WorksPhysics and Astronomy
Observation of multipartite spin entanglement in a cuprate chain
Abstract
Quantum materials are believed to host highly entangled states of matter, but probing and quantifying such entanglement has long remained experimentally elusive. Here, we use resonant inelastic X-ray scattering (RIXS) as a probe of multipartite spin entanglement. By projecting the RIXS cross section onto the appropriate spin excitation channel, we show that the Quantum Fisher Information can be extracted directly from the measured spin fluctuation spectrum. We detect at least 7-partite spin entanglement in the model one-dimensional cuprate SrCuO, revealing one of the largest entanglement depths yet reported in a solid. The observed entanglement depth agrees with Hubbard-model expectations and persists to elevated temperatures. Our results establish RIXS as an experimental probe of many-body spin entanglement and provide a platform for extensions to charge and orbital sectors and nonequilibrium settings.
Introduction
Many-body entanglement is a fundamental ingredient in the physics of quantum materials [1, 2]. In frustrated magnets, strong quantum correlations suppress classical order and stabilize highly entangled states with long-range coherence and fractionalized excitations [3, 4, 5, 6]. Fractional quantum Hall states likewise arise from highly entangled composite-fermion wavefunctions [7, 8, 9, 10], while pseudogap and strange-metal behavior in unconventional superconductors have similarly been linked to entangled electronic states [11, 12]. Quantum critical systems should exhibit entanglement over broad length and time scales, which amplifies quantum fluctuations and reshapes the response to external perturbations [13, 14] [15]. Direct signatures of entanglement can reveal phases for which conventional symmetry-breaking order parameters are absent or ambiguous, as in quantum spin liquids [3] [16], fluctuating superconductors [17], and many nonequilibrium states of matter [18, 19]. Quantitative knowledge of entanglement also informs whether a material can serve as a resource for quantum information systems [20, 21]. More broadly, entanglement metrics define the microscopic correlation patterns that quantum simulators must reproduce [22], and thus the gate requirements needed to emulate real materials [23–25]. However, in contrast to isolated quantum systems used in quantum
metrology and quantum information science, direct experimental probing of entanglement in solid-state quantum materials is a challenging task and an exciting frontier of modern physics [1] [26, 27, 28, 29].
A practical route to probe entangled states in solids relies on entanglement witnesses [30–37], operators whose expectation values are directly linked to experimental observables. Like Bell inequalities, these witnesses certify entanglement through the violation of bounds satisfied by all separable states, without requiring full state tomography. A particularly powerful example is the Quantum Fisher Information (QFI), which certifies multipartite entanglement and can be extracted directly from the dynamical susceptibility [36, 38]. Recent inelastic neutron scattering (INS) and muon spin rotation experiments have used the QFI to analyze quantum spin chains [39–41], proximate quantum spin liquids [42–44], and heavy-fermion systems [45], establishing a concrete protocol for quantifying the depth and distribution of spin entanglement in quantum magnets [46]. These developments call for entanglement witnesses in spectroscopic probes that can access a broader range of microscopic degrees of freedom across a wider class of quantum materials [47–49].
RIXS is a natural platform for probing entangled states in quantum materials. Over the past two decades, it has become an effective probe of charge, orbital, and spin degrees of freedom, with unique advantages for small samples, thin films, and two-dimensional (2D) materials [50–52]. Its scattering amplitude is phase coherent and highly sensitive to the presence of distinct electronic pathways [53], as underscored by pioneering RIXSinterferometry studies of dimer and trimer iridates [54–57]. Turning RIXS into an entanglement probe, however, is not straightforward, particularly compared with INS. Because RIXS couples to multiple microscopic degrees of freedom, its excitation operator is more complex than a local spin flip, and the corresponding QFI bounds become material- and geometry-dependent. Moreover, RIXS does not generally yield an absolute scattering cross section, so its use as an entanglement witness demands careful normalization. Local spin degrees of freedom provide a natural starting point for witnessing entanglement with RIXS. Yet, despite the widespread use of RIXS as a probe of magnetic excitations, spin entanglement has not yet been detected with this technique.

Figure 1. Collective excitations of the cuprate chain SrCuO. (a) Sketch of the Cu -edge RIXS experiment. We use incident -polarized light to maximize the scattering cross section, and tune the momentum transfer along the chain direction by varying the sample angle, . () and () denote initial (final) momentum and energy of the X-ray photons. (b) Cu -edge RIXS intensity map along the chain direction at 35 K, measured by tuning to the maximum of the Cu X-ray absorption peak. Below 1 eV, the spectra exhibit dispersing two-spinon excitations; between 1.7 and 3.2 eV they reveal a manifold of -orbital excitations. Owing to spin-orbital separation, and excitations become propagating orbitons and exhibit dispersive behavior. (c) SrCuO chains can be described as a 1D extended Hubbard model with hopping amplitude , on-site Coulomb repulsion , and intersite Coulomb interaction .
Here, we establish RIXS as a probe of spin entanglement in strongly correlated electron systems. We focus on copper oxides, where two decades of scattering experiments [58–60] and a well-characterized cross section [61–63] provide stringent experimental and theoretical benchmarks. Using high-resolution RIXS and theoretical calculations on the prototypical one-dimensional (1D) cuprate SrCuO, we project the measured signal onto the spin dynamical structure factor (accounting for multiple spin channels and core-hole effects) and extract the QFI through a sum-rule integral. The resulting momentum-resolved QFI reveals multipartite spin entanglement, at least 7-partite at low temperature, in quantitative agreement with theory. Our results establish RIXS as a spectroscopic probe of many-body entanglement in quantum materials and open a path to detecting entanglement in charge and orbital sectors, as well as in nonequilibrium states of matter.
A paradigmatic 1D cuprate
SrCuO is an ideal platform for probing many-body entanglement with RIXS Fig. 1(a). This half-filled Mott insulator consists of chains of corner-sharing CuO plaquettes aligned along the axis [64, 65], giving rise to a local structure, chemical environment, and microscopic interactions closely related to those of the superconducting cuprates. At the same time, the interchain coupling is extremely weak, rendering the chains effectively decoupled and realizing nearly ideal 1D physics [66–69].
In one dimension, strong quantum fluctuations suppress long-range order and stabilize a critical, strongly entangled ground state, while the elementary excitations fractionalize through spin–charge and spin–orbital separation [68, 70, 71]. These features are directly visible in our high-resolution ( meV) Cu -edge RIXS spectra along the chain direction Fig. 1(b). Between 1.5 and 3.0 eV, we observe dispersive and orbitons, consistent with prior results in Ref. 70, providing a clear signature of spin–orbital separation. Below eV, the spectra show a prominent continuum of strongly dispersing magnetic excitations, corresponding to pairs of spin- quasiparticles propagating along the chain [72]. This two-spinon continuum is consistent with prior work identifying SrCuO as a near-ideal spin- chain with a large superexchange scale, meV [70, 73]. At still lower energies, we resolve bond-stretching phonons, as observed across other cuprate families. Related corner-sharing quasi-1D cuprates show the same underlying phenomenology. CaCuO displays dispersive two-spinon and orbiton excitations in the undoped chain limit, while doped BaCuO reveals how the two-spinon spectrum evolves away from half filling [74, 75]. Further experimental details are provided in Appendices A and B, and Supplemental Material Secs. S1 and S2 [76].
