Frustrated spin order and stripe fluctuations in FeSe

The charge and spin dynamics of the structurally simplest iron-based superconductor, FeSe, may hold the key to understanding the physics of high temperature superconductors in general. Unlike the iron pnictides, FeSe lacks long range magnetic order in spite of a similar structural transition around 90\,K. Here, we report results of Raman scattering experiments as a function of temperature and polarization, compared to simulations from a frustrated spin model. Both experiment and theory find a persistent low energy peak close to 500\,cm$^{-1}$ in $B_{1g}$ symmetry, which softens slightly around 100\,K, that we assign to spin excitations. By comparing with results from neutron scattering, this study provides evidence for nearly frustrated stripe order in FeSe.

The charge and spin dynamics of the structurally simplest iron-based superconductor, FeSe, may hold the key to understanding the physics of high temperature superconductors in general. Unlike the iron pnictides, FeSe lacks long range magnetic order in spite of a similar structural transition around 90 K. Here, we report results of Raman scattering experiments as a function of temperature and polarization, compared to simulations from a frustrated spin model. Both experiment and theory find a persistent low energy peak close to 500 cm −1 in B1g symmetry, which softens slightly around 100 K, that we assign to spin excitations. By comparing with results from neutron scattering, this study provides evidence for nearly frustrated stripe order in FeSe. Fe-based pnictides and chalcogenides, similar to cuprates, manganites or some heavy fermion compounds, are characterized by the proximity and competition of various phases including magnetism, charge order and superconductivity. Specifically, the magnetism of Febased systems has various puzzling aspects which do not straightforwardly follow from the Fe valence or changes in the Fermi surface topology [1][2][3][4]. Some systems have a nearly ordered localized moment close to 2 µ B [5], such as FeTe or rare-earth iron selenides, whereas the moments of AFe 2 As 2 -based compounds (A = Ba, Sr, Eu or Ca) are slightly below 1 µ B [6] and display aspects of itinerant spin-density-wave (SDW) magnetism with a gap in the electronic excitation spectrum [7]. In contrast others do not order down to the lowest temperatures, such as FeSe [8] or LaFePO [9].
The material specific differences are a matter of intense discussion; and low-as well as high-energy electronic and structural properties may contribute [1,2,4,[10][11][12]. In fact, the main fraction of the electronic density of states at the Fermi energy E F originates in t 2g Fe orbitals, and a substantial part of the Fe-Fe hopping occurs via the pnictogen or chalcogen atoms, hence via the xz and yz orbitals. For geometrical reasons, the resulting exchange coupling energies between nearest (J 1 ) and next-nearest † School of Solar and Advanced Renewable Energy, Department of Physics and Astronomy, University of Toledo, Toledo, Ohio 43606, USA * hackl@wmi.badw.de neighbour (J 2 ) iron atoms have the same order of magnitude, and small changes in the pnictogen (chalcogen) height above the Fe plane influence the ratio J 2 /J 1 , such that various orders are energetically very close [12].
The reduced overlap of the in-plane xy orbitals decreases the hopping integral and increases the influence of the Hund's rule interactions and correlation energy U , even though it may only be in the range of 1-2 eV. Thus the electrons in the xy orbitals have a considerably higher effective mass m * and smaller quasiparticle weight Z than those of the xz and yz orbitals. This effect was coined orbital selective Mottness [13][14][15] and observed by photoemission spectroscopy in Fe-based chalcogenides [16]. It is similar in spirit to what was found by Raman scattering in the cuprates as a function of momentum [17].
Here we address the question as to whether systematic trends can be found across the families of the Febased superconductors, and their relationship to other highly correlated systems. Specifically, Raman scattering reveals the type of magnetic "ordering" in FeSe, which shows substantial differences to other families of Fe-based systems, but similarities with the cuprates.
