate the dual radiative recombination pathways originating from both the free () and self-trapped () excitons. The bottom panels schematically illustrate the microscopic structural evolution, tracking the progressive distortion of the local octahedra during the polaronic collapse.
The central mechanistic implication of these observations is a cooperative two-phonon contribution of the 2.7 and 8.0 THz modes to exciton self-trapping. In the photoinduced transient state, both modes transform as odd-parity , such that their linear coupling to a parity-definite free exciton vanishes () and neither alone can produce the symmetric deformation that defines a localization potential. By contrast, their direct product () yields the quadratic term , which describes a cooperative distortion compatible with the lattice relaxation that stabilizes self-trapped excitonic states. Because this quadratic term combines the zone-boundary 2.7 THz mode with the zone-center 8.0 THz mode, it conserves crystal momentum when localization removes the C-centering translation. This is reflected in the measured dynamics, in which the 8.0 THz mode is renormalized during exciton formation, whereas the 2.7 THz mode appears alongside the negative dielectric response that marks localization. Their combined quantum energy also falls within the uncertainty of and is compatible with this cooperative coupling. Although the THz overtone () satisfies the same symmetry and energy criteria, the prompt kinetic decoupling and steady amplitude (Figure 3C) point to its non-dominant participation (Supplementary Note 8. and Note 9.). Ultimately, the convergence of energetic, dynamical and parity constraints supports a cooperative two-phonon mechanism underlying self-trapped exciton formation in NiPS.
The present results provide a unified framework for understanding the diverse nonequilibrium states reported previously in NiPS. Earlier ultrafast studies primarily employed resonant or near-resonant excitation within the low-energy excitonic manifold, where photoexcitation directly addresses preexisting Zhang–Rice-like excitons and deposits little excess energy into the lattice [14, 15]. Under these conditions, transient metallic behavior, population inversion, and long-lived nonequilibrium states have been reported. Excitation at 1.91 eV, by contrast, drives the inter-site charge-transfer transition, generating high-energy unbound carriers and producing substantial charge perturbation of the local Ni–S bonding network. This excess energy enables activation of lattice coordinates that are not efficiently accessed under resonant excitonic pumping and promotes rapid carrier localization into self-trapped excitons. These observations indicate that the contrasting non-adiabatic evolution of NiPS arises not from conflicting interpretations of the same excitation regime, but from fundamentally different photoexcitation pathways that access distinct regions of the coupled charge-lattice energy landscape.
A key enabling aspect of this work is the use of ultrabroadband terahertz spectroscopy. Extending the probe bandwidth to 15 THz provides simultaneous access to the emergent 2.7 THz vibrational modes and the intrinsic 8.0 THz Ni–S optical phonons that are involved in energy dissipation and lattice relaxation. This capability enables direct spectro-temporal tracking of the lattice coordinates involved in exciton localization and reveals correlations that would be inaccessible using conventional narrow-band THz probes. More broadly, the results highlight the importance of mode-selective exciton-phonon coupling in correlated van der Waals materials. The identification of specific lattice coordinates that mediate excitonic localization opens new opportunities to steer carrier self-trapping and manipulate emergent quantum states in low-dimensional materials through selective vibrational excitation with intense terahertz or mid-infrared fields.
Materials and Methods
Sample synthesis and characterization
Single crystals of NiPS were synthesized by chemical vapor transport (CVT). Stoichiometric amounts of Ni, P, and S powders (molar ratio 1:1:3) together with 0.15 g of I transport agent were sealed in an evacuated quartz ampoule. Crystal growth was carried out in a two-zone furnace with source and growth-zone temperatures of 973 K and 923 K, respectively, for approximately two weeks. The furnace was subsequently cooled to room temperature at a rate of 100 K h. For optical measurements, thin platelet-like crystals with typical dimensions of $2 \times 2 \times 0.09\ \mathrm{mm}^{3}$ were mechanically exfoliated from the bulk material to obtain optically flat surfaces. Magnetic characterization was performed using a Quantum Design Magnetic Property Measurement System (MPMS-5T). The Néel temperature was determined from the maximum in , yielding $T_{\mathrm{N}} \approx 155 K, consistent with previous reports for high-quality single crystals. Crystal structure and phase purity were further confirmed by X-ray diffraction (Supplementary fig. S1).
Ultrabroadband optical-pump terahertz-probe spectroscopy
Time-resolved optical-pump terahertz-probe (OPTP) measurements were performed using an air-plasma-based terahertz spectrometer. Ultrabroadband THz probe pulses were generated by two-color laser-induced plasma filamentation in nitrogen using 800 nm pulses (40 fs) from a Ti:sapphire regenerative amplifier and their second harmonic generated in a -barium borate (BBO) crystal. The THz electric fields were detected using air-biased coherent detection (ABCD), yielding phase-resolved measurements with spectral bandwidths exceeding 20 THz and an overall temporal resolution of better than 60 fs. The combination of ultrabroad spectral coverage and ultrafast temporal resolution enabled simultaneous tracking of the transient low-frequency vibrational modes together with the intrinsic optical phonons at 5.7 and 8.0 THz that participate in the localized dynamics discussed in the main text. Photoexcitation was provided by an optical parametric amplifier driven by the regenerative amplifier. Unless otherwise stated, the excitation pulse was fixed at 650 nm central wavelength (1.91 eV) to selectively drive the ligand-to-metal charge-transfer transition in .
All OPTP measurements were performed in reflection geometry (RL path in Figure 2A) to preserve the full detection bandwidth by avoiding strong attenuation of the THz probe by the infrared-active phonons of in transmission measurements. One-dimensional dynamics were recorded by scanning the pump–probe delay while fixing the THz sampling position at the peak of the reflected THz waveform. Two-dimensional spectro-temporal measurements were obtained by independently scanning both the pump–probe and THz sampling delays, enabling extraction of the transient complex dielectric function. Temperature-dependent measurements were performed between 75 and 300 K using a continuous-flow cryostat (MicrostatHe, Oxford Instruments) equipped with CVD diamond windows to minimize THz absorption and dispersion across the measured spectral range.