The reduced dimensionality of SrCuO makes it particularly well suited for quantitative comparison with theory, which is far more controlled in one dimension than in higher dimensions. Its low-energy physics and the corresponding RIXS intensity are captured by the extended Hubbard model (EHM) with nearest-neighbor hopping meV, on-site interaction , and attractive nearest-neighbor interaction [Fig. 1(c)] and Appendix C, consistent with earlier work [71, 77, 78] and recent photoemission and RIXS measurements [75, 79]. Here, denotes an effective attractive Coulomb interaction at low energy, distinct from the bare repulsive intersite Coulomb interaction. All model parameters used are fixed by independent spectroscopic experiments. In the large- limit, the magnetic sector reduces to a Heisenberg model [80] with meV (Supplemental Material Sec. S5 [76]). The orbital sector is not captured within this effective description and would require a multiband treatment beyond the scope of this work [81].
Witnessing multipartite spin entanglement
The main focus of this work is multipartite spin entanglement. Because entanglement is a property of the wavefunction, we characterize it through the notion of -producibility. A pure spin state is -producible if it factorizes into non-overlapping partitions,
where each is an irreducible many-body state defined on a partition with . The entanglement depth is then if the state is -producible but not -producible, meaning that it cannot be decomposed into product wavefunctions containing fewer than spins. This definition of entanglement in a pure state is generalizable to mixed quantum states (Supplemental Material Sec. S6 [76]), such as the finite-temperature spin chain we consider here. Larger entanglement depth therefore reflects entanglement distributed across increasingly large portions of the system and provides a direct measure of many-body complexity.
We use the QFI density as a witness of multipartite entanglement. For a mixed state of an ensemble of particles, the QFI density is defined as
where is the density matrix and is a sum of local Hermitian operators with bounded spectra [82, 83]. For an -producible spin state, the QFI density is bounded by [36]
where and are the extremal eigenvalues of the local observable . For spins with local moment , these become and , and a value exceeding therefore certifies at least -partite entanglement. The QFI thus provides a Bell-like test for the presence of multipartite entanglement with a given depth.
Crucially, the QFI density is experimentally accessible through a sum rule over the dynamical response [38],
where acts as a filter that suppresses thermal contributions and precisely isolates quantum fluctuations. is the momentum-, frequency-, and temperature-dependent dynamical structure factor [38]. For spin fluctuations in an isotropic magnet,
where denotes the component of the spin operator () on site at position and time , and the brackets denote a thermal ensemble average. Total moment sum rules require that for each of the three components of [84]. For simplicity, we omit the explicit temperature dependence of from now on. Further details of the connection between and the dynamical spin structure factor are given in Supplemental Material Sec. S6 [76].
The correlation function can be extracted from INS through well-established data treatment procedures, including phonon subtraction, normalization through the atomic form factor, appropriate scaling with , and often the use of overall sum rules [84]. In RIXS, by contrast, the richer scattering cross section contains multiple excitation channels and correlation functions and therefore does not map directly onto . Extracting the dynamical spin structure factor thus requires an additional, theory-assisted cross-section projection, which in the present work is obtained from an experimentally benchmarked model of SrCuO (as discussed in the next section). For SU(2)-symmetric systems such as the isotropic spin-1/2 chain studied here, one can conceptualize direct RIXS in terms of spin creation and destruction operators , corresponding to excitations [61, 62, 85]. By contrast, indirect RIXS processes can probe higher-order spin-conserving correlations [85–87], which do not enter the present QFI formulation. Witnessing spin entanglement with RIXS therefore requires isolating the spin correlations from the scattered signal, a step at which X-ray and neutron protocols diverge because of their distinct cross sections.
RIXS intensity analysis
As spins do not couple directly to light, RIXS probes spin excitations through two distinct pathways mediated by the intermediate state [50, 61, 62]. At the Cu absorption edge, two-spinon excitations arise through both spin-flip (direct RIXS) and spin-conserving (indirect RIXS) channels Fig. 2(a). A core electron is resonantly promoted into the valence band, creating a transient doubly occupied site (orange). In the direct channel (dark-blue box), a valence electron refills the core hole and, through spin–orbit coupling in the intermediate state, leaves behind a flipped spin, generating spin-flip excitations (2 spinons with the same spin alignment). In the indirect channel (light-blue box), the spin system evolves under a locally modified superexchange Hamiltonian due to the transient configuration. During this intermediate state lifetime, spins may scatter from the excited site, leaving behind spin-conserving excitations (2 spinons with opposite spin alignment) after a valence electron refills the core hole without flipping a spin [88].
To extract the momentum-dependent structure factor, we must separate the two spin scattering channels. In systems, including many cuprates, spin-flip and spin-conserving contributions can be distinguished experimentally in certain scattering geometries through polarization selection rules, with spin-conserving scattering occurring only in the polarization-conserving channel and spin-flip scattering occurring only in the polarization-rotating channel (Supplemental Material Fig. S4 [76]). In SrCuO, however, the scattering geometry imposed by the natural cleavage plane prevents this approach: -polarized light is perpendicular to the valence orbital at all angles, causing both the absorption and emission cross sections for -polarized X-rays to vanish (Supplemental Material Sec. S3A [76]), while spin-conserving and spin-flip scattering processes both contribute to the cross section for -polarized incident and emitted X-rays. Future work can take advantage of the full tensorial nature of the RIXS cross section to experimentally isolate the spin-conserving and spin-flip contributions, as each RIXS spin channel has a distinct symmetry for its azimuthal intensity dependence. This has been demonstrated for magnons and bimagnons in 2D cuprates [89] and can be generalized to other materials platforms.
Here, we determine the relative spin-flip and spin-conserving cross sections for SrCuO by exploiting the fact that SrCuO is well described by the EHM and using exact diagonalization (ED) to compute the momentum-dependent RIXS intensity for each spin channel Appendix Fig. A1. The channel-resolved intensity depends on the intermediate-state core-hole lifetime and can be written as a fundamental scattering amplitude multiplied by a polarization- and geometry-dependent matrix element (Appendix C; Supplemental Material Sec. S3 [76]). Although the two channels carry comparable total integrated intensity, their momentum dependencies are markedly different Fig. 2(b) [88]. At r.l.u., the spectral weight arises entirely from two-spinon excitations, whereas near r.l.u. it is dominated by excitations. Thus, approaching the antiferromagnetic wavevector progressively isolates the channel relevant to the spin structure factor.
In the ultrashort core-hole lifetime limit, the direct spin-flip () RIXS channel approaches [61, 85]. In practice, the finite lifetime of the localized core hole in the intermediate state redistributes spectral weight among spin excitations at different momenta and energies Appendix Fig. A2. Therefore, we further project the RIXS channel onto assuming an inverse core-hole lifetime of ( being the hopping amplitude), as estimated from the Cu -edge X-ray absorption spectroscopy (XAS) (Supplemental Material Sec. S1 [76]).
The full projection factor for SrCuO, which accounts for both magnetic excitation channels and the finite core-hole lifetime, is shown in Fig. 2(c). We define this projection factor as
where is the total RIXS intensity from small-cluster ED calculations and is the spin structure factor of an extended Hubbard chain calculated with density matrix renormalization group (DMRG) (see Appendix C for details). In evaluating , we fix to the energy of the Cu -edge resonance. This factor converts the sum-rule integral of the measured RIXS intensity at each momentum into the corresponding sum-rule integral of required to extract the QFI. Finally, to place the spin QFI on an absolute scale, we normalize the projected RIXS intensity by fixing the total inelastic spin spectral weight integrated over energy loss and the entire Brillouin zone to , as required by the total-moment sum rule [40, 41, 84, 90–92].