To set the stage, Fig. 1a shows Raman spectra in B 1g (x 2 − y 2 ) symmetry (see Methods) for La 2 CuO 4 at two characteristic temperatures well below and above the Néel temperature T N = 325 K, as an example from a material with localized moments [18,19]. The well-defined peak at approximately 2.84 J 1 [20,21] possesses a weak and continuous temperature dependence across T N [22]. The origin of the scattering in La 2 CuO 4 and other in-  sulating cuprates [23] can be traced back to Heisenbergtype physics of localized moments [24], which need only include the nearest-neighbour exchange interaction J 1 for simplicity. In contrast, most iron-based superconductors are metallic antiferromagnets in the parent state and exhibit rather different Raman signatures. In BaFe 2 As 2 ( Fig. 1b) abrupt changes are observed in B 1g symmetry upon entering the SDW state: the fluctuation peak below 100 cm −1 vanishes, a gap develops below some 500-600 cm −1 , and intensity piles up in the range 600-1,500 cm −1 [25,26], typical behaviour for a SDW or charge density wave (CDW) [27] in weak-coupling, resulting from Fermi surface nesting. Yet, even for itinerant systems such as these, longer range exchange interactions become relevant and can lead to magnetic frustration [28].
Thus, there is no consensus on a limit which best describes all Fe-based systems. Early work considered Heisenberg-type magnetism the most appropriate approach [29], and the high-energy maxima observed by Raman scattering in AFe 2 As 2 -based compounds were in-terpreted in terms of localized spins [26,30]. On the other hand, the low-energy spectra are reminiscent of CDW or SDW formation [26,27,31,32]. In principle, both effects can coexist if the strength of the correlations varies for electrons from different orbitals, where itinerant electrons form a SDW, while those on localized orbitals give rise to a Heisenberg-like response.
In contrast to the AFe 2 As 2 -based compounds, FeSe seems to be closer to localized order , with a larger mass renormalization than the iron pnictides [1]. Apart from low lying charge excitations, the remaining, presumably spin, degrees of freedom in FeSe may be adequately described by a spin-1 J 1 -J 2 -J 3 -K Heisenberg model [12] (see Fig. 1c), which provides a consistent description of our results shown in this work. The J 1 -J 2 -J 3 -K model allows for the presence of different spin orders as indicated in the phase diagram in Fig. 1d [12]. Since various types of spin order are energetically in close proximity [12,33,34], frustration may quench long-range order down to the lowest temperatures [8], even though neutron scattering experiments in FeSe find large values for the exchange energies [34,35].
Raman scattering can contribute significantly to answering these questions, covering an energy range of approximately 1 meV to 1 eV with sensitivity to both spin and charge excitations [36]. Recent experiments on FeSe analyzed the low-energy results in B 1g symmetry connected with charge excitations [37]. Here, we obtain similar experimental results at low energies, and analyze all symmetries at higher energies, to uncover crucial information about the behaviour of the spin degrees of freedom in FeSe. By comparing experimental and simulated Raman data we find a persistent low energy peak at roughly 500 cm −1 in B 1g symmetry, which softens slightly around 100 K. We assign the B 1g maximum and the related structures in A 1g and B 2g symmetry to spin excitations. The theoretical simulations aim at establishing a link between light and neutron scattering data with respect to the spin degrees of freedom and to furnish evidence for nearly frustrated stripe order at low temperature. We arrive at the conclusion that frustrated order of localized spins dominates the physics in FeSe, while the critical and charge fluctuations are not the main focus of the paper.

RESULTS
Symmetry-resolved Raman spectra of single-crystalline FeSe (see Methods) in the energy range up to 0.45 eV (3,600 cm −1 ) are shown in Fig. 2. The spectra are linear combinations of the polarization dependent raw data (see Methods and Supplemetary Material S1). For A 1g , A 2g , and B 2g symmetry we show spectra at 40, 90 and 300 K (Fig. 2b-d). For B 1g symmetry we plot only two temperatures, 40 and 300 K, to highlight the persistence of the peak at approximately 500 cm −1 as a function of temperature, while R a m a n S h i f t Ω ( c m -1 ) 2. Symmetry-resolved Raman spectra of FeSe at various temperatures for large energy transfers. a B1g spectra. b d A1g, A2g and B2g spectra as indicated. In A1g and B2g symmetry particle-hole excitations dominate the response. In agreement with the simulations weak additional peaks from spin excitations appear at low temperature (blue shaded areas). B2g shows a loss of spectral weight (shaded red). The narrow lines close to 200 cm −1 are the A1g and B1g phonons.