Absorption, photoluminescence, and transient absorption spectroscopy
Optical absorption spectroscopy. Steady-state optical spectra were measured using a UV–Vis–NIR spectrophotometer (Cary 5000, Varian) over the wavelength range of 200–1500 nm. Measurements were performed in diffuse-reflectance geometry to enhance sensitivity to weak optical transitions, including the parity-forbidden crystal-field (–) excitations of .
Photoluminescence spectroscopy. Photoluminescence (PL) measurements were carried out using a 637 nm continuous-wave laser for excitation. The emitted PL was collected in a backscattering geometry, dispersed by a Czerny–Turner spectrometer (SR500i, Andor), and detected using a cooled InGaAs array detector (DU491A-1.7, Andor).
Transient absorption spectroscopy. Room-temperature transient absorption spectroscopy (TAS) measurements were performed using femtosecond pump pulses centered at 592 nm (2.09 eV) with a fluence of 1.5 mJ . The excitation conditions were chosen to generate a carrier density comparable to that used in the OPTP experiments, enabling direct correlation between the excited-state population dynamics measured by TAS and the transient electrodynamic response obtained from terahertz spectroscopy.
Acknowledgments
Funding: This work was partially supported by the Independent Research Fund Denmark (Grant Nos. 2035-00365B and 6211-00182B), the Novo Nordisk Foundation (Grant No. NNF24OC0094858), and the Villum Fonden (VIL88455). X.H. and S.X. acknowledge support from Villum Fonden (VIL70202, VIL78289). X.L. and Y.S. acknowledge support from the National Key Research and Development Program of China (Grants No. 2021YFA1600200 and No. 2023YFA1607402) and the National Natural Science Foundation of China (Grant No. 12274412).
Author contributions: L.C. and B.Z. conceived and coordinated the study. L.C. and B.Z. designed and performed the THz spectroscopy experiments and analyzed the experimental data, with assistance from A.H.O., O.N., and P.U.J.; G.Li and K.Z. performed the transient absorption spectroscopy measurements, with K.Z. also contributing to data analysis and optical absorption spectroscopy. Q.P. contributed to the optical absorption spectroscopy measurements. X.H. and S.X. performed sample exfoliation and photoluminescence measurements. L.C. led the density functional theory calculations, with substantial contributions to the theoretical modeling and analysis from S.M., C.W., and Y.W.; X.L., Y.S., and Y.Z. grew the single crystals and performed the X-ray diffraction measurements. S.Z., G.Liao, and Y.L. contributed to scientific discussions, interpretation of the results, and critical review of the manuscript. L.C. and B.Z. prepared the initial manuscript, with input from all authors. B.Z. and P.U.J. supervised the project. All authors reviewed and approved the final manuscript.
Competing interests: There are no competing interests to declare.
Data, code and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials.
Supplementary materials
Supplementary Text
Figs. S1 to S8
References (32-41)
Supplementary Materials for Charge-transfer excitation drives cooperative phonon-assisted exciton self-trapping in NiPS₃
Long Cheng; Alexander Holm Ohrt; Oliver Nagy; Yu Zhao; Chenyu Wang; Yaxian Wang; Sheng Meng; Guohui Li; Qinying Pan; Xingyu Huang; Sanshui Xiao; Sijie Zhang; Guoqian Liao; Yutong Li; Xuan Luo; Yuping Sun; Kaibo Zheng; Peter Uhd Jepsen; Binbin Zhou
Supplementary Text
Structural Characterization of NiPS Single Crystals
Structural integrity and phase purity. To assess the crystalline quality and phase purity of the NiPS single crystals synthesized by chemical vapor transport (CVT), X-ray diffraction (XRD) measurements were performed on the mechanically exfoliated flakes. As shown in Figure S1, the diffraction pattern is dominated by the (00) family of Bragg reflections, spanning the (001) to (005) peaks, consistent with the layered crystal structure and the preferred orientation of the basal plane. No impurity-related diffraction peaks are observed, confirming the high phase purity of the synthesized crystals. The optical microscopy image (Figure S1, right inset) reveals a smooth and uniform surface obtained by cleavage along the van der Waals basal plane. This absence of visible surface damage or large-scale morphological defects indicates the excellent surface quality and provides an optically flat interface for ultrafast spectroscopy measurements.

Figure S1. X-ray diffraction and surface morphology of the NiPS single crystal. X-ray diffraction measured from the exfoliated basal plane exhibits exclusively Bragg reflections, confirming the high crystallinity and phase purity of the crystal. Insets: Photograph of a representative bulk NiPS (left) and a optical microscope image of the mechanically exfoliated basal plane (right), revealing a smooth, optically flat, and highly reflective surface.
Electronic Band Structure and Orbital Hybridization
Orbital-projected electronic structure and density of states The electronic structure of NiPS was calculated within density functional theory (DFT) using the Vienna Ab initio Simulation Package (VASP) [32]. To capture the correlated electronic states and charge-transfer-insulator character of NiPS, structural optimization and electronic structure calculations were performed using the strongly constrained and appropriately normed (SCAN) meta-generalized gradient approximation (meta-GGA) [18, 33]. Previous studies have shown that the SCAN functional reproduces both the experimental band gap and the localized electronic states of this correlated van der Waals antiferromagnet [16].
Figure S2 shows the orbital-projected band structure (left panel) and corresponding partial density of states (PDOS, right). The projections reveal the characteristic charge-transfer electronic structure of NiPS. The valence band maximum (VBM, aligned to 0 eV) consists predominantly of strongly hybridized S and Ni states, as evidenced by the substantial overlap of the corresponding orbital weights. Deeper valence bands originate primarily from non-bonding sulfur states. This electronic configuration is consistent with a correlated ground state containing mixed and character, where denotes a ligand hole on the sulfur sublattice. In contrast, the conduction band minimum (CBM) is dominated by localized Ni states, primarily of character, with comparatively weaker contributions from sulfur. This orbital configuration provides the microscopic basis for the ligand-to-metal charge-transfer (LMCT) excitation employed in the ultrafast measurements.