The polarization- and geometry-dependent matrix elements largely set the relative weights of spin-flip and spin-conserving excitations in RIXS, and can be well-approximated for any material and scattering geometry within an atomic-model framework (Supplemental Material Sec. S3A [76]). The core-hole lifetime, in turn, is fixed by the chosen absorption edge (Appendix C; Supplemental Material Sec. S1 [76]). Thus, while the projection factor in this work is benchmarked using a specific Hamiltonian, its inputs are largely spectroscopic and geometric, making the procedure extendable across cuprates at the Cu edge and adaptable to other absorption edges, provided that the projection factor is recalculated or experimentally constrained for each material and scattering geometry (Supplemental Material Secs. S7 and S8 [76]). Future developments in absolute-intensity calibration against well-characterized scattering standards [93] or suitable Bragg reflections (such as the Bragg reflection of BiSrCaCuO) [94] could yield an experimentally determined absolute RIXS cross section and further reduce reliance on material-specific theoretical modeling.

Figure 2. Two distinct two-spinon excitation pathways. (a) Following resonant X-ray absorption, two-spinon excitations are generated via spin-flip () and spin-conserving () channels. The former arise from a direct spin flip upon de-excitation from the intermediate state, while the latter involve a spin rearrangement during the intermediate state. (b) Calculated RIXS intensities of (dark blue) and (light blue) channels in SrCuO as a function of momentum transfer along the chain direction. (c) Calculated projection factor relating the total RIXS response to for .
Multipartite spin entanglement in SrCuO
We now turn to the main result of this work, namely the observation of multipartite spin entanglement in SrCuO. Below 150 meV, the RIXS spectra contain nondispersive contributions from the elastic line, phonons and their harmonics, together with a clear dispersing two-spinon continuum [Fig. 3(a)]. To isolate the spin response, we first correct for self-absorption [95] and normalize each spectrum to the expected angle-dependent intensity of the nondispersing orbital, calculated within a single-site cluster, thereby accounting for variations of beam footprint and collection efficiency with sample angle (Supplemental Material Sec. S2 [76]). We use the orbital as a standard candle because it remains localized, whereas the other orbital excitations disperse through spin–orbital separation and are not reliably captured by few-site cluster calculations [81, 96].
After fitting and subtracting the elastic and phonon contributions, the remaining signal is the two-spinon continuum [Fig. 3(b)], with spectral weight concentrated near its lower edge around and r.l.u. (the continuum bounds for meV are overlaid in white). Weak intensity near at eV, above the upper edge of the two-spinon continuum, is assigned to four-spinon excitations [73, 97] and excluded from the QFI integral in Eq. (4).
We then evaluate the normalized QFI (nQFI) integral at each momentum by weighting the measured RIXS intensity with the projection factors in Fig. 2(c). The resulting momentum-dependent QFI, shown in Fig. 3(c), peaks sharply at the zone boundary, r.l.u. (), where it exceeds six and certifies an entanglement depth of at least 7 spins at 35 K. The large entanglement depth follows from the critical nature of the spin- chain, as its gapless ground state sustains quantum correlations over all length scales. The same criticality also implies that the QFI at the antiferromagnetic wavevector diverges logarithmically upon cooling [98]. Because the superexchange in SrCuO is exceptionally large, the chain remains effectively close to its zero-temperature critical ground state even at 35 K (), yielding a correspondingly large QFI.
The zone-boundary maximum of the QFI density arises due to the presence of staggered antiferromagnetic correlations. The QFI density is related to the coherent part of the fluctuations of operators with the general form
Due to the antiferromagnetic nature of the interactions in SrCuO, at the dominant staggered contributions add constructively and provide the most sensitive witness of entanglement. Away from , the residual phase winds with distance, causing longer-range contributions to interfere destructively and yielding a progressively looser bound on the entanglement [99].
Figure 3. Observation of 7-partite spin entanglement. (a) Experimental RIXS spectra at 35 K at selected momenta. Shaded areas correspond to the quasielastic, phonon, multiphonon, and two-spinon continuum contributions. The energy resolution is meV. (b) Experimental RIXS intensity map as a function of momentum and energy loss, after subtracting elastic and phonon contributions and normalizing as described in the text. White dashed lines are the boundaries of the two-spinon continuum. (c) Normalized QFI (nQFI) extracted from the RIXS spectrum (dark purple), compared with theoretical expectations for an extended Hubbard chain of length (blue). Error bars are obtained by propagating the Poisson uncertainties on the integrated measured RIXS photon numbers. The normalized QFI exceeds 6 at r.l.u., witnessing at least 7-partite spin entanglement.
We emphasize that the observed entanglement depth is robust against modest uncertainties in the projection factor. At r.l.u., the magnetic RIXS spectral weight is overwhelmingly dominated by the channel and the projection factor is maximized. Although an ideal 1D spin chain at zero temperature has divergent entanglement depth near r.l.u., finite momentum resolution necessarily averages over the vicinity of the zone boundary and thus yields a finite measured depth. After accounting for experimental resolution broadening, the extracted normalized QFI quantitatively follows the EHM result at every measured momentum, despite the simplicity of the single-band description.
The 7-partite entanglement witnessed in is among the deepest reported in solids to date. Inelastic neutron scattering has witnessed at least four-partite entanglement in the Heisenberg chain KCuF ( meV, ) [40], and tripartite entanglement in the XXZ chain CsCoCl ( meV, ) [41]. Four-partite entanglement has also been reported in the proximate spin-liquid KYbSe [43], while comparable depths have been reached in quantum-critical heavy-fermion metals, where high-resolution neutron measurements at millikelvin temperatures find near a Kondo-destruction critical point [45].
By contrast, in two-dimensional cuprates, sum-rule integrations of uncalibrated inelastic neutron scattering data yield at most bipartite entanglement in the pseudogap regime [11], consistent with weaker quantum fluctuations and with superconductivity cutting off the low-temperature growth of spin entanglement. The comparison with two-dimensional cuprates shows that, even for comparable exchange scales, one-dimensional criticality is essential for enhancing the entanglement depth of the spin sector.
We next examine the temperature dependence of the observed spin entanglement. A spin-1/2 Heisenberg chain is critical at , but its entanglement is expected to weaken with increasing temperature, following a characteristic logarithmic scaling [38, 98] Fig. 4(a). We compare RIXS spectra measured at 35 K and 220 K Fig. 4(b). At 220 K, the two-spinon continuum broadens and loses spectral weight, particularly near the zone boundary r.l.u., where staggered correlations are strongest. Spectra at intermediate temperatures near r.l.u. evolve continuously (Supplemental Material Sec. S4 [76]), consistent with the absence of a phase transition. The corresponding normalized QFI is shown in Fig. 4(c). Near the zone center, changes in the inferred entanglement depth are marginal and reflect the reduced sensitivity of the QFI away from the antiferromagnetic wavevector. Near the zone boundary, by contrast, the loss and broadening of two-spinon spectral weight suppress the QFI to a value still above 4, certifying at least 5-partite entanglement at 220 K. thus retains substantial multipartite spin entanglement at elevated temperature, as its large keeps the chain in the quantum regime even at 220 K (, Supplemental Material Sec. S9 [76]).