In the 1 Fe unit cell used here the B1g phonon appears in B2g symmetry since the axes are rotated by 45 • with respect to the crystallographic (2 Fe) cell. c The A2g cross section is completely temperature independent.
dependence. Out of the four symmetries, the A 1g , B 1g , and B 2g spectra display Raman active phonons, magnons or electron-hole excitations, while the A 2g spectra are weak and vanish below 500-1,000 cm −1 . As intensity in A 2g symmetry appears only under certain conditions not satisfied in the present study, we ignore it here. In the high-energy limit the intensities are smaller in all symmetries than those in other Fe-based systems such as BaFe 2 As 2 ( Fig. 1b and Supplementary S2). However, in the energy range up to approximately 3,000 cm −1 there is a huge additional contribution to the B 1g cross section in FeSe (Fig. 2a). The response is strongly temperature dependent and peaks at 530 cm −1 in the low-temperature limit. Between 90 and 40 K the A 1g and B 2g spectra increase slightly in the range around 700 and 3,000 cm −1 , respectively (indicated as blue shaded areas in Fig. 2b and d). The overall intensity gain in the A 1g and B 2g spectra in the shaded range is a fraction of approximately 5% of that in B 1g symmetry. The B 2g spectra exhibit a reduction in spectral weight in the range from 600 to 1900 cm −1 [shaded red] which is already fully developed at the structural transition at T s = 89.1 K in agreement with earlier work [37]. In contrast to A 1g and B 2g symmetry, the temperature dependence of the B 1g intensity is strong, whereas the peak energy changes only weakly, displaying some similarity with La 2 CuO 4 (Fig. 1a). This similarity, along with the considerations of Glasbrenner et al. [12], motivated us to explore a spin-only, Heisenberg-like model for describing the tem- perature evolution of the Raman scattering data. In Fig. 3 we show the low-temperature data along with results from numerical simulations at zero temperature for Raman scattering in a frustrated spin-1 model as described in the Methods section. The energy scale for the simulations (Fig. 3b) is given in units of J 1 which has been derived to be 123 meV or 990 cm −1 [12], allowing a semi-quantitative comparison with the experiment. As already mentioned, the experimental A 1g and B 2g spectra are not dominated by spin excitations and we do not attempt to further analyze the continua extending to energies in excess of 1 eV, considering them a background. The opposite is true for B 1g symmetry, also borne out in the simulations. For the selected values of J 1 = 123 meV, J 2 = 0.528 J 1 , J 3 = 0, and K = 0.1 J 1 , the positions of the spin excitations in the three symmetries are qualitatively reproduced. The choice of parameters is motivated by the previous use of the J 1 -J 2 Heisenberg model, with J 1 = J 2 to describe the stripe phase of iron pnictides [29]. Here we use a value of J 2 smaller than J 1 to enhance competition between Néel and stripe orders when describing FeSe.
The comparison of the different scattering symmetries, the temperature dependence, and our simulations indicate that the excitation at 500 cm −1 is an additional scattering channel superimposed on the particle-hole continuum and fluctuation response, as shown in Fig. S5 of the Supplementary Material. Here we focus on the peak centered at approximately 500 cm −1 which, in agreement with the simulations, originates from two-magnon excitations in a highly frustrated spin system, although the features below 500 cm −1 also are interesting, and were interpreted in terms of quadrupolar orbital fluctuations [37]. It is enlightening to look at the B 1g spectra across the whole range in temperature as plotted in Fig. 4. The well-defined two-magnon peak centered at approximately 500 cm −1 in the low temperature limit loses intensity, and becomes less well defined with increasing temperature up to the structural transition T s = 89.1 K. Above the structural transition, spectral weight continues to decrease and the width of the two-magnon feature grows, while the peak again becomes well-defined and the energy hardens slightly approaching the high temperature limit of the study. What may appear as a gap opening at low temperature is presumably just the reduction of spectral weight in a low-energy feature at approximately 22 cm −1 . The intensity of this lower energy response increases with temperature, leading to a well-formed peak at an energy around 50 cm −1 near the structural transition. Above the structural transition this feature rapidly loses spectral weight, hardens, and becomes indistinguishable from the two-magnon response in the high temperature limit. This low energy feature develops in a fashion very similar to that found in Ba(Fe 1−x Co x ) 2 As 2 for x > 0 [38][39][40].
Now we compare the measurements with numerical simulations for the temperature dependence of the Raman B 1g susceptibility in Fig. 5. For the simulations we use the same parameters as at T = 0 (Fig. 3). At zero temperature the simulations show a single low energy B 1g peak around 0.3 J 1 . As temperature increases, a weak shoulder forms on the low energy side of the peak, and the whole peak softens slightly and broadens over the simulated temperature range. Except for the additional intensity at low energies, Ω < 200 cm −1 , (Fig. 5a) Supplementary Fig. S5. b Simulated Raman response at temperatures as indicated. Only magnetic excitations are included. The coupling constant was derived as J1 = 123 meV in Ref. [12].
there is good qualitative agreement between theory and experiment. As shown in Fig. S4 in the Supplementary Material, a similar agreement between experiment and simulations is obtained for the temperature dependence in A 1g and B 2g symmetries, indicating that both the gain in intensity (blue shaded areas in Fig. 2) as well as the reduction in spectral weight in B 2g from 600 to 1900 cm −1 (shaded red in Fig. 2d) can be attributed to the frustrated localized magnetism.
To support our explanation of the Raman data, we simulated the dynamical spin structure factor and compared the findings to results of neutron scattering experiments [34]. While clearly not observing long-range order, above the structural transition neutron scattering finds similar intensity at finite energy for several wave vectors along the line (π, 0) − (π, π). Upon cooling, the spectral weight at these wave vectors shifts away from (π, π) to directions along (π, 0), although the respective peaks remain relatively broad. In Fig. 6a and b we show the results of the simulations for two characteristic temperatures. As temperature decreases, spectral weight shifts from (π, π) towards (π, 0) in agreement with the experiment [34]. In Fig. 6c we show the evolution of the spectral weights around (π, π) and (π, 0) in an energy window of (0.4 ± 0.1) J 1 as a function of temperature, similar to the results shown in Ref. [34]. In the experiment, the temperature where the integrated dynamical spin structure factor changes most dramatically is close to the structural transition. From our simulations, the temperature where similar changes occur in comparison to neutron scattering corresponds to the temperature at which the simulated B 1g response (Fig. 5) shows the most pronounced shoulder, and the overall intensity begins to decrease. Not surprisingly, the low-energy peak in the Raman scattering experiment is also strongest near the At high temperature there is intensity at (π, π) indicating a tendency towards Néel order. At low temperature the intensity at (π, π) is reduced and the stripe-like antiferromagnetism with (π, 0) ordering wave vector becomes stronger. c Integrated (π, π) and (π, 0) intensities as a function of temperature.
structural transition. The agreement of the experiment with the theory in both neutron and Raman scattering suggests that a dominant influence in FeSe comes from frustrated magnetism of essentially local spins. The differences between the classes of ferro-pnictides and -chalcogenides, in particular the different degrees of itineracy, may then originate in a subtle orbital differentiation across the families [1]. If FeSe were frustrated, near such a phase boundary between magnetic states, then its behavior would be consistent with the observed sensitivity to intercalation [41,42], layer thickness [43], and pressure [44], which could affect the exchange interactions through the hopping. Relative to the theoretical results below 200 cm −1 , critical fluctuations of any origin, which are characterized by a diverging correlation length close to the transition, can neither be described nor distiguished in such a small cluster calculation. Here, only experimental arguments can be applied similar to those in Ref. [40], but will not be further discussed, since they are not the primary focus of the analysis. A brief summary may be found in the Supplementary Material S5.
In summary, we have measured the Raman response of FeSe in all symmetries and compared to simulations of a frustrated spin-1 system. We decompose the experimental data in order to determine which parts of the spectra originate in particle-hole excitations, fluctuations of local spins, and low energy critical fluctuations. Comparison of the experimental data with these simulations gives evidence that the dominant contribution of the Ra-man spectra comes from magnetic competition between (π, 0) and (π, π) ordering vectors. These features of the Raman spectra, which agree qualitatively with a spin only model, consist of a dominant peak in B 1g symmetry around 500 cm −1 along with a peak at similar energy but lower intensity in A 1g and at higher energy in B 2g symmetry. These results will likely help to unravel the mechanism behind the superconducting phase found in FeSe.

Experiment
The FeSe crystals were prepared by the vapor transport technique. Details of the crystal growth and characterization are described elsewhere [45]. Before the experiment the samples were cleaved in air and the exposure time was minimized. The surfaces obtained in this way have several atomically flat regions allowing us to measure spectra down to 5 cm −1 . At the tetragonal-toorthorhombic transition T s twin boundaries appear and become clearly visible in the observation optics. As described in detail by Kretzschmar et al. [40] the appearance of stripes can be used to determine the laser heating ∆T L and T s to be (0.5 ± 0.1) K/mW and (89.1 ± 0.2) K, respectively.
