In iron pnictides, it has been widely perceived that superconductivity is driven by spin fluctuations, which supports the sign reversal between order parameters (OP) on the electron and hole pockets1,2. The discovery of superconductivity in intercalated or monolayer FeSe at a critical temperature of the order of 40 K rekindled interest in Fe-based superconductivity and sent many theorists back to the drawing board3,4,5,6,7,8,9,10,11. Indeed, the simple, transparent and largely accepted idea of spin fluctuations scattering electron pairs between hole and electron pockets was shaken by the absence of hole pockets in KxFe2−ySe2. The fact that the superconducting phase was formed by filamentary inclusions in a strongly magnetic matrix12 spoke against a conventional single-sign s-wave (‘s++’) pairing13, and model calculations based on the spin-fluctuation scenario predicted a d-wave state14,15, which, by symmetry, would have generated gap nodes16. On the other hand, later photoemission experiments indicated a nodeless superconducting state17,18. After other materials with similar properties were discovered, including (Li1−xFex)OHFeSe, which could be synthesized in a single-phase form6,7, as well as FeSe monolayers8, it became increasingly clear that superconductivity at 40 K and higher is possible without hole pockets.

We do not discuss here possible mechanisms for this superconductivity, nor even whether it may or may not be similar to the superconductivity in Fe-pnictides. Instead, we will concentrate on a phenomenological question of utmost importance: is superconductivity in FeSe derivatives (assuming they all belong to the same class) of a constant OP sign, or does it have a sign-changing OP? The most natural superconductivity of the latter sort is of x2y2 type (where x and y are the directions of the Fe–Fe bonds). As discussed in refs 5,16, crystallographic symmetry-lowering due to the Se positions, and thus doubling of the unit cell, results in this state acquiring gap nodes, although in principle the nodal area may be very small. Moreover, for KxFe2−ySe2, which exhibits another electron pocket at the zone centre, this should also lead to nodal lines on this pocket, with a typical cosinusoidal angular dependence of the gap. Neither of these effects has been observed17. The other type of (truly gapless) sign-changing superconductivity was suggested in ref. 16 and a detailed theory developed by Khodas and Chubukov19. They observed that upon accounting for the spin–orbit interaction, the folded electron ellipses anti-cross and form two concentric pockets, of which the inner one is mostly of dxz/dyz, and the outer one of dxy orbital character. This ‘bonding–antibonding’ scenario5 postulates that the OP on the inner barrel has one sign, and on the outer barrel the other. The goal of this paper is not to distinguish between the dx2 − y2 and s± symmetries, but to eliminate another popular hypothesis, namely that all electron pockets have the same sign of the OP20. We emphasize that this question has a principal conceptual importance; it has been widely argued that no high-temperature superconductivity is possible at all, unless the order parameter averages to (nearly) zero over all Fermi surfaces (thus eliminating the effect of the on-site Coulomb repulsion), and it is generally accepted that spin-fluctuation-driven superconductivity necessarily requires a sign-changing order parameter.

Unfortunately, the phase-sensitive probes developed for d-wave cuprates either fail or are more questionable in Fe-based materials. Probes based on Josephson loops with π-contacts, instead of providing a qualitative test, offer only a quantitative probe, since all possible junctions have currents arising from various Fermi surface sheets corresponding to both same-sign and opposite-sign order parameters21,22. Quasiparticle interference (QPI) due to scattering from vortex cores is, in principle, phase-sensitive, but the interpretation requires specific models of the superconducting states23,24. The technique of identifying bound states formed at a non-magnetic impurity by means of scanning tunnelling microscopy (STM) is more promising, and straightforward to measure. However, here the problem is that it is often hard to prove that the investigated impurities are indeed non-magnetic.

In this paper, we shall report, first, an observation of the above-mentioned bound state, which, notwithstanding the reservations above, strongly suggests a sign-changing order parameter. Second, we will present a set of QPI measurements, and an analysis that does not rely upon the scattering from vortex cores, a poorly understood process, but upon zero-field integrated quantities, as suggested in ref. 24. As discussed in the following, this analysis involves a minimal number of assumptions and utilizes qualitative differences between the integrated QPI intensities for the sign-changing and sign-conserving QP scatterings. This methodology does not require a separation between intra- and inter-pocket scatterings, and unambiguously identifies the presence of a sign-changing OP. However, this universality comes at a cost; it cannot distinguish between a nodeless s± and a nearly nodeless d-state in these systems. This choice has to be made based on separate information.