Figure 4. Multipartite spin entanglement at high temperature. (a) Sketch of the multipartite spin entanglement as a function of temperature. While at low temperature spins are highly entangled, upon increasing temperature quantum correlations become weaker and shorter range. (b) Low-energy RIXS spectra at selected momenta at 35 K (purple) and 220 K (orange). Spectra are shifted for clarity. With increasing temperature, the spectra are modestly reshaped. (c) Normalized QFI (nQFI) extracted from the RIXS data at 35 K (purple circles) and 220 K (orange circles). Error bars are obtained by propagating the Poisson uncertainties on the integrated measured RIXS photon numbers. The normalized QFI is slightly reduced at the zone boundary and remains above 4, witnessing at least 5-partite entanglement at 220 K.
Conclusions and outlook
Detecting multipartite spin entanglement with RIXS in has broad implications for the microscopic understanding of cuprates and other quantum materials. Owing to their large magnetic energy scales, cuprates can sustain substantial spin entanglement up to elevated temperatures, plausibly beyond the 1D limit. Extending entanglement witnesses to 2D cuprates would place direct constraints on the interactions shaping their magnetic continua and shed light on the reported fractionalization of magnons into two spinons [100, 101], akin to observations in other 2D square-lattice quantum antiferromagnets [102]. Because candidate resonating-valence-bond [102, 103], AF* [104], valence-bond-solid [105–107], -flux spin-liquid [108], and disordered phases [109–111] differ in their entanglement structure, the QFI provides a direct route to distinguish them through their spectroscopic signatures. Combined with accurate theory, the momentum and doping dependence of the paramagnon QFI above could further constrain the putative quantum critical point [112–114] and test whether the pseudogap and strange-metal regimes are highly entangled states of matter, as suggested by a recent analysis of inelastic neutron scattering data from Hg-based cuprates and [11].
The present implementation extracts the spin QFI through a material- and geometry-specific mapping of the RIXS cross section onto , constrained by the benchmarked Hamiltonian of . A natural next step is to make this mapping increasingly reliant on purely experimental quantities. Entanglement witnesses may be formulated directly in terms of the full RIXS operator [48], while momentum- and polarization-resolved measurements could yield higher-order correlators, including the two-particle cumulant reduced density matrix [49]. In parallel, absolute-intensity calibration against well-characterized scattering standards [93] or suitable Bragg reflections [94] could place RIXS on an experimentally determined absolute scale and further reduce reliance on material-specific modeling.
Beyond spin, RIXS couples directly to charge, orbital, and lattice degrees of freedom [50–52], thus constituting a natural platform for entanglement metrology across multiple sectors [47, 48, 115] and for the identification of novel quantum phase transitions. Proposed RIXS-based protocols can already witness orbital entanglement [47, 48] and, when combined with angle-resolved photoemission (ARPES), resolve entanglement of indistinguishable particles [49]. By extending entanglement metrology to time-resolved RIXS studies of optically driven quantum states [116–123], our work also enables the use of entanglement as a figure of merit for photoinduced and coherent phases of matter [124, 125].
Acknowledgments
We thank P. Abbamonte, S. R. Clark, R. Fazio, M. Hafezi, T. Roscilde, S. Sachdev, A. O. Scheie, Q. Si, and A. Tennant for insightful discussions. We acknowledge technical support by F. Glerean. Experiments were primarily supported by the U.S. Department of Energy, Office of Basic Energy Sciences, Early Career Award Program, under Award No. DE-SC0022883. Theoretical and numerical work (Z.S., P.S., and Y.W.) was primarily supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, under Early Career Award No. DE-SC0024524. Work performed at Brookhaven National Laboratory (RIXS and RIXS data interpretation) was supported by the U.S. Department of Energy (DOE), Division of Materials Science, under Contract No. DE-SC0012704. The dynamical DMRG simulations (D.B. and S.J.) were supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Award Number DE-SC0022311. S. F. R. T. acknowledges support by the U.S. Department of Energy, Office of Science, Office of Workforce Development for Teachers and Scientists, Office of Science Graduate Student Research (SCGSR) program. The SCGSR program is administered by the Oak Ridge Institute for Science and Education (ORISE) for the DOE. ORISE is managed by ORAU under contract number DE-SC0014664. All opinions expressed in this paper are the authors’ and do not necessarily reflect the policies and views of DOE, ORAU, or ORISE. M. M. also acknowledges support by a Mercator Fellowship of the German Research Foundation (DFG) [Research Unit 5750, “Optical Control of Quantum Materials” (OPTIMAL)]. V.M. was funded by a Harvard Quantum Initiative fellowship. P.S. is also partially supported by the Scialog Grant No. SA-QMI2025-083a from Research Corporation for Science Advancement and Kevin Wells. The simulations used resources of the National Energy Research Scientific Computing Center, a U.S. Department of Energy Office of Science User Facility located at Lawrence Berkeley National Laboratory, operated under Contract No. DE-AC02-05CH11231.
Data Availability
The data supporting the findings of this study are provided in the paper and its Supplemental Material. Additional data are available from the corresponding authors upon reasonable request.
Appendix
Sample growth and characterization
High-quality single crystals were grown by the traveling solvent floating-zone method. and (99.99%) powders were mixed in their metal ratio, ground, and calcined (24 h at 980 °C and 48 h at 1050 °C for feed-rod material; 950 °C for solvent material). The powders were reground, packed into a rubber tube, and hydrostatically pressed at 4200 kg/cm to form feed rods (8 mm diameter, 25 cm length), which were sintered in air for 72 h at 1100 °C. Crystal growth was performed in a floating-zone furnace with two ellipsoidal mirrors, and the feed and seed rods were counter-rotated to ensure homogeneous heating and mixing. Single-crystal rods were grown at 1 mm/h under 1 bar , yielding rods of mm diameter and cm length. The rods were cut into mm-sized crystals and characterized by X-ray diffraction to verify the degree of crystallinity and rule out undesired secondary phases. The lattice parameters are Å, Å, and Å, with chains oriented along the direction.
RIXS measurements
We conducted high-resolution RIXS measurements at beamline 2-ID of the National Synchrotron Light Source II, Brookhaven National Laboratory. Single-crystal samples were oriented using Laue diffraction, cleaved in rough vacuum, and immediately transferred to ultra-high vacuum for the RIXS measurements. The incident X-rays were tuned to resonance with the Cu -edge peak (932.5 eV). The X-rays were -polarized to maximize sensitivity to the spin fluctuations. The plane was oriented in the scattering plane and the -axis perpendicular to it. The scattering angle was fixed at to maximize the total momentum transfer, while the sample angle was varied to control the momentum transfer along the -axis chain direction. Given the 1D character of the chains, we neglect dispersion along the and directions. We verified that the cleaved surfaces remained unoxidized by monitoring the XAS spectrum and, prior to each measurement, set the incident energy to the XAS absorption peak to ensure consistency among RIXS spectra and rule out energy drifts. The combined RIXS energy resolution was 35 meV. Further details of the RIXS data processing are discussed in the Supplemental Material Sec. S2 [76].
Numerical Methods
Exact Diagonalization Calculations of RIXS Intensities
We performed exact diagonalization calculations of the RIXS intensity, , on chains of up to sites with periodic boundary conditions. We consider the 1D EHM Hamiltonian at half filling,
with nearest-neighbor hopping meV, on-site interaction , and nearest-neighbor interaction . We obtained the full many-body spectrum and eigenstates using Lanczos methods to enable the calculation of dynamical correlation functions.
The RIXS intensity is defined as [50]
where the final-state wavefunction is given by
In these equations, denotes the momentum transfer, the incident energy, the energy loss, and () the incident (emitted) X-ray polarization. denotes the ground-state energy and the excitation site. The intermediate-state Hamiltonian includes the spin-orbit-coupled core hole and its interaction with valence electrons. The intermediate-state lifetime is defined by the phenomenological parameter , which is incorporated by introducing a Lorentzian broadening in frequency space. We varied between and to assess its effect on spectral line shapes. Experimentally, the inverse lifetime is found to be from the XAS spectrum (Supplemental Material Sec. S1 [76]), comparable to other cuprates [60].
The dipole operators () represent orbital-and polarization-specific transitions between the core and valence bands by absorbing (emitting) an X-ray photon at a specific absorption edge at site . For Cu -edge RIXS in a system, these operators correspond to transitions between the core level and the valence state and can be expressed as , where creates a hole with spin in orbital (in the representation), and annihilates a hole with spin , at site . The dipole matrix element contains the geometric overlap between the absorbed (scattered) X-ray polarization, (), and the atomic core and valence orbitals, which depends on the scattering geometry. To facilitate calculation, ED is used to calculate the fundamental excitation intensities, omitting the individual dipole matrix elements, . The total scattering matrix element, , combining all contributions from individual , is separately computed from an atomic model [96]. The fundamental excitation intensities and total scattering matrix elements are then multiplied together to give the RIXS intensity for each polarization-and spin-specific excitation channel [61, 85] (Supplemental Material Sec. S3 [76]).
The total RIXS intensity, , is the sum of two processes: a spin-flip process which reverses a spin between the initial and final states due to the core-hole spin-orbit coupling in the intermediate state , and a spin-conserving process which preserves the total spin of the
system. The spin-conserving () and spin-flip () RIXS intensities are each computed separately by theoretically selecting the appropriate dipole transition operators and final matrix elements. The polarization configuration, together with the sample and scattering angles, sets the selection rules and thus determines whether spin-conserving or spin-flip excitations dominate the spectra. For the scattering geometry used here, both spin-conserving and spin-flip excitations occur in the - scattering channel. In the scattering geometry used for typical 2D cuprate measurements, the spin-flip excitations occur in channels which rotate the linear X-ray polarization by , while the spin-conserving excitations occur in the polarization-conserving channels (Supplemental Material Sec. S3A [76]).
Ground-State DMRG
To determine and the QFI of the Hubbard chain, we use DMRG to compute the ground state of a 128-site EHM with open boundary conditions. Upon convergence, the resulting ground state is numerically exact within the DMRG representation [126, 127]. We then simulate the wavefunction following a local spin excitation and evaluate the equal-time correlation function in real space. The zero-temperature QFI density is then obtained by computing an appropriately normalized Fourier transform of this correlation function.
Numerical results
The ED-simulated RIXS (at resonance energy) and DMRG-simulated spectra are shown in Figure A1(a) and (Figure A1b), respectively. Although the peak positions are similar, the intensities clearly have different momentum dependencies, primarily due to the mixed contributions to RIXS from spin-conserving and spin-flip 2-spinon excitations, which are shown individually in Figure A1(c) and (Figure A1d), respectively. The spin-conserving excitations are strongest near r.l.u. In contrast, at r.l.u., only excitations are possible, and the RIXS spectrum is directly proportional to . The total projection factor in Eq. (6) is determined by dividing the calculated by the calculated total RIXS intensity.
Figure A1. Numerical calculations of a 1D EHM. (a) Total calculated RIXS intensity. (b) Spin structure factor. (c) Spin-conserving 2-spinon RIXS excitation intensity. (d) Spin-flip 2-spinon RIXS excitation intensity.
Since for many materials it is possible to experimentally distinguish spin-conserving and spin-flip RIXS channels through their polarization dependence, we also calculate the projection factor from RIXS to the structure factor for different values of the core-hole lifetime, , by taking the ratio between the RIXS and the structure factor (Figure A2). As expected, in the limit of an infinitely short intermediate state lifetime (), the projection factor approaches 1 for all momenta and the spin-flip RIXS channel reflects the structure factor . As the inverse lifetime decreases, the spectral weight is redistributed across different . In experiments at the Cu -edge, the core-hole inverse lifetime , estimated from the peak width of the XAS (Supplemental Material Sec. S1 [76]). We use this value of to obtain the total projection factor shown in Fig. 2(c) of the main text.
Figure A2. Projection of spin-flip () RIXS onto as a function of inverse core-hole lifetime .
supplemental_arXiv.pdf
XAS
X-ray absorption spectroscopy (XAS) was measured using an avalanche photodiode (APD) placed in the scattering plane at . Representative XAS spectra are shown in Fig. 1 for -polarized (purple solid markers) and -polarized (red open markers) incident light. exhibits a single, strongly dichroic peak at the Cu edge ( eV), with strong absorption for light polarized within the plaquettes ( polarization) and negligible absorption for light polarized perpendicular to the plaquettes ( polarization). Strong XAS dichroism indicates high-quality samples, whereas over- and under-oxidized samples show appreciable additional peaks in both polarization channels.

Figure 1 (PDF p. 18). XAS spectra measured with -polarized (purple solid markers) and -polarized (red open markers) incident X-rays at an incident angle of . A Lorentzian fit to the -polarized spectrum (gray solid line) yields a half-width at half-maximum of eV.
We estimate the inverse core-hole lifetime from the half-width at half-maximum of the main absorption edge. A Lorentzian fit yields eV for , consistent with typical values for copper oxides [1].
RIXS DATA PROCESSING
Here we describe the data processing for the RIXS datasets presented in the main text. In order to determine the Quantum Fisher Information (QFI), the magnetic intensities must be accurately determined in absolute units. Thus we correct the data for self-absorption effects, which suppress scattering intensity for grazing-exit conditions (§ 4 A). We then benchmark the intensities by comparing the orbital intensity to that expected by an atomic model calculation (§ 4 B), correcting for variations in beam footprint and collection efficiency at each angle. Finally, we fit and subtract all non-magnetic spectral weight from the resonant inelastic X-ray scattering (RIXS) spectra (§ 4 C).
Self-absorption correction
The RIXS data are corrected for self-absorption effects, accounting for the energy dependence of the emitted light and the sample geometry, and assuming an infinitely thick sample relative to the penetration depth of the incident light according to the equations in Ref. [2]. Since the penetration depth at the Cu -edge is of order 50 nm and far smaller than the sample thickness, this approximation is accurate. The correction factor, , is given by
Figure 2 (PDF p. 19). Integrated intensity of the orbital. The calculated intensity is shown as a solid dark blue line. The integrated measured intensity before self-absorption is identified by light blue circles, and the integrated measured intensity after correcting for self-absorption effects by dark blue circles. Raw and self-absorption-corrected data in this figure contain independent datasets from the two RIXS detectors of the SIX beamline.
where accounts for the angle-dependent path lengths, according to
and accounts for the difference in absorption along different crystallographic directions, according to
is the X-ray absorption below the absorption edge; is the X-ray absorption at a given X-ray energy and along a given crystallographic direction, given by
We assume the outgoing light is -polarized, as single-site cluster calculations indicate that in the absence of orbital excitations, the cross-section for -polarized light is zero for the scattering geometry used here (§ 5 A). Self-absorption effects are primarily important for outgoing energies near the main XAS absorption edge, and thus mainly affect the low-energy-loss region of the spectrum, below the orbital excitation energy.
Orbital normalization
To correct for angle-dependent variations in beam footprint and collection efficiency, we normalize the RIXS spectra at each angle to the calculated intensity of the orbital, after correcting for energy- and angle-dependent self-absorption. We use only the orbital because it points perpendicular to the CuO chains and is therefore well localized. The remaining orbitals have appreciable overlap along the chain direction and, due to the one-dimensional (1D) interactions in SrCuO, undergo spin-orbital fractionalization. Their excitations are consequently dispersive along the chains and are not well described by local models [3, 4].
We calculate the orbital intensity using an atomic model of a single Cu d site in EDRIXS [5], in the same way the matrix elements in § 5 A were computed. We consider a square-planar crystal field splitting, set to match the energies of the measured orbital excitations, with the orbital lying at 3 eV energy loss in the RIXS spectra, the and orbitals at 2.2 eV energy loss, and the orbital at 1.8 eV energy loss. The RIXS intensity is summed over both spin-conserving and spin-flip excitations to the and over both and outgoing polarizations.
Figure 3 (PDF p. 20). RIXS data with low-energy elastic and phonon fits at (a) 35 K and (b) 220 K. (c) Comparison of low-energy spectra and fit components for r.l.u. at 35 K and 220 K. Experimentally measured RIXS spectra are shown in gray. The fitted quasielastic contribution is shown in pink, phonons in orange, and a broad multiphonon component is shown in yellow. The remaining spectral weight, attributed to spinon excitations, is shown in purple.
The experimentally measured intensity (light blue circles), determined by integrating the RIXS spectra from 2.85 to 3.5 eV to avoid contributions from nearby dispersive orbiton excitations, agrees well with the expected trend from the atomic model calculations (solid blue line) [Fig. 2]. After applying the self-absorption correction (dark blue circles), the agreement is further improved, with some variations only at large momentum transfers, where either the incoming X-rays or outgoing X-rays are quite grazing, and thus the alignment of the sample more strongly affects the collection efficiency.
Fitting and subtracting non-magnetic features
To isolate the magnetic contributions to RIXS, we fit and subtract the quasielastic line and phonons. We model the RIXS spectra as containing four low-energy features: a quasielastic peak at zero energy loss, two phonons with energies below 100 meV, and a multiphonon contribution at higher energy. We assume all peaks are Gaussian. Prior to fitting, we subtract the average value of the RIXS spectrum between and eV on the energy gain side from each RIXS spectrum, which constitutes a weak and constant background contribution. We do not include any additional background in the fitting model.
At most momenta ( r.l.u.), these non-magnetic features are energetically well-separated from the 2-spinon continuum. The phonons are visibly nondispersive, as expected for optical phonons. We simultaneously fit these RIXS spectra over the energy loss region from to 0.18 eV: the energy of the phonon peaks and the peak widths are shared global free parameters, optimized over all spectra, while the amplitudes of each peak are independent in each spectrum. From these fits, we determine that the phonons are at 26 meV and 61 meV, respectively. At 35 K, the quasielastic peak is determined to have a Gaussian width meV, and the phonons a Gaussian peak width of meV. We use these fit parameters to constrain the fit parameters for the remaining spectra where the 2-spinon continuum overlaps with the other low-energy excitations.
For the data at 220 K, we additionally include a small peak on the energy gain side, capturing weak anti-Stokes phonon spectral weight. The peak widths at 220 K are found to be slightly broader, with a width of meV, likely due to the thermal activation of additional low energy phonon modes at higher temperature. The spectra and their fits are shown in Fig. 3.
Spin-flip and spin-conserving processes
The total RIXS intensities shown in Figs. 2(b) and Figure A1 of the main text depend both on fundamental material properties and on the RIXS total scattering matrix elements, which depend on the experimental geometry. The total RIXS intensity is a sum of the intensities of the spin-flip () and spin-conserving () channels. The intensity of each channel can be expressed as a product of a geometrical total scattering matrix element and the fundamental scattering intensity, which can then be calculated separately. Thus, the total magnetic RIXS intensity for each polarization channel can be expressed as:
() is the total spin-flip (spin-conserving) RIXS intensity. () is the geometrical total scattering matrix element for spin-flip (spin-conserving) processes for incident polarization and scattered polarization ; () is the fundamental bare intensity of spin-flip (spin-conserving) excitation for momentum transfer , incident X-ray energy , and energy loss . We now discuss the separate contributions of the total scattering matrix elements (§ 5 A) and fundamental excitation intensities (§ 5 B) to the total magnetic RIXS scattering.
RIXS total scattering matrix elements
The total scattering matrix elements were computed for an atomic model using EDRIXS [5] by calculating the X-ray transition amplitudes between specific states for a single Cu site. We assumed a tetragonal crystal field, with the orbital having the lowest energy, followed by the degenerate and orbitals, then the orbital, and the half-filled orbital as the valence orbital. We specified a small magnetic field to account for the exchange field near the real sample and break the degeneracy of spin-up and spin-down states. We diagonalized the resulting Hamiltonian to determine the eigenvectors and then calculated the transition amplitudes, , from the ground state to each orbital configuration . For spin-conserving excitations, the final orbital and spin configuration is the same as the initial configuration, while for spin-flip excitations, the final configuration has the same orbital occupancies as the initial state but with a reversed spin in the orbital.
We then determined the total scattering matrix elements for spin-flip () and spin-conserving () excitations to and from the ground-state orbital by determining the polarization vectors for the incident and scattered X-rays. The total scattering matrix element, , is given by
where are the outgoing and incident polarization vectors. captures the geometric and polarization dependences which can alternatively be encoded in individual dipole matrix elements, , in the RIXS intensity calculations. We use to denote light polarized in the scattering plane, while denotes light polarized perpendicular to the scattering plane.
The scattering geometry influences the total scattering matrix elements [Fig. 4]. For the scattering geometry [Fig. 4(a)] imposed by the natural cleavage planes of SrCuO, only matrix elements for -polarized incident and emitted light are non-zero [Fig. 4(c)]. As the orbital lies in the scattering plane, the overlap between the orbital and the light polarization is independent of the sample angle , and thus both spin-conserving and spin-flip total scattering matrix elements are angle-independent. The total scattering matrix element for the spin-conserving channel is larger than the spin-flip channel for . In contrast, for the scattering geometry [Fig. 4(b)] used to study most two-dimensional (2D) cuprates, all combinations of incident and emitted polarizations are possible, with the spin-conserving () channel corresponding to conserved polarization and the spin-flip () channel corresponds to flipping the polarization between absorption and emission Fig. 4(d). As the orbital is oriented perpendicular to the scattering plane, the overlap with -polarized light is strongly dependent on the sample angle . For , the spin-flip channel () dominates for RIXS measurements with -polarized light near grazing-exit conditions.
Figure 4 (PDF p. 22). Total scattering matrix elements versus sample angle at fixed scattering angle for each polarization channel in (a), (c) the scattering plane and (b),(d) the scattering plane. Spin-flip () and spin-conserving () channels are shown in dark and light blue, respectively. Solid (dashed) lines denote incident () polarization.