Optical excitation at 1.91 eV promotes electrons from the hybridized S –Ni states near the VBM into unoccupied Ni () states at the CBM. Because the lobes of the orbitals are directed toward the surrounding sulfur ligands within the NiS octahedra, this excitation induces substantial redistribution of charge density along the Ni–S bonds and strongly perturbs the local bonding environment. Furthermore, the localized character of the Ni states gives rise to a relatively narrow conduction bandwidth. Within strong-coupling polaron models, the effective electron–phonon coupling strength increases as the electronic bandwidth decreases [24]. The calculated effective conduction bandwidth ( meV) inherently limits carrier itinerancy. Consequently, photoexcitation into the Ni manifold is expected to enhance carrier–lattice interactions and promote lattice-assisted carrier localization. These considerations provide a microscopic framework for understanding the pronounced phonon renormalization and exciton localization dynamics observed in the time-resolved measurements.
Crystal-field Splitting and Steady-State Optical Transitions
Crystal-field splitting and charge-transfer transitions The electronic structure and optical response of NiPS are governed by the local coordination environment of the Ni () cations. As illustrated in Figure S3 A, the intrinsic trigonal distortion lowers the local symmetry from to , lifting the degeneracy of the crystal-field levels. As a result, the low-lying manifold splits into the and states, giving rise to the weak absorption features near ∼0.9 and ∼1.1 eV, respectively. Because these intra-atomic transitions are parity-forbidden by the Laporte selection rule, they exhibit relatively small oscillator strengths. These weak features are resolved because the spectra were acquired in diffuse-reflectance geometry on cleaved bulk crystals, where multiple scattering lengthens the effective optical path; spectra are reported as instrument absorbance, , with the specular component included. Since is bounded below by the front-surface Fresnel reflectivity once the crystal becomes opaque, strongly absorbing bands are compressed while weak sub-gap features are retained, so the weight of the crystal-field bands relative to the charge-transfer continuum in Figure S3B represents an upper bound. In contrast, the spectrum between 1.2 and 2.1 eV is dominated by a broad absorption continuum that can be decomposed into four overlapping components (Figure S3B). The feature centered near ∼1.4 eV is assigned to the Zhang–Rice-like excitonic resonance, whereas the broad continuum near ∼1.6 eV originates from the ligand-to-metal charge-transfer (LMCT) transition (). The onset of this charge-transfer excitation shifts from approximately 1.8 eV at low temperatures down to ∼1.6 eV at room temperature, reflecting the influence of lattice expansion and electron–phonon coupling on the electronic structure [34].

Figure S3. Crystal-field splitting and room-temperature optical absorption of NiPS3. (A) Schematic illustration of the local NiS octahedral coordination and the associated crystal-field splitting of the Ni () multiplet states. The intrinsic trigonal distortion lowers the local symmetry from to , lifting the degeneracy of the crystal-field levels and giving rise to the additional splitting observed in the optical spectra. (B) Room-temperature optical absorption spectrum of NiPS together with a multi-peak decomposition. Weak absorption features at 0.9 and 1.1 eV are assigned to the parity-forbidden on-site transitions from the ground state to the split and levels derived from the manifold. The broad absorption continuum spanning 1.2–2.1 eV is decomposed into four components: an excitonic resonance centered near 1.4 eV (pink), a ligand-to-metal charge-transfer (LMCT) continuum near 1.6 eV (green), and higher-energy crystal-field excitations near 1.78 eV (cyan) and 2.0 eV (purple) associated with states derived from the manifold. Additional features near 2.6 and 4.0 eV are attributed to higher-lying crystal-field excitations and charge-transfer transitions [14].
Two additional absorption features appear near ∼1.78 eV and ∼2.0 eV and are assigned to crystal-field excitations derived from the split manifold. Compared with the lower-energy transitions, these excitations exhibit substantially larger oscillator strengths and broader linewidths. Such enhancement is commonly attributed to intensity borrowing from the nearby optically allowed charge-transfer continuum [35, 36]. Because these states lie within or close to the LMCT absorption band, hybridization between sulfur and nickel orbitals partially relaxes the dipole-selection constraints and enhances their optical activity [6]. The corresponding linewidth broadening is consistent with stronger coupling to the surrounding electronic and lattice continua. This optical landscape is particularly relevant for the ultrafast measurements presented in the main text.
Excitation at 1.91 eV directly accesses the LMCT continuum (Figure S3B), generating high-energy charge-transfer carriers within a strongly hybridized electronic manifold. Although the split states lie nearby, their oscillator strengths arise from mixing with the allowed charge-transfer continuum rather than from the intra-atomic configuration. Moreover, a purely intra-atomic excitation redistributes charge within a single Ni site and is charge-neutral and cannot generate the mobile-carrier response that dominates the transient terahertz signal at early delays. Any residual crystal-field absorption at 1.91 eV would therefore open a parallel, non-conductive relaxation channel contributing neither to the measured conductivity nor to the subsequent localization. The associated redistribution of charge density along the Ni–S bonds provides favorable conditions for strong carrier–lattice coupling and the subsequent localization dynamics observed in the time-resolved experiments.
Phonon Dispersion and Projected Density of States
Phonon dispersion and density of states. Phonon dispersion relations and the corresponding density of states (DOS) were calculated using the finite-displacement method as implemented in the VASP and Phonopy packages [37]. The interatomic force constants were obtained from a supercell. Electronic exchange and correlation were treated using the Perdew–Burke–Ernzerhof (PBE) functional with a Hubbard correction of eV applied to the Ni states in order to reproduce the localized electronic character and zigzag antiferromagnetic ground state of NiPS.
The calculated atom-projected phonon dispersion identifies the microscopic origin of the lattice modes observed experimentally. The branch at THz corresponds to the infrared-active mode and is dominated by vibrations of the Ni atoms within the NiS octahedra [25]. In contrast, the optical phonon branches near THz are primarily associated with sulfur displacements. This frequency range contains the infrared-active mode arising from Davydov splitting of the corresponding vibration [26]. The mode involves collective sulfur motions that modulate the Ni–S octahedral environment and dynamically perturb the crystal-field splitting of the Ni orbitals [13, 26, 27]. Such lattice distortions are expected to couple efficiently to charge-transfer excitations because they directly modify the local Ni–S bonding geometry.
Beyond the zone-center infrared-active modes, the calculations reveal a weakly dispersive phonon branch near THz along several zone-boundary directions, including the – and – paths. The small group velocity of these modes gives rise to an enhanced phonon density of states near this frequency, characteristic of a flat phonon branch. The atom-projected DOS indicates that these vibrations involve coupled Ni–S displacements. In the C-centered NiPS lattice, the zone-boundary phonons at the point are antisymmetric under the centering translation, so the dipoles of translation-related atoms cancel and leave these modes dark in equilibrium. A localized exciton could remove this translation at its site, making the two centering-related halves of the cell inequivalent and rendering these modes optically active. Within the subgroup that retains inversion through the midpoint of a nearest-neighbor Ni–Ni bond, the THz mode transforms as . The dipole of each mode scales with the relative displacement between these two halves in its equilibrium eigenvector. By this measure, the THz mode ranked first among the 30 Y-point modes. Due to the centering translation belonging to the crystal structure rather than the magnetic order, the same activation operates at 300 K, above the Néel temperature.
To explicitly investigate the origin of the transiently activated modes, we calculated the structural and vibrational response of the charge-transfer excited state. Promoting an electron from the hybridized valence-band maximum to the localized Ni conduction-band minimum induces a localized structural relaxation. Rather than a macroscopic phase transition, this local distortion microscopically mimics a primitive-cell doubling within the spatial extent of the exciton. By folding the phonon spectrum of this extended cell onto the equilibrium Brillouin zone (Figure S4B), the origin of the emergent vibrational branches could be identified through their spectral weights. The calculations reveal that the transiently emergent modes originate from weakly dispersive zone-boundary phonons and acquire transient optical activity. For instance, the emergent 4.5 THz mode is attributed to a second zone-boundary branch, in which Ni displacements couple to the bipyramidal units. This branch is activated by the same loss of the C-centering translation, but less efficiently than the 2.7 THz case. Notably, the 2.7 THz branch exhibits a substantially stronger transient dielectric response than would be expected solely from its phonon density of states. This observation suggests that the nonequilibrium terahertz response is governed not only by the available vibrational phase space, but also by the strength of the coupling between specific lattice coordinates and the transient charge-transfer dipole. The strong Ni–S participation of the 2.7 THz vibration therefore also makes it a particularly sensitive indicator of the lattice response during exciton localization.
Power-density Dependence of the PL Spectrum
High-density exciton dynamics and exciton–exciton annihilation. To further investigate the population dynamics at elevated excitation densities, we measured the excitation-power dependence of the steady-state photoluminescence (PL) spectra. As shown in Figure S5, the emission exhibits a crossover near an excitation power density of . Below this threshold, the integrated intensities of the 1.409 eV () and 1.363 eV () emission features scale approximately linearly with excitation power, consistent with a regime dominated by independent excitonic recombination. In contrast, the low-energy emission band near 1.295 eV is negligible in this regime.

Figure S5. Power density dependence of the photoluminescence emission in NiPS. (A) Log–log plot of the integrated photoluminescence (PL) intensity of the 1.295 eV emission band as a function of excitation power density. (B) Corresponding log–log plots of the integrated intensities of the 1.409 eV () and 1.363 eV () excitonic emission features. Above an excitation threshold of approximately 11 kW/cm, both excitonic peaks exhibit pronounced sublinear behavior, consistent with the onset of exciton–exciton interactions and saturation effects. Concurrently, the 1.295 eV emission band emerges and follows an approximately power-law dependence over the measured excitation range. Black dashed lines indicate fits to the corresponding power-law scaling.
Above this threshold, both excitonic emission features exhibit pronounced sublinear behavior, indicating the onset of density-dependent exciton interactions. Such behavior is consistent with exciton–exciton annihilation (EEA), which becomes increasingly important as the average exciton separation decreases. Within a simple rate-equation framework, the exciton population density obeys
where is the excitation efficiency, is the pump power density, and is the EEA coefficient. Under steady-state conditions (), the exciton density scales as .
Concurrently with the onset of EEA, the 1.295 eV emission band becomes clearly visible and exhibits an approximately power-law dependence over the measured excitation range up to 154 kW/cm. The correspondence between the observed scaling and the EEA-limited exciton population suggests that this low-energy emission is populated through density-dependent relaxation channels that become active at high excitation densities. While the microscopic origin of this feature cannot be uniquely determined from the present measurements, its emergence only in the high-density regime is consistent with a localized excited state that is efficiently fed by exciton–exciton interactions and subsequent energy relaxation processes.
Excitation-energy Dependence and Thermally Activated Localization Dynamics
Optical excitation at 800 nm (1.55 eV) predominantly addresses the Zhang–Rice excitonic resonance and deposits relatively little excess energy into the lattice. Under these conditions, the transient response remains largely positive and carrier-like throughout the measured time window. In contrast, excitation at 650 nm (1.91 eV) directly accesses the ligand-to-metal charge-transfer (LMCT) continuum, generating initially unbound carriers and depositing substantially greater excess energy into the coupled charge–lattice system. The resulting dynamics evolve toward a localized dielectric state, highlighting the strong dependence of the nonequilibrium response on the excitation pathway.
The temperature dependence of the transient dynamics provides further insight into the localization process. Both the formation rate of the negative transient response () and the amplitude of the localized component () are well described by Arrhenius-type expressions, and , where is the build-up rate of the negative response following exciton formation, with and denoting temperature-independent offsets [38]. Fitting (upper panel of Figure S6B) and (Figure 2E, inset) yields meV and meV, respectively. The two energies are numerically close but physically distinct, with setting the activation barrier that governs the localization rate and describing the thermally activated growth of the localized population.
The subsequent recovery of the negative transient response reflects the depletion of the localized excitonic population. Unlike the initial localization process, this slower relaxation involves multiple competing channels, including thermal detrapping, non-radiative recombination, and transfer to intermediate dark states. Consequently, the activation energy extracted from the recovery dynamics ( meV) should be regarded as an effective phenomenological energy scale rather than a unique microscopic detrapping barrier. This energy characterizes the thermal destabilization of the localized state and provides a measure of its stability against thermally activated relaxation processes. Given that is comparable to at room temperature, localization does not exhaust the free exciton population. Both configurations therefore remain populated and retain their own radiative channel, which accounts for the two excitonic emission features resolved in the steady-state PL spectrum (Figure 1D). Their relative weights are consistent with the fraction of the photoexcited population transferred into the long-lived localized state (Figure 2B). Such coexistence is well documented in molecular crystals [38].
Extraction of Equilibrium and Transient THz Optical Constants
Extraction of intrinsic terahertz optical constants via parameterized global forward modeling (PGFM). The extraction of complex optical constants from terahertz time-domain spectroscopy (THz-TDS) is commonly performed through direct numerical inversion of the measured transfer function. For thin van der Waals crystals, however, multiple internal reflections generate overlapping Fabry–Pérot echoes that complicate this procedure. Conventional approaches often require temporal windowing of the time-domain waveform, which can reduce low-frequency spectral fidelity, or become susceptible to phase-unwrapping ambiguities near strong phonon resonances. To overcome these limitations, we employ the parameterized global forward modeling (PGFM) approach [39]. Rather than extracting the optical constants independently at each frequency, PGFM describes the dielectric response using a physically constrained model and determines the corresponding parameters through global optimization. The measured complex transfer function is reproduced by explicitly accounting for the coherent Fabry–Pérot reflections within the sample, thereby enabling a self-consistent determination of the equilibrium dielectric response across the full spectral range. For NiPS, the equilibrium dielectric function is represented by a hybrid Drude–Smith and double-Lorentz-oscillator model:
where denotes the high-frequency dielectric constant. The Lorentz oscillators describe the infrared-active phonons, with , , and corresponding to the oscillator strength, resonance frequency, and linewidth, respectively. The Drude–Smith contribution accounts for any residual carrier response in the equilibrium state.
The complex refractive index is obtained from . The corresponding complex transmission function is then constructed by explicitly including multiple internal reflections within a crystal of thickness ,
where and are the Fresnel transmission and reflection coefficients at the corresponding interfaces, and is the speed of light. This expression is equivalent to summing all coherent Fabry–Pérot contributions to infinite order. The model parameters are optimized by minimizing the difference between the simulated transfer function and the experimentally measured transfer function . To improve the robustness of the extraction, the fitting procedure was cross-validated using NiPS crystals of different thicknesses. The resulting optical constants reproduce both the amplitude and phase spectra across the full measured frequency range (Figure S7), yielding a self-consistent set of equilibrium refractive-index () and extinction-coefficient () spectra. These parameters serve as the reference optical constants for the perturbed transfer-matrix method (PTMM) analysis of the photoinduced terahertz response presented in the main text.

Figure S7. Extraction of equilibrium terahertz optical constants of NiPS₃ using parameterized global forward modeling (PGFM) [39]. (A) Transmittance and (B) phase-difference () spectra measured by steady-state THz time-domain spectroscopy at room temperature. Solid blue curves show the experimental data, including Fabry–Pérot interference arising from multiple internal reflections within the crystal. Dashed orange curves represent the corresponding PGFM fits obtained using a hybrid Drude–Smith and double-Lorentz-oscillator dielectric model. (C) Refractive index () and (D) extinction coefficient () extracted from the optimized dielectric function. By explicitly incorporating Fabry–Pérot reflections and simultaneously fitting the amplitude and phase response, PGFM provides a self-consistent description of the optical properties across the full measured spectral range, including the strong phonon resonances near 5.7 and 8.0 THz. The resulting equilibrium optical constants serve as the reference parameters for the perturbed transfer-matrix method (PTMM) analysis of the photoinduced terahertz response.
Extraction of the transient dielectric response with PTMM. Although the photoexcited layer depth of the 1.91 eV charge-transfer pump ( nm) [14] is much smaller than the terahertz probe wavelength and therefore formally satisfies the thin-film limit, the extraction of transient optical parameters remains complicated by the strong frequency dependence of the equilibrium refractive index near the infrared-active phonon resonances. Under these conditions, changes in the reflected THz field contain contributions from both the intrinsic photoinduced dielectric response and the electrodynamic propagation of the probe field within the excited region. To account for these effects, we employ the perturbed transfer-matrix method [40], which explicitly incorporates the depth-dependent excitation profile together with the measured equilibrium optical constants. This approach provides a self-consistent description of the interaction between the THz probe and the photoexcited layer, enabling reliable extraction of the transient dielectric function across the full measured spectral range. Within the PTMM formalism, the measured relative change in the reflected THz field, , is related to the complex photoconductivity through:
where is the complex photoconductivity at the crystal surface, is the equilibrium complex refractive index, is the impedance of free space, and is the speed of light. The exponential terms account for the depth-dependent carrier distribution generated by optical absorption and the phase accumulation of the terahertz field within the photoexcited region. Evaluation of the integral yields the transient complex photoconductivity, from which the pump-induced change in the dielectric function is obtained through:
where is the vacuum permittivity. The resulting transient dielectric spectra provide direct access to the spectro-temporal evolution of the electronic response and the phonon resonances discussed in the main text, enabling analysis of the coupling between photoexcited carriers and lattice vibrations.
Spectro-temporal Mapping and Supplementary Features
Spectro-temporal mapping and supplementary phonon dynamics. Figure S8A, B presents the two-dimensional spectro-temporal maps of the pump-induced changes in the reflected terahertz electric field () measured at 300 K and 75 K following 1.91 eV charge-transfer excitation. By applying PTMM to the experimental (Supplementary Note 7.) and subtracting the divergent low-frequency tail, we can extract the intrinsic, phonon-resolved transient dielectric response shown in Figure 3A, B. Because 75 K lies well below the Néel temperature of NiPS ( K), these measurements probe the nonequilibrium electrodynamic response within the zigzag antiferromagnetic phase. Despite the distinct magnetic environments, the overall spectro-temporal evolution remains qualitatively similar at both temperatures, indicating that the primary localization dynamics are robust against long-range magnetic ordering. In addition, the 75 K data reveal a long-lived, low-frequency oscillatory feature centered at THz, consistent with the previously reported antiferromagnetic magnon persisting up to [14].
The transient evolution of the intrinsic optical phonons (Figure 3) provides further insight into the lattice response following charge-transfer excitation. Following the peak itinerant carrier response at , photoexcited carriers condense into precursor bound excitons on a timescale of fs, driving the bound-state polarization recovery observed in . The blueshift of the 5.7 THz () and 8.0 THz () modes in the imaginary dielectric response () is triggered after this condensation window, occurring as the excitonic wavepacket undergoes spatial collapse. This dynamic stiffening reflects the non-adiabatic electrostrictive stress exerted on local Ni–S bonds during spatial wavepacket confinement. Throughout the remaining ps trajectory, the blueshifted frequencies persist without pronounced lattice softening, signaling a stabilized structural deformation that accommodates the localized state.
To microscopically isolate the roles of individual lattice coordinates in this polaronic collapse, we compare their mode-specific dielectric dynamics (Figure 3C and Figure S8C,D):
2.7 THz mode (): Coincident with the peak itinerant carrier response () and prior to mode activation, the imaginary dielectric component () exhibits a positive amplitude, reflecting conventional Drude-like dissipation. It crosses to negative values once the fs condensation window ends and deepens as the mode is activated during the ps localization.
Concurrently, the real component () climbs to a peak at ∼3 ps (a ∼90% increase relative to its itinerant-carrier peak). This electrodynamic crossover from itinerant absorption to a negative localized dielectric state marks the activation of the 2.7 THz mode as a spectroscopic indicator of exciton localization.
4.5 THz mode (): Although sharing the same folded zone-boundary origin and a similar zero-crossing transition with the 2.7 THz mode, this coordinate exhibits a weaker dielectric response. Its imaginary component () similarly transitions from positive Drude dissipation to a negative localized state over an identical timescale of ∼ fs, yet its negative excursion remains shallow. Concurrently, its real dielectric modulation () attains less than half the amplitude of the 2.7 THz mode. Lattice-dynamics calculations attribute this weaker response to the smaller dynamic dipole that this branch acquires when the C-centering translation is lost (Supplementary Note 4.).
5.7 THz mode (): Driven impulsively by the photoinduced dipolar field across the charge-transfer gap, this coordinate exhibits a steep real dielectric jump (), reaching its peak within the precursor exciton condensation window. This large response without progressive accumulation indicates prompt charge-transfer polarization but limited participation in the slower excitonic localization process. Moreover, its imaginary dielectric component () remains strictly positive without sign inversion, displaying a sub-100 fs transient drop and rapid upward turnaround (a kinetic bounce) before evolving into monotonic decay. This sustained positive response and kinetic bounce are consistent with continued energy exchange with the photoexcited manifold. The mode-selective kinetic bounce may originate from partial non-adiabatic reflection of the coherent lattice wavepacket involving an cross-coupling component. Accordingly, the 5.7 THz vibration plays a non-dominant role during spatial wavepacket confinement.
8.0 THz mode (): Its real dielectric component () rises promptly within ∼ fs but reaches only ∼33% of the 5.7 THz response. By contrast, bleaches by ∼70% over the same window and remains at a reduced plateau without recovery. The component-resolved dynamics distinguish this behavior from the 4.5 THz mode, whose real modulation and negative imaginary excursion are both weak. Their divergence shows that the 8.0 THz
oscillator is strongly perturbed despite its limited real modulation, supporting pinning within locally distorted octahedra. The sustained bleaching tracks a redistribution of oscillator spectral weight associated with changes in the local Ni–S restoring force rather than dissipative absorption. Occupation of antibonding Ni orbitals deforms the octahedra, and the 8.0 THz mode, which modulates the Ni–S bonds, tracks this persistent local distortion.
Parity Considerations for Cooperative Phonon-assisted Self-trapping
The temperature-dependent localization dynamics yield an activation energy of meV, establishing the characteristic energy scale associated with the formation of the localized state. This value exceeds the quantum energy of either the intrinsic 8.0 THz mode ( meV) or the transiently activated 2.7 THz mode ( meV) individually, but closely matches their combined energy ( meV). This observation motivates consideration of a cooperative multi-phonon pathway for exciton localization. Self-trapping requires the formation of a static, totally symmetric lattice distortion that lowers the electronic energy at the trapping site. This is the symmetry requirement, rather than any particular energy scale, that determines which lattice coordinates can participate. Microscopically, the exciton-lattice interaction is described by the electron-phonon () Hamiltonian. Expanding the interaction potential with respect to the normal mode lattice coordinates yields both linear (first-order) and quadratic (second-order) terms [41]:
where () creates (annihilates) a charge-transfer excitation, and and denote the linear and quadratic electron–phonon coupling coefficients, respectively. Within the subgroup of , both the intrinsic 8.0 THz Ni–S stretching mode and the transiently activated 2.7 THz mode transform as odd-parity . Because the photoinduced distortion is local and perturbative, this classification remains a valid reference for identifying the modes. For a parity-definite free exciton, the diagonal matrix element associated with an isolated odd-parity distortion vanishes, . Consequently, a single phonon cannot by itself generate the totally symmetric lattice deformation associated with the formation of a static localization potential.
The prominence of the infrared-active phonons nevertheless reflects the nature of the charge-transfer excitation itself. The ligand-to-metal charge-transfer process (S 3 Ni 3) generates a transient redistribution of charge density and an associated local dipolar character. Odd-parity lattice vibrations efficiently modulate this transient polarization and are therefore expected to couple strongly to the photoexcited state. By contrast, Raman-active phonons correspond to fully symmetric distortions that primarily couple through deformation-potential interactions and do not directly modulate the transient charge-transfer dipole. Within this framework, self-trapping need not be governed by a single phonon coordinate. Instead, theoretical descriptions of self-trapped excitons frequently invoke cooperative lattice distortions involving multiple vibrational modes [38]. Group-theoretical considerations provide a natural illustration of such a mechanism. The direct product of two odd-parity representations transforms as a totally symmetric representation, . Accordingly, the quadratic coupling channel generates a non-vanishing diagonal matrix element, , allowing two odd-parity distortions to cooperate in constructing a symmetry-allowed localization potential. Both modes deform the Ni–S bonds of the NiS octahedra, so the quadratic term acts on the local unit that hosts the charge-transfer exciton. Combining the zone-boundary 2.7 THz mode with the zone-center 8.0 THz mode, this term conserves crystal momentum once localization removes the C-centering translation.
Parity alone, however, is insufficient to uniquely identify the relevant quadratic channel because the same symmetry condition can also be satisfied by overtone processes (). The experimentally determined activation barrier () provides the additional constraint. Among the two-phonon combinations of the observed modes, 2.7+8.0 THz, 4.5+5.7 THz, and 25.7 THz are energetically compatible with meV (Supplementary Note 6). The remaining combinations lie outside this range. This energy screen defines the competing channels against which the dominant localization pathway can be assessed. The overtone of the 5.7 THz mode ( meV) formally satisfies both the totally symmetric parity requirement () and the energetic budget. However, as established by the mode-resolved dynamics in Supplementary Note 8, the 5.7 THz mode makes a transient, non-dominant contribution to the photoinduced lattice response. By contrast, the 8.0 THz mode exhibits persistent renormalization of the local Ni–S restoring force, while the 2.7 THz response evolves over the localization timescale. The 5.7 THz mode therefore contributes to early energy redistribution, but combinations containing it are unlikely to provide the sustained structural component that dominates localization. For the THz channel, this limitation is reinforced by the weaker activation of the 4.5 THz branch (Supplementary Note 4). Together with their combined energy, these observations support assignment of the THz pair as the dominant cooperative contribution to the localization coordinate. Within the point group associated with the subgroup, provides symmetry support for this cooperative pair.
The weak participation of the 5.7 THz overtone channel may further reflect non-adiabatic coordinate redirection, as suggested by its mode-resolved dielectric signature (Supplementary Note 8). During the initial wavepacket collapse, cross-coupling between simultaneously driven and coordinates yields an asymmetric shear distortion () that provides no polaronic stabilization. The coherent wavepacket along the coordinate may therefore be reflected, consistent with the sub-100 fs kinetic bounce in its response, which decouples the 5.7 THz mode from the subsequent structural relaxation.
References
- [1]D. N. Basov, R. D. Averitt, D. Hsieh, Towards properties on demand in quantum materials. Nat. Mater. 16 (11), 1077–1088 (2017).DOI
- [2]A. S. Disa, T. F. Nova, A. Cavalleri, Engineering crystal structures with light. Nat. Phys. 17 (10), 1087–1092 (2021).DOI
- [3]A. de la Torre, et al., Colloquium: Nonthermal pathways to ultrafast control in quantum materials. Rev. Mod. Phys. 93 (4), 041002 (2021).DOI
- [4]C. Franchini, M. Reticcioli, M. Setvin, U. Diebold, Polarons in materials. Nat. Rev. Mater. 6 (7), 560–586 (2021).DOI
- [5]E. Baldini, et al., Exciton control in a room temperature bulk semiconductor with coherent strain pulses. Sci. Adv. 5 (11), eaax2937 (2019).DOI
- [6]S. Kang, et al., Coherent many-body exciton in van der Waals antiferromagnet NiPS₃. Nature 583 (7818), 785–789 (2020).DOI
- [7]B. Huang, et al., Layer-dependent ferromagnetism in a van der Waals crystal down to the monolayer limit. Nature 546 (7657), 270–273 (2017).DOI
- [8]Y. J. Bae, et al., Exciton-coupled coherent magnons in a 2D semiconductor. Nature 609 (7926), 282–286 (2022).DOI
- [9]K. S. Burch, D. Mandrus, J.-G. Park, Magnetism in two-dimensional van der Waals materials. Nature 563 (7729), 47–52 (2018).DOI
- [10]J. Pei, J. Yang, T. Yildirim, H. Zhang, Y. Lu, Many-Body Complexes in 2D Semiconductors. Adv. Mater. 31 (2), 1706945 (2019).DOI
- [11]M. Piacentini, F. S. Khumalo, C. G. Olson, J. W. Anderegg, D. W. Lynch, Optical transitions, XPS, electronic states in NiPS₃. Chem. Phys. 65 (3), 289–304 (1982).DOI
- [12]K. Hwangbo, et al., Highly anisotropic excitons and multiple phonon bound states in a van der Waals antiferromagnetic insulator. Nat. Nanotechnol. 16 (6), 655–660 (2021).
- [13]E. Ergeçen, et al., Magnetically brightened dark electron-phonon bound states in a van der Waals antiferromagnet. Nat. Commun. 13 (1), 98 (2022).DOI
- [14]C. A. Belvin, et al., Exciton-driven antiferromagnetic metal in a correlated van der Waals insulator. Nat. Commun. 12 (1), 4837 (2021).
- [15]J. A. Warshauer, et al., Long-lived population inversion in resonantly driven excitonic antiferromagnet. Phys. Rev. Lett. 134 (1), 016901 (2025).DOI
- [16]C. Lane, J.-X. Zhu, An ab initio study of electron-hole pairs in a correlated van der Waals antiferromagnet: NiPS₃ (2022), arXiv:2209.13051.arxiv.org/abs/2209.13051
- [17]Y. Li, G. Liang, C. Kong, B. Sun, X. Zhang, Charge-Transfer-Mediated Exciton Dynamics in van der Waals Antiferromagnet NiPS₃. Adv. Funct. Mater. 34 (40), 2402161 (2024).DOI
- [18]C. Lane, J.-X. Zhu, Thickness dependence of electronic structure and optical properties of a correlated van der Waals antiferromagnetic NiPS₃ thin film. Phys. Rev. B 102 (7), 075124 (2020).DOI
- [19]D. Afanasiev, et al., Controlling the anisotropy of a van der Waals antiferromagnet with light. Sci. Adv. 7 (23), eabf3096 (2021).
- [20]T. Klaproth, et al., Origin of the Magnetic Exciton in the van der Waals Antiferromagnet NiPS₃. Phys. Rev. Lett. 131 (25), 256504 (2023).DOI
- [21]S. Y. Kim, et al., Charge-Spin Correlation in van der Waals Antiferromagnet NiPS₃. Phys. Rev. Lett. 120 (13), 136402 (2018).DOI
- [22]X. Wang, et al., Electronic Raman scattering in the 2D antiferromagnet NiPS₃. Sci. Adv. 8 (2), eabl7707 (2022).DOI
- [23]S. Fang, et al., Pressure-induced enhancement and retainability of optoelectronic properties of NiPS₃. Mater. Res. Lett. 11 (2), 134–142 (2023).
- [24]A. S. Alexandrov, ed., Polarons in advanced materials, vol. 103 of Springer Series in Materials Science (Springer, Dordrecht) (2007).DOI
- [25]N. Pistawala, et al., Probing electron–phonon coupling in magnetic van der Waals material NiPS₃: A non-magnetic site-dilution study. 2D Mater. 11 (2), 025035 (2024).DOI
- [26]M. H. Nguyen, G. Park, J.-G. Park, H. Cheong, Interlayer interaction and Davydov splitting in antiferromagnetic few-layer NiPS₃. npj Quantum Mater. 10 (1), 65 (2025).DOI
- [27]Z. Muhammad, et al., Anisotropic phonon and magnon vibration and gate-tunable optoelectronic properties of nickel thiophosphite. 2D Mater. 10 (2), 025001 (2023).DOI
- [28]A. Hashemi, H.-P. Komsa, M. Puska, A. V. Krasheninnikov, Vibrational Properties of Metal Phosphorus Trichalcogenides from First-Principles Calculations. J. Phys. Chem. C 121 (48), 27207–27217 (2017).DOI
- [29]C. Tovatello, et al., The ultrafast onset of exciton formation in 2D semiconductors. Nat. Commun. 11 (1), 5277 (2020).
- [30]P. Steinleitner, et al., Direct observation of ultrafast exciton formation in a monolayer of WSe₂. Nano Lett. 17 (3), 1455–1460 (2017).
- [31]M. Ueta, H. Kanzaki, K. Kobayashi, Y. Toyozawa, E. Hanamura, Theory of Excitons in Phonon Fields, in Excitonic Processes in Solids (Springer Berlin Heidelberg, Berlin, Heidelberg), vol. 60 of Springer Series in Solid-State Sciences, pp. 203–284 (1986).DOI
- [32]G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169–11186 (1996).pubmed.ncbi.nlm.nih.gov/9984901
- [33]J. Sun, A. Ruzsinszky, J. P. Perdew, Strongly Constrained and Appropriately Normed Semilocal Density Functional. Phys. Rev. Lett. 115, 036402 (2015).arxiv.org/abs/1504.03028
- [34]F. Wang, et al., New Frontiers on van der Waals Layered Metal Phosphorous Trichalcogenides. Adv. Funct. Mater. 28 (37), 1802151 (2018).DOI
- [35]A. B. P. Lever, Inorganic Electronic Spectroscopy, vol. 33 of Studies in physical and theoretical chemistry (Elsevier), 2nd ed. (1984).
- [36]P. A. Joy, S. Vasudevan, Optical-absorption spectra of the layered transition-metal thiophosphates MPS₃ (M=Mn, Fe, and Ni). Phys. Rev. B 46, 5134–5141 (1992).pubmed.ncbi.nlm.nih.gov/10004288
- [37]A. Togo, I. Tanaka, First principles phonon calculations in materials science. Scr. Mater. 108, 1–5 (2015).arxiv.org/abs/1506.08498
- [38]K. S. Song, R. T. Williams, Self-trapped excitons (Springer, Berlin), 2nd ed. (1996).
- [39]D. Stock, A. K. Wigger, P. Haring Bolívar, Reducing errors in THz material parameter determination by model-based time-domain extraction methods. J. Opt. Soc. Am. B 38 (3), 815–824 (2021).DOI
- [40]H.-K. Nienhuys, V. Sundström, Intrinsic complications in the analysis of optical-pump, terahertz probe experiments. Phys. Rev. B 71, 235110 (2005).DOI
- [41]G. D. Mahan, Many-particle physics (Kluwer Academic/Plenum Publishers, New York), 3rd ed. (2000).