Calibrated Raman scattering equipment was used for the experiment. The samples were attached to the cold finger of a He-flow cryostat having a vacuum of approximately 5 · 10 −5 Pa (5 · 10 −7 mbar). For excitation we used a diode-pumped solid state laser emitting at 575 nm (Coherent GENESIS MX-SLM 577-500) and various lines of an Ar ion laser (Coherent Innova 304). The angle of incidence was close to 66 • for reducing the elastic stray light entering the collection optics. Polarization and power of the incoming light were adjusted in a way that the light inside the sample had the proper polarization state and, respectively, a power of typically P a = 4 mW independent of polarization. For the symmetry assignment we use the 1 Fe unit cell (axes x and y parallel to the Fe-Fe bonds) which has the same orientation as the magnetic unit cell in the cases of Néel or single-stripe order (4 Fe cell). The orthorhombic distortion is along these axes whereas the crystallographic cell assumes a diamond shape with the length of the tetragonal axes preserved. Because of the rotated axes in the 1 Fe unit cell the Fe B 1g phonon appears in the B 2g spectra. Spectra at low to medium energies were measured with a resolution σ ≈ 5 cm −1 in steps of ∆Ω = 2.5 or 5 cm −1 below 250 cm −1 and steps of 10 cm −1 above where no sharp peaks need to be resolved. Spectra covering the energy range up to 0.5-1 eV were measured with a resolution σ ≈ 20 cm −1 in steps of ∆Ω = 50 cm −1 .

Simulations
We use exact diagonalization to study a Heisenberglike model on a 16 site square lattice, which contains the necessary momentum points and is small enough that exact diagonalization can reach high enough temperatures to find agreement with the temperature dependence in the experiment. This was solved using the parallel Arnoldi method [46]. The Hamiltonian is given by where S i is a spin-1 operator. The local moments of iron chalcogenides have been found to give 2µ B [47], the sum over nn is over nearest neighbours, the sum over 2nn is over next nearest neighbours, and the sum over 3nn is over next next nearest neighbours.
We determine the dominant order according to the largest static spin structure factor, given by the following expression.
Due to the possible spontaneous symmetry breaking we adjust the structure factor by the degeneracy of the momentum. To characterize the relative strength of the dominant fluctuations we project the relative intensity of the dominant static structure factor onto the range [0,1] using the following where d q is the degeneracy of momentum q, q max is the momentum with the largest d q S q , and q next is the momentum with the second largest d q S q . The Raman susceptibilities for B 1g , B 2g , and A 1g symmetries for non-zero temperatures were calculated using the Fleury Loudon scattering operator [29] given bŷ where J ij are the exchange interaction values used in the Hamiltonian,d ij is a unit vector connecting sites i and j andê in/out are the polarization vectors. For the symmetries calculated we use the polarization vectorŝ (where x and y point along the Fe-Fe directions). We use this operator to calculate the Raman response R(ω) using the continued fraction expansion [48], where R(ω) is given by and Z is the partition function. The sum over n rep-resents a sum over the eigenstates of the Hamiltonian with energy less than E 0 + 2J 1 , where E 0 is the ground state energy. The Raman susceptibility is given by The dynamical spin structure factor was calculated using the same method witĥ O replaced with S z q = 1 √ N l e iq · R l S z l .  Fig. S1a shows the complete set of polarization resolved Raman spectra we measured for FeSe at T = 40 K up to a maximum energy of 0.45 eV. The measured spectra have been corrected for the sensitivity of the instrument and divided by the Bose factor 1 − exp(− Ω kBT ) . In Fig. S1b sums of corresponding pairs of spectra are shown. Each sum contains the full set of all four symmetries (A 1g + A 2g + B 1g + B 2g ) accessible with the light polarizations in the Fe plane. All three sets exhibit the same spectral shape. The spectra measured with linear light polarizations at 45 • with respect to the Fe-Fe bonds (x x and x y ) were multiplied by a factor of 0.65 to fit the other configurations. The same factor was applied when calculating the sums for extracting the pure symmetries. The reason for this deviation from the expected x x and x y intensities lies in small inaccuracies in determining the optical constants. Since we never observed polarization leakages the main effect pertains obviously on the power absorption and transmission rather than phase shifts between the parallel and perpendicular light polarizations.