Fundamental scattering intensities
We calculate the fundamental RIXS scattering intensities for the spin-conserving () and spin-flip () channels for an EHM chain with parameters describing SrCuO, as discussed in Appendix C. Figure 5 shows the fundamental scattering intensities integrated over the energy-loss axis. The spin-flip intensity peaks at the zone boundary, r.l.u., and is stronger than the spin-conserving intensity at all momenta. The spin-conserving intensity instead peaks at intermediate momenta, around r.l.u., and vanishes at the zone boundary.

Figure 5. Integrated intensity of the fundamental scattering intensity for the spin-conserving (, light blue) and spin-flip (, dark blue) 2-spinon excitations in an EHM.
The material-dependent core-hole lifetime influences the relative contributions of the spin-flip and spin-conserving channels. As spin-flip excitations are created directly from the absorption and emission dipole processes without any intermediate state dynamics, is independent of the core-hole lifetime, . In contrast, spin-conserving excitations arise due to a rearrangement of spins during the intermediate state lifetime. The rearrangements of neighboring spins occur on a characteristic superexchange timescale of , where is the spin superexchange energy [6]. The longer the superexchange timescale is relative to the core-hole lifetime, the larger the contribution from becomes. In SrCuO, meV, thus the timescale for spin rearrangements can be estimated as fs, while the core-hole lifetime is fs. Thus in SrCuO at the Cu -edge we expect a non-negligible contribution of spin-conserving excitations to the overall RIXS intensity.
Temperature dependence
Figure 6(a) shows the full measured dispersion at 35 K (purple) and 220 K (orange). At each momentum, the low-energy quasielastic and phonon scattering increase slightly due to thermal occupation. The strong bond-stretching phonon at 60 meV is minimally affected as this energy scale is above the thermal energy scale even at 220 K. Additionally, the 2-spinon continuum broadens prominently, particularly near the zone boundary, . We also measured the RIXS intensity at intermediate temperatures at fixed momentum transfers of , , , and r.l.u., shown in Figure 6(b)-(Figure 6e). For each momentum, we fixed the momentum transfer and measured the spectrum at temperatures from 35 to 300 K, allowing the sample to stabilize at each temperature. The spectra evolve continuously with temperature, with the largest changes in the two-spinon intensity near the zone boundary, r.l.u.
Figure 6. Temperature-dependent RIXS spectra. (a) Measured RIXS intensity as a function of momentum at 35 K (purple) and 220 K (orange). Measured RIXS intensity at fixed momentum for intermediate temperatures at (b) r.l.u., (c) r.l.u., (d) r.l.u., and (e) r.l.u.
Impact of on spin dynamical structure factor
To test robustness against alternative parameterizations reported in the literature, we compare the spectrum obtained for the alternative parameterization [7] with expectations for the weak- and strong-coupling limits of the Hubbard model [8] Figure 7. The spectral weight is concentrated near the lower edge of the continuum, consistent with Heisenberg-like behavior and the ratio . This supports interpreting the low-energy dynamics of SrCuO in terms of independent spin degrees of freedom.
Figure 7. Calculation of the dynamical structure factor for a Hubbard model with .
Additional notes on witnessing spin entanglement with QFI
Here we expand upon the derivations underlying the entanglement analysis in Sec. III of the main text. We provide additional details on the relations between the relevant equations, the normalization conditions required for entanglement metrology, and the sum rules obeyed by the dynamical spin structure factor. To improve readability, we repeat several key equations from the main text.
Entanglement in pure and mixed states
Because entanglement of a pure quantum state is a property of the wavefunction, we characterize it through the notion of -producibility. A pure spin state is -producible if it factorizes into non-overlapping partitions,
where each is an irreducible many-body state defined on a partition with . The entanglement depth is then if the state is -producible but not -producible. The QFI can be computed simply as
A mixed state is -producible if the density matrix admits a convex decomposition
where , , and every is a pure -producible state, although its partition may differ between terms. In this case, the QFI is then defined as
with being the density matrix and a sum of local Hermitian operators. For with eigenvalues bounded between
certifies at least -partite entanglement.
Connection between entanglement and dynamical susceptibility
Hauke et al. showed that, for a thermal Gibbs state in equilibrium, is experimentally accessible through a sum rule over the dynamical susceptibility [9]
where is the energy and is the imaginary part of the retarded correlation function [10]
Within this formalism, the factor suppresses classical low-energy thermal fluctuations from contributing to . To apply this formalism to spin entanglement, we consider the spin operator , where is the spin direction probed. We can then define a dynamical structure factor [11]
where angle brackets denote thermal averages , is the momentum transfer in reciprocal lattice units (r.l.u.) [12], denotes the position of site in units of the lattice constant, and
As such, the most general spin susceptibility is a tensor of different spin directions , where . For an isotropic magnet, the diagonal components are equivalent, so it is sufficient to consider a single component
Through the fluctuation-dissipation theorem, the dynamical susceptibility can be connected to a generalized dynamical structure factor [13]
Substituting Eq. (26) into Eq. (21) yields
as given in the main text.
The single spin component of the dynamical structure factor is constrained by the sum rule for
Note that this is one third of the typical total sum rule intensity, because only one of the three spin components is measured. This assumes spin-rotation invariance, so the total on-site moment is shared equally among the three Cartesian coordinates. For the special case of , and we obtain the QFI density corresponding to the spin operator
Generalization to arbitrary
A QFI generator must be Hermitian. The Fourier component is itself Hermitian when modulo a reciprocal lattice vector. In the one-dimensional Brillouin zone used here, the important examples are
At the antiferromagnetic wavevector , each local term still has spectral width one, so the entanglement criterion Eq. (20) applies directly.
The construction can be generalized to witness entanglement at arbitrary using the method described in Ref. [14]. While is not generally Hermitian, it can be separated into two components that are themselves Hermitian at every :
As both and are Hermitian, they define valid QFI generators associated with dynamical susceptibilities in analogy to Eq. (26), but based on correlators and , respectively. The experimentally accessible witness at a generic momentum is their summed QFI density,
The second equality follows from detailed balance. If the system is inversion symmetric, so that , then Eq. (33) reduces to a form that is effectively equivalent to Eq. (29). At the self-inverse momenta , the sine quadrature vanishes and this is the QFI of the single Hermitian operator . At a generic momentum it should instead be understood as the sum of the QFIs of the cosine and sine standing-wave spin modulations. For spin , their local squared spectral widths obey
so the normalization of the multipartite-entanglement bound is unchanged: certifies at least -partite entanglement.
Impact of on QFI
To test the sensitivity of the calculated QFI to the nearest-neighbor interaction , we compare the parameterization used in the main text, , with both and the repulsive value used in other parameterizations of one-dimensional cuprates. As shown in Figure 8, the momentum-dependent QFI changes only weakly across these values of , demonstrating that the inferred entanglement depth is robust to the choice of nearest-neighbor interaction within the parameter range considered here.
Figure 8. QFI calculated for an EHM chain with an attractive interaction, (solid blue line), a repulsive interaction, (solid purple line), and no nearest-neighbor interaction, (dashed black line).