In this work, we synthesized zinc-doped (Li1−xFex)OHFeSe and measured its material characteristics. The superconducting transition temperature determined using the midpoint of the resistive transition was suppressed from about 37.5 K to about 34.8 K, and the transition width becomes also clearly broadened by Zn-doping. More detailed information is presented as Supplementary Note 1. We have measured the scanning tunnelling spectra (STS) at 1.5 K near a single Fe-substituting impurity in (Li1−xFex)OHFe1−yZnySe, and show the results in Fig. 1. The topographic image measured with Vbias = 30 mV, It = 100 pA around the single impurity is shown in Fig. 1a. One can see that the impurity exhibits a typical image of a dumbbell shape with the centre located at the Fe site beneath the top Se layer. This dumbbell shape looks quite similar to the non-magnetic Cu impurities in NaFe0.96Co0.03Cu0.01As (ref. 25), which suggests that the Zn impurity is the origin of the dumbbell shape. This is also supported by the fact that simple counting of the impurities visible in the field of view (FOV) of the STM agrees well with the analysis of X-ray energy dispersive spectrum (EDX). Both yield an approximately 2% Zn/Se atomic ratio in the sample, which is, as expected, less than the nominal 10% concentration. Indeed, we see essentially one type of surface defect with a concentration in the right range, and, while intrinsic defects such as vacancies may or may not occur in a sizeable amount, Zn atoms that exist in the sample must manifest themselves in the STM. Thus, although we cannot definitely identify this impurity as a Zn atom, it seems exceedingly likely. In Fig. 1c, d we show the spatial evolution of the tunnelling spectra along the line shown in Fig. 1a when the magnetic field is zero and 11 T, respectively. One can see that a strong resonant peak appears at about 4.0 meV at the impurity site. When moving away from the impurity, the spectrum recovers a typical shape with two superconducting gaps, similar to the pristine system without Zn (ref. 26). The two gaps Δ1 = 14 meV and Δ2 = 8.5 meV, determined here from spectra far away from the impurity (see the inset of Fig. 1b) are also quite close to those in the Zn-free samples26. To illustrate how the magnetic field affects the resonant impurity states, we show in Fig. 1b two selected spectra measured at the impurity site with and without magnetic field. It is clear that the magnetic field suppresses the resonant peak, but does not shift or split the peak.

Figure 1: In-gap resonant states of a non-magnetic impurity.
figure 1

a, STM topography (Vbias = 30 mV, It = 100 pA) around a single impurity. The colour bar represents height of the background with units of picometres. b, Tunnelling spectra measured at the centre of the defect with a dumbbell shape under B = 0 T and 11 T, respectively. The peak position of the impurity bound state does not shift under an 11 T magnetic field, indicating its non-magnetic origin. The inset shows a spectrum measured far away from the impurity site with the two gaps Δ1 = 14 meV, Δ2 = 8.5 meV, as marked by the arrows. c,d, Tunnelling spectra measured along the red line in a at B = 0 T and 11 T, respectively. All data shown in this figure were taken at 1.5 K.

Both observations are of great importance; the former proves that the peak in question has superconducting origin, and the latter indicates that the impurity is non-magnetic. Indeed, for a magnetic impurity in the Born limit, the energy of a Bogoliubov–de-Gennes (BdG) quasiparticle contains three terms: Hqp = H0 + BS H + ∑ k, σ, k′, σJck, σ+σσ, σ S ckσ (ref. 27). Here the first term H0 is just the bare energy of a BdG quasiparticle. The second term arises from the interaction between the magnetic moment S and external field H; g is the Landé factor and μB is the Bohr magneton. This term shifts the resonant peak position in a magnetic field. The third term is due to the interaction between the BdG quasiparticles and the local magnetic moment, and will lead to a splitting of the resonant peak. J in this term is the exchange coupling parameter between the BdG quasiparticles and the magnetic moment, while σσ, σ denotes the Pauli matrix in spin space and ck σ+(ckσ) is the quasiparticle creation (annihilation) operation at momentum k(k′) with quasiparticle spin σ(σ′). We have monitored more than ten individual ‘dumbbell’ impurities and have never observed either a shift or splitting of the peak under a magnetic field in the present sample, so we conclude the impurity of this type (presumably Zn atom) is non-magnetic. According to theory28,29,30, in multiband superconductors, if the pairing gap has a sign reversal, the non-magnetic impurity with a moderate scattering potential will induce in-gap resonant bound states. For a sign-preserving gap, such as plain s-wave (s++), these bound states are not expected. The observation of the bound state peak at the non-magnetic Zn impurity reflects the sign-reversal gap of the system. Our conclusion that the bound states are associated with Zn atoms seems at odds with ref. 31, which found no such states for Zn atoms dosed onto the surface. We note, however, that the effective potential for a dosed Zn adatom is significantly weaker than for a substitutional Zn impurity. It is highly likely that the former is simply too weak to create a midgap resonance5,28,29,30. This point is discussed further in Supplementary Note 2.

Earlier angle-resolved photoemission10,11 and STM measurements26 revealed that the hole Fermi surfaces at the Brillouin centre (Γ point) are absent in (Li1−xFex)OHFeSe, with only electron pockets at the zone corners. In Fig. 2a, we present a sketch of the Fermi surface topology in the folded (2-Fe) Brillouin zone. Here is the reciprocal vector of the 2-Fe lattