Time-Resolved Resonant Inelastic X-ray Scattering

• Time-resolved RIXS is an advanced spectroscopic technique that probes dynamic molecular and quantum excitations using femtosecond pump–probe methods.
• It achieves element-specific insights by tuning to core-level edges, revealing spin, orbital, vibrational, and charge-transfer channels with precise energy and momentum resolution.
• The method overcomes stationary limitations by incorporating time-dependent intermediate-state dynamics and explicit four-time correlation functions.

Time-resolved resonant inelastic X-ray scattering (tr-RIXS) extends the resonant Raman-type, photon-in/photon-out X-ray spectroscopy into the femtosecond domain, enabling direct observation of pump-prepared, evolving molecular wavepackets as they are interrogated by a short, coherent X-ray probe, and, in quantum materials, measuring with simultaneous energy and momentum resolution how the spectrum of collective excitations evolves after ultrafast photoexcitation (Freibert et al., 2024, Cao et al., 2018). By tuning the probe X-ray energy to an element-specific absorption edge, tr-RIXS accesses spin, orbital, charge-transfer, lattice, excitonic, and vibronic channels, while the pump–probe delay resolves how intermediate-state dynamics, core-hole decay, and nonequilibrium correlations reshape the emitted spectrum (Cao et al., 2018, Mitrano et al., 2020).

1. Spectroscopic definition and observables

tr-RIXS records the energy loss and momentum transfer of scattered X-rays following resonant absorption and re-emission. The transferred variables are

$$\hbar \omega = \hbar \omega_{\mathrm{in}}-\hbar \omega_{\mathrm{out}}, \qquad \mathbf{q}=\mathbf{k}_{\mathrm{out}}-\mathbf{k}_{\mathrm{in}},$$

so the measured spectrum encodes the excitations created in the sample. In quantum materials, this enables mapping of dispersions and lifetimes of bosonic quasiparticles such as magnons, orbitons, phonons, and charge-transfer modes across large fractions of the Brillouin zone. In molecular systems, the same pump–probe logic is used to monitor photoinduced dynamics in an evolving rovibronic wavepacket rather than in a crystal momentum manifold (Cao et al., 2018, Freibert et al., 2024).

The method is resonant because the incident photon is tuned to a core-level edge. The intermediate core hole makes the response element and orbital specific, and the relevant dipole selection rules depend on the edge and polarization. At transition-metal L-edges, strong core-level spin–orbit coupling allows conversion of photon angular momentum into a spin flip in the valence shell, enabling direct access to single-magnon channels. At K-edges, indirect processes emphasize density-like operators and can enhance bimagnon or charge channels. At soft X-ray ligand edges such as O K or C K, the probe can emphasize hybridized ligand states, local symmetry breaking, and electron–phonon coupling (Cao et al., 2018, Paris et al., 2021, Malvestuto et al., 17 Apr 2025).

A central feature of the time-resolved modality is that the measured spectrum is no longer a stationary property. A pump pulse prepares a non-equilibrium state whose correlators evolve with delay. In materials language, the dynamical magnetic structure factor $S(\mathbf{q},\omega,t)$ and related correlation functions become explicit functions of pump–probe delay; in molecular language, the signal depends on evolving nuclear wavepackets in the ground, valence-excited, core-excited, and final valence manifolds (Cao et al., 2018, Freibert et al., 2024).

2. Cross section, time-domain theory, and limits of stationary formulations

The equilibrium baseline is the Kramers–Heisenberg expression,

$$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$

with $|g\rangle$, $|n\rangle$, and $|f\rangle$ the initial, intermediate, and final states, and $\Gamma_n$ the inverse lifetime of the core-hole intermediate state. In stationary applications this is commonly evaluated in an eigenstate basis and typically assumes a stationary initial state, weak and time-independent fields, core-hole lifetime entering as a Lorentzian broadening, and vibrational dynamics treated in the frequency domain. Those constraints limit the description of explicit pulse envelopes, ultrafast non-adiabatic dynamics, and correlated intermediate-state propagation (Freibert et al., 2024, Cao et al., 2018).

Time-resolved theory generalizes the cross section to multi-time correlation functions. In the time-dependent Kramers–Heisenberg formulation with a weak probe, the tr-RIXS intensity can be written as a four-time integral,

$$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$

where $g$ is the probe envelope and $l$ encodes the core-hole lifetime. The loss of time-translation invariance is therefore structural rather than cosmetic: the signal depends on excitation time, emission time, probe bandwidth, and explicit propagation in the intermediate manifold (Chen et al., 2019, Wang et al., 2019).