Independence of the projection factor from and in RIXS and INS analyses
The dynamical structure factor is defined as:
where is the ground state of the Hamiltonian with energy , and is the spin operator in momentum space. can be rewritten by summing over intermediate states:
where the imaginary part of the resolvent is given by:
Integrating over , we find:
Similarly, for the RIXS intensity , we have:
Expanding , we get:
and substituting as in the operators
we obtain
The projection factor is defined as:
Since both RIXS intensity and depend on similar ground-state correlation functions, the ratio eliminates dependence on Hamiltonian parameters and . Hence, is determined solely by experimental parameters such as core-hole lifetime and matrix elements.
Purity of finite-temperature states
The multipartite spin entanglement exhibited by the sample at even a relatively high temperature of 220K can be explained by the large superexchange constant meV of the effective Heisenberg model that describes the material. At K, . Therefore, we expect that the thermal (Gibbs) state that describes the state of the system
where is the Heisenberg Hamiltonian and is the partition function for energies , is not highly mixed and remains close to the ground state at K.
In this section, we present numerical evidence that indicates that the state of the sample is indeed close to being in a pure state at temperatures 35 K and 220 K. A standard measure of how pure a mixed quantum state with density matrix is the purity:
where are the eigenvalues of . The purity measures how mixed a quantum state is, with pure states having and for the maximally mixed state. For a thermal state of a gapless Hamiltonian at fixed temperature as in Eq. (45), the purity is expected to decay exponentially in the system size . This can be seen by observing that the probability of occupying an individual eigenstate decays as , while the number of terms in the sum in Eq. (46) scales as . Consequently, the appropriate intensive quantity that is bounded in is , which we refer to as intensive purity. Using exact diagonalization of the Heisenberg Hamiltonian for system sizes up to , we compute this quantity numerically for both temperatures K and K, shown in Figure 9.
Figure 9. Intensive purity for the Heisenberg model thermal state at temperatures K and K for various system sizes with periodic boundary conditions. Purity is calculated from the eigenvalues of the Heisenberg Hamiltonian using exact diagonalization.
The intensive purity at K can be seen to saturate at as a function of system size (it is exactly 1 up to numerical precision). At K, the intensive purity decays with in the range of accessible system sizes with exact diagonalization techniques. Nonetheless, the intensive purity remains high at the largest system size , although system sizes that would allow a determination of as for K via extrapolation cannot be accessed numerically at present (the purity is intractable to compute using matrix product state methods for even modest bond dimensions). A numerically computed positive second derivative of the blue curve in Figure 9 serves as a weak indication that indeed asymptotes as to a non-zero value.
[10] Note that here we use a site-normalized definition for , whereas Ref. [9] uses an extensive definition that scales with the number of sites in the system.
[11] The sign convention in the exponential is tied to the operator ordering: reversing the sign is equivalent to interchanging and .
[12] This unit choice leads to the integral limits being rather than later on.
[13] Sometimes one sees an additional proportionality constant of in this relation. In our formalism, instead appears in Eq. (25).
[14] Tianhao Ren, Yao Shen, Marton Lajer, Sophia F. R. TenHuisen, Jennifer Sears, Wei He, Mary H. Upton, Diego Casa, Petra Becker, Matteo Mitrano, Mark P. M. Dean, and Robert M. Konik, “Witnessing Quantum Entanglement Using Resonant Inelastic X-ray Scattering,” (2024), arXiv:2404.05850.
References
- [1]M. P. M. Dean, R. S. Springell, C. Monney, K. J. Zhou, J. Pereira, I. Božović, B. Dalla Piazza, H. M. Rønnow, E. Morenzoni, J. van den Brink, T. Schmitt, and J. P. Hill, “Spin excitations in a single La₂CuO₄ layer,” Nature Materials 11, 850–854 (2012).DOI
- [2]Ru-Pan Wang, Hebatalla Elnaggar, Charles J. Titus, Keisuke Tomiyasu, Jaap Geessinck, Gertjan Koster, Federica Frati, Jun Okamoto, Di-Jing Huang, and Frank M. F. de Groot, “Saturation and self-absorption effects in the angle-dependent 2p3d resonant inelastic X-ray scattering spectra of Co³⁺,” J. Synchrotron Rad. 27, 979–987 (2020).DOI
- [3]Krzysztof Wohlfield, Satoshi Nishimoto, Maurits W. Haverkort, and Jeroen van den Brink, “Microscopic origin of spin-orbital separation in Sr₂CuO₃,” Phys. Rev. B 88, 195138 (2013).
- [4]J. Schlappa, K. Wohlfeld, K. J. Zhou, M. Mourigal, M. W. Haverkort, V. N. Strocov, L. Hozoi, C. Monney, S. Nishimoto, S. Singh, A. Revcolevschi, J.-S. Caux, L. Patthey, H. M. Rønnow, J. van den Brink, and T. Schmitt, “Spin–orbital separation in the quasi-one-dimensional Mott insulator Sr₂CuO₃,” Nature 485, 82–85 (2012).DOI
- [5]Y. L. Wang, G. Fabbris, M. P. M. Dean, and G. Kotliar, “EDRIXS: An open source toolkit for simulating spectra of resonant inelastic x-ray scattering,” Computer Physics Communications 243, 151–165 (2019).DOI
- [6]Valentina Bisogni, Stefanos Kourtis, Claude Monney, Kejin Zhou, Robert Kraus, Chinnathambi Sekar, Vladimir Strocov, Bernd Büchner, Jeroen van den Brink, Lucio Braicovich, Thorsten Schmitt, Maria Daghofer, and Jochen Geck, “Femtosecond Dynamics of Momentum-Dependent Magnetic Excitations from Resonant Inelastic X-Ray Scattering in CaCu₂O₃,” Phys. Rev. Lett. 112, 147401 (2014).pubmed.ncbi.nlm.nih.gov/24766010
- [7]Shaozhi Li, Alberto Nocera, Umesh Kumar, and Steven Johnston, “Particle-hole asymmetry in the dynamical spin and charge responses of corner-shared 1D cuprates,” Communications Physics 4, 217 (2021).
- [8]A. Nocera, N. D. Patel, J. Fernandez-Baca, E. Dagotto, and G. Alvarez, “Magnetic excitation spectra of strongly correlated quasi-one-dimensional systems: Heisenberg versus Hubbard-like behavior,” Phys. Rev. B 94, 205145 (2016).DOI
- [9]Philipp Hauke, Markus Heyl, Luca Tagliacozzo, and Peter Zoller, “Measuring multipartite entanglement through dynamic susceptibilities,” Nat. Phys. 12, 778–782 (2016).DOI
- [10]Note that here we use a site-normalized definition for χ″OO(ℏω), whereas Ref. [9] uses an extensive definition that scales with the number of sites in the system.
- [11]The sign convention in the exponential is tied to the operator ordering: reversing the sign is equivalent to interchanging Ŝᵞ(t) and Ŝᵞ(0).
- [12]This unit choice leads to the integral limits being ±1/2 rather than ±π later on.
- [13]Sometimes one sees an additional proportionality constant of π in this relation. In our formalism, π instead appears in Eq. (16).
- [14]Tianhao Ren, Yao Shen, Marton Lajer, Sophia F. R. TenHuisen, Jennifer Sears, Wei He, Mary H. Upton, Diego Casa, Petra Becker, Matteo Mitrano, Mark P. M. Dean, and Robert M. Konik, “Witnessing Quantum Entanglement Using Resonant Inelastic X-ray Scattering,” (2024), arXiv:2404.05850.arxiv.org/abs/2404.05850