S2: RAMAN SPECTRA OF BAFE2AS2 Fig. S2 shows the Raman spectra of BaFe 2 As 2 as a function of symmetry and temperature. Towards high energies the spectra increase almost monotonically over an energy range of approximately 0.7 eV. We could not observe the pronounced nearly polarization-independent maxima in the range 2,000 -3,000 cm −1 reported in Ref. [49]. At high energies our spectra are temperature independent. At low energies pronounced changes are observed in A 1g and B 1g symmetry upon entering the striped spin density wave (SDW) state below T SDW = 135 K as described by various authors [25,26,31]. In A 2g and B 2g symmetry the changes are small but probably significant in that polarization leakage is unlikely to be the reason for the weak low-temperature peaks in the range 2,000 cm −1 and the gap-like behaviour below approximately 1,000 cm −1 . The changes are particularly pronounced in B 1g symmetry. As shown in Figs. S2c, in Fig. 1b of the main text and in more detail elsewhere [40] the fluctuation peak vanishes very rapidly and the redistribution of spectral weight from low to high energies sets in instantaneously at T SDW . All these observations show that the polarization and temperature dependences here are fundamentally different from those of FeSe (Fig. 2).  Fig. S3 shows Raman spectra of the FeSe sample at temperatures below (blue line) and above (red line) the superconducting transition temperature T c , which was determined to be T c = 8.8 K by measuring the third harmonic of the magnetic susceptibility [50]. Both spectra show a sharp increase towards the laser line which can be attributed to increased elastic scattering due to an accumulation of surface layers at low temperatures. Below T c a broad peak emerges centred around approximately 28 cm −1 which we identify as pair breaking peak at 2∆ ≈ (4.5 ± 0.5) k B T c . Above 50 cm −1 the spectra at T < T c and T ≥ T c are identical. We could not resolve the second peak close to 40 cm −1 as observed in Ref. [37]. The gap ratio of (4.5 ± 0.5) k B T c is comparable to what was found for Ba(Fe 0.939 Co 0.061 ) 2 As 2 [51] but smaller than that found for Ba 1−x K x Fe 2 As 2 [52]. The existence of a superconducting gap and a pair-breaking peak in the Raman spectra shows that the magnetic features are superposed on an electronic continuum.

S4: TEMPERATURE DEPENDENCE IN A1g AND
B2g SYMMETRIES Figure S4 compares experimental and simulated Raman spectra in A 1g and B 2g symmetry up to high energies at room temperature (red), slightly above T s (green) and below T s (blue). The choice of temperatures for the simulated spectra corresponds to Fig. 5 of the main text. Sharp phonon peaks (labelled ph) appear in the experimental spectra at 200 cm −1 the shape of which is not reproduced properly since resolution and sampling width are reduced. With J 1 ≈ 123 meV (990 cm −1 ) as found in Ref. 12 the experimental and simulated spectra can be compared semi-quantitatively. Both theory and experiment consistently show a gain in intensity for A 1g at medium energies and for B 2g at high energies (blue shaded areas in the respective spectra) as well as a reduction of spectral weight in B 2g in the range from 600 to 1900 cm −1 (shaded red). The changes appear to be more continuous in the simulations than in the experiment where the gain in intensity in both symmetries only occurs at T < T s . The reduction in spectral weight  [12]. The sharp peaks in the experimental spectra marked with "ph" are phonons.
in B 2g symmetry has already taken place at T s (green spectra).

S5: DELINEATION OF THE CONTRIBUTIONS TO THE B1g SPECTRA
The temperature and symmetry dependence of the Raman response show that the B 1g spectra are a superposition of various scattering channels as shown in Fig. S5: (i) particle-hole excitations and presumably also a weak contribution from luminescence in the range up to 1 eV and beyond, (ii) critical fluctuations of either spin or charge in the range below 250 cm −1 , and (iii) excitations of neighboring spins with the response centered at 500 cm −1 in A 1g and B 1g symmetry and at 3,000 cm −1 in B 2g symmetry.
(i) An estimate of electron-hole excitations may be obtained by comparing the B 1g with the A 1g and B 2g spectra at various temperatures including T < T c . In a first approximation we assume that luminescence has a weak symmetry and temperature dependence and find that the intensities in all channels have the same order of magnitude. We use the B 2g continuum for deriving an analytical approximation for modeling the particle-hole spectrum (blue in Fig. S5).
(ii) There are various ways to derive the Raman response of critical fluctuations with finite wave vector Q c .