Several complementary time-domain frameworks have been developed. A fully time-dependent, wavepacket-based molecular formalism propagates nuclear wavepackets with ML-MCTDH on diabatic vibronic Hamiltonians and treats the Gaussian X-ray probe explicitly, yielding a Raman wavefunction that carries the core-excited propagation and decay (Freibert et al., 2024). A real-time scattering approach solves the time-dependent Schrödinger equation without explicit eigenstate summations, includes photons and core electrons, and naturally handles pump–probe sequences in large strongly correlated systems with time-dependent DMRG (2002.04142). A non-equilibrium DMFT formulation embeds the probe pulse directly in the impurity model, avoids explicit four-point correlation functions, and makes tr-RIXS accessible in multi-orbital settings, albeit in single-site form without momentum resolution (Eckstein et al., 2020).

In important limits the theory simplifies. In the ultrashort core-hole lifetime limit, the cross section can reduce to a probe-windowed two-point correlator of an effective operator; in magnetic Mott insulators, crossed polarizations can isolate spin channels, and under UCL the cross section reduces to a weighted spin correlation function. By contrast, when explicit intermediate-state propagation is retained, the signal can depend strongly on detuning, finite probe duration, and coherent nonequilibrium superpositions. This suggests that UCL-like reductions are controlled approximations rather than a universal description of tr-RIXS (Shi et al., 27 Jul 2025, Chen et al., 2019, Freibert et al., 2024).

3. Sources, spectrometers, timing, and high-repetition-rate detection

The experimental realization of tr-RIXS is shaped by the small inelastic scattering cross section and by the intrinsic time–energy trade-off. The measured tr-RIXS intensity is often described as the convolution of a generalized dynamical structure factor with energy- and time-resolution functions, and the intrinsic limit obeys $S(\mathbf{q},\omega,t)$0. For a coherent Gaussian pulse with $S(\mathbf{q},\omega,t)$1 meV, $S(\mathbf{q},\omega,t)$2 fs is the lower bound. In practice the resolution is further limited by monochromators, spectrometers, pulse durations, shot-to-shot arrival-time jitter, and geometry-induced path-length differences (Cao et al., 2018).

These constraints explain the reliance on XFEL infrastructure. Facilities discussed in the tr-RIXS literature include FLASH, FERMI, LCLS, LCLS-II, SACLA, European XFEL, PAL-XFEL, SwissFEL, and dedicated beamlines such as Bernina, SCS/hRIXS, and chemRIXS (Cao et al., 2018, Chen et al., 2023, Hoffman et al., 24 Mar 2026). Hard-X-ray implementations at SwissFEL use a compact $S(\mathbf{q},\omega,t)$3 m Johann-type spectrometer with up to 3 crystal analyzers and demonstrated an elastic-line width of $S(\mathbf{q},\omega,t)$4 meV at the Ir $S(\mathbf{q},\omega,t)$5-edge, with $S(\mathbf{q},\omega,t)$6 meV estimated for Si(555) monochromatization and tighter focusing (Chen et al., 2023). Soft-X-ray solution-phase operation at chemRIXS uses monochromatized 300–1600 eV X-rays, a windowless liquid-sheet endstation, APD-based shot-by-shot timing diagnostics, and a measured instrument response of $S(\mathbf{q},\omega,t)$7 fs FWHM in water TFY-XAS, with tr-RIXS planes containing more than $S(\mathbf{q},\omega,t)$8 events collected in 5 minutes (Hoffman et al., 24 Mar 2026).

Detector technology has become part of the spectroscopy rather than a downstream detail. At EuXFEL SCS, the hRIXS spectrometer with a JUNGFRAU detector equipped with an iLGAD sensor achieved a spatial resolution of $S(\mathbf{q},\omega,t)$9, resolving power $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$0 at $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$1 eV, and a burst frame rate of 47 kHz with 16 sequential gates; each gate integrated 19 pulses at 1.1 MHz (Duarte et al., 15 Nov 2025). The same study showed that, for CuO at $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$2, the integrated emitted signal decreased by $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$3 over $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$4 within a train, establishing that intra-train FEL-induced effects must be monitored in high-repetition-rate tr-RIXS (Duarte et al., 15 Nov 2025). This is not a generic statement about all samples, but it is a concrete warning that repetition-rate gains can be coupled to accumulation and reversible modification.

Probe geometry and polarization are equally consequential. Near $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$5 scattering with horizontal incident polarization suppresses nonmagnetic Thomson scattering and enhances magnetic visibility at L-edges. Grazing-incidence X-rays or thin samples are used to match pump and probe penetration depths. In the soft-X-ray regime, on-resonance penetration depths are sub-micron, making collinear geometries more practical; in the hard-X-ray regime, larger penetration depths often force compromises among footprint, fluence, and time resolution (Cao et al., 2018). This suggests that “instrumentation” in tr-RIXS is inseparable from the microscopic observable: timing, footprint matching, polarization control, and emitted-photon detection all enter the effective cross section.

4. Molecular wavepackets, vibronic coupling, and core-excited symmetry breaking

In molecules, tr-RIXS is a probe of correlated vibronic dynamics across the entire electronic state manifold. A fully dynamical treatment of pyrazine at the nitrogen K-edge used ML-MCTDH on a diabatic vibronic-coupling Hamiltonian including all 24 mass/frequency-scaled normal modes, with a core-hole lifetime $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$6 fs incorporated through a non-Hermitian term in the core-excited Hamiltonian (Freibert et al., 2024). The model grouped the electronic structure into ground and low valence states, higher valence states, and two subsets of core-excited states, and propagated nuclear motion in the ground, valence-excited, core-excited, and final valence manifolds.

The pyrazine calculation showed that probe tuning selects distinct dynamical pathways. For $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$7 eV, addressing $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$8, the emission contains bands at approximate energy losses $$I(\omega_{\mathrm{in}},\omega_{\mathrm{out}})\propto \sum_f \left| \sum_n \frac{\langle f|\hat D^\dagger|n\rangle \langle n|\hat D|g\rangle} {\omega_{\mathrm{in}}-(E_n-E_g)+i\Gamma_n} \right|^2 \delta\!\big(\omega_{\mathrm{in}}-\omega_{\mathrm{out}}-(E_f-E_g)\big),$$9, $|g\rangle$0, $|g\rangle$1, and $|g\rangle$2 eV, together with a pronounced anti-Stokes band at $|g\rangle$3 eV. That anti-Stokes feature is attributed to ultrafast vibronic symmetry breaking in the core manifold: $|g\rangle$4 and $|g\rangle$5 are nearly degenerate and coupled through $|g\rangle$6 modes, localizing the N $|g\rangle$7 orbitals and enabling otherwise forbidden channels within the core-hole lifetime. For $|g\rangle$8 eV, probing $|g\rangle$9 and early $|n\rangle$0, two dominant bands appear, but the absence of anti-Stokes features reflects the different symmetry of the addressed core-excited states (Freibert et al., 2024). The delay dependence tracks the $|n\rangle$1, $|n\rangle$2, and $|n\rangle$3 population flow, including oscillatory population exchange on the $|n\rangle$4–21 fs scale.

A related but distinct vibronic problem appears in graphite at the C K-edge. One of the first implementations of femtosecond tr-RIXS in this regime followed vibronically dressed core excitons at the $|n\rangle$5 resonance, using the integrated phonon sideband to extract an effective Huang–Rhys parameter (Malvestuto et al., 17 Apr 2025). In equilibrium the fitted coupling was $|n\rangle$6 with coupling strength $|n\rangle$7 eV, while at $|n\rangle$8 fs after pumping it became $|n\rangle$9 with $|f\rangle$0 eV, accompanied by a positive energy shift of $|f\rangle$1 meV in the resonance maximum (Malvestuto et al., 17 Apr 2025). Near resonance, $|f\rangle$2 showed a rapid decrease with characteristic time $|f\rangle$3 fs; at larger detuning, the dynamics slowed to $|f\rangle$4 fs. The analysis used an effective scattering time

$|f\rangle$5

so detuning directly controlled whether optical phonons participated in the inelastic channel (Malvestuto et al., 17 Apr 2025).

An alternative route to chemically specific ultrafast RIXS is stochastic stimulated RIXS. In CO at the O $|f\rangle$6 resonance, a two-color SASE pump–seed scheme was proposed in which the limited spectral coherence of the XFEL radiation defines the energy resolution, and covariance analysis of transmitted spectra reconstructs high-resolution SRIXS features without a monochromator (Kimberg et al., 2015). The simulated covariance maps displayed diagonal stripes obeying the energy-loss law $|f\rangle$7, with vibrational resolution set by the SASE spike width rather than by monochromator bandwidth. This suggests that, for dilute or photon-hungry molecular targets, stimulated or covariance-based variants may complement conventional spontaneous tr-RIXS rather than replace it.

5. Correlated solids: magnetic correlations, charge transfer, and nonequilibrium excitations

In condensed matter, tr-RIXS was first established as a probe of nonequilibrium collective excitations rather than only transient absorption edges. In the spin–orbit Mott insulator Sr$|f\rangle$8IrO$|f\rangle$9, hard-X-ray tr-RIXS at the Ir $\Gamma_n$0 edge observed a dispersing magnon branch below $\Gamma_n$1 meV and an orbital excitation near $\Gamma_n$2 meV, while time-resolved REXS showed strong suppression of 3D long-range antiferromagnetic order. Crucially, at $\Gamma_n$3 ps, when the 3D Bragg peak remained suppressed, tr-RIXS already detected short-range magnons, showing robust 2D in-plane magnetic correlations in the transient state (Cao et al., 2018).

The sensitivity of tr-RIXS to strictly local short-range correlations was demonstrated in CuGeO$\Gamma_n$4 at the O K-edge. There the Zhang–Rice singlet exciton appears at $\Gamma_n$5 eV energy loss, and its spectral weight is governed by the local spin configuration on neighboring CuO$\Gamma_n$6 plaquettes, because the O K-edge dipole step does not flip spin and the nonlocal de-excitation pathway projects efficiently onto the singlet final state only when neighboring spins are antiparallel (Paris et al., 2021). Pumping at $\Gamma_n$7 eV with 50 fs pulses caused an immediate suppression of the ZRS intensity within $\Gamma_n$8 ps, a non-monotonic feature around $\Gamma_n$9 ps, and a continued reduction on longer timescales, saturating after $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$0 ps and persisting to $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$1–500 ps (Paris et al., 2021). Model analysis combined a coherent phonon displacement $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$2 with a time-dependent effective temperature $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$3, showing how femtosecond lattice dynamics modulate short-range AFM correlations and how the magnetic subsystem decouples from the lattice on longer timescales (Paris et al., 2021).

In NiO, high-resolution tr-RIXS at the Ni $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$4 edge isolated transient charge-transfer excitons. After $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$5 eV ultraviolet pumping, trXAS showed a pre-edge at $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$6 eV, and tr-RIXS at that pre-edge showed an energy-gain peak at $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$7 eV and a second feature at $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$8 eV, both assigned to Ni sites transiently in the $$\begin{aligned} I(\omega_{\rm i},\omega_{\rm f},\mathbf q,t) =&\int dt_2 \int^{t_2} dt_1 \int dt'_2 \int^{t'_2} dt'_1 \; e^{i\omega_{\rm i}(t'_1-t_1)-i\omega_{\rm f}(t'_2-t_2)} \ &\times l(t_1,t_2)\,l(t'_1,t'_2)\,g(t_1,t)\,g(t_2,t)\,g(t'_1,t)\,g(t'_2,t)\, \sum_{m,n} e^{i\mathbf q\cdot(\mathbf r_m-\mathbf r_n)}\, S^{mn}(t_1,t_2,t'_2,t'_1), \end{aligned}$$9 configuration (Merzoni et al., 23 Apr 2025). The energy-gain peak rose within the $g$0 fs instrument response and decayed with $g$1 ps, whereas the $g$2 excitation at the main resonance softened by $g$3 meV and remained shifted for at least 50 ps (Merzoni et al., 23 Apr 2025). The separation between a few-picosecond localized charge-transfer exciton and a long-lived itinerant photo-doped response is a characteristic example of the local-versus-delocalized selectivity of tr-RIXS.

Hard-X-ray tr-RIXS on $g$4-Li$g$5IrO$g$6 at SwissFEL extended the method into the 0–2 eV itinerant sector of a honeycomb 5$g$7 oxide. Using $g$8 nm pumping and overall $g$9 meV resolution, the pumped-minus-unpumped spectra exhibited changes in the energy-loss region below 2 eV, with transient features centered near $l$0 eV and $l$1–1.6 eV that were ascribed to modulations of Ir-to-Ir intersite transition scattering efficiency, associated with transient screening of the on-site Coulomb interaction (Chen et al., 2023). This was not presented as a full quantitative extraction of $l$2, but as an interpretation consistent with the hopping nature of the 5$l$3 electrons and with intersite pathways in the RIXS cross section.

Theoretical work has broadened the scope further. Exact-diagonalization studies of a pumped 2D Hubbard model at an indirect K-edge showed that tr-RIXS can track bimagnons at $l$4, Mott-gap excitations near $l$5, in-gap doublon and hole features, and anti-Stokes channels generated during relaxation (Wang et al., 2019). A pump-driven transverse-field Ising chain analysis showed that the high-energy two-kink continuum is bounded by instantaneous dispersions, while low-energy oscillatory spectral weight can encode dynamical quantum phase transitions through interference terms in the nonequilibrium tr-RIXS cross section (Shi et al., 27 Jul 2025). These are theoretical rather than experimental results, but they demonstrate that tr-RIXS can be sensitive not only to transient populations but also to coherent critical dynamics.

6. Trade-offs, interpretive pitfalls, and emerging directions

A recurring interpretive pitfall is to treat tr-RIXS as a delayed equilibrium spectrum. The literature does not support that simplification. Explicit probe envelopes, finite core-hole lifetimes, intermediate-state propagation, and final-state dynamics can change lineshapes, redistribute intensity between overlapping bands, and even produce channels that would be missed or severely underestimated in static Kramers–Heisenberg treatments. In pyrazine, the $l$6 eV anti-Stokes feature is a direct consequence of explicit core-state nuclear motion and vibronic coupling; in graphite, the detuning-dependent effective scattering time determines whether phonon sidebands survive; in DMFT-based simulations, the way the probe is embedded in the impurity action determines whether certain vertex corrections are retained (Freibert et al., 2024, Malvestuto et al., 17 Apr 2025, Eckstein et al., 2020).

A second limitation is throughput. tr-RIXS is extremely photon-hungry, and the trade-off between spectral resolution and temporal resolution is not only fundamental but operational. Compact XFEL spectrometers have reached $l$7 meV in soft X-rays and $l$8 meV at $l$9 keV, while synchrotron spectrometers routinely deliver sub-50 meV FWHM and approach $S(\mathbf{q},\omega,t)$00 meV in select ranges (Cao et al., 2018). High repetition rates partly compensate, but they also introduce accumulation, sample heating, and normalization demands, as demonstrated by the intra-train signal decrease in CuO at EuXFEL (Duarte et al., 15 Nov 2025). This suggests that repetition rate alone is not an unqualified figure of merit; timing correction, detector architecture, and sample-refresh strategy are equally decisive.

New geometries and sample environments are extending the reach of the method. chemRIXS at LCLS-II is explicitly designed for solution-phase soft-X-ray tr-RIXS with windowless liquid sheets, APD-based shot-by-shot timing, and multimodal detection, enabling dilute samples and full RIXS planes on femtosecond timescales (Hoffman et al., 24 Mar 2026). In quantum materials, seeded FEL operation, analyzer multiplexing, polarization analysis, Fourier-transform or stimulated RIXS, and higher-throughput spectrometers are identified as near-term directions (Cao et al., 2018, Kimberg et al., 2015). In theory, extensions to cluster DMFT, multi-orbital Hamiltonians with full intermediate-state dynamics, and explicit environment couplings are natural next steps when nonlocal correlations, solvent effects, or dissipation are essential (Eckstein et al., 2020, Freibert et al., 2024).

Across these developments, tr-RIXS has emerged as a method whose distinctive value lies in the simultaneous treatment of resonance selectivity, finite-$S(\mathbf{q},\omega,t)$01 or local final-state sensitivity, and ultrafast delay dependence. In molecular systems it can resolve wavepacket dynamics, core-state symmetry breaking, and vibronic dressing; in correlated materials it can separate short-range magnetism, local charge-transfer excitons, and itinerant photo-doping; and in emerging high-repetition-rate implementations it can be extended to liquids and other previously inaccessible sample classes (Freibert et al., 2024, Paris et al., 2021, Merzoni et al., 23 Apr 2025, Hoffman et al., 24 Mar 2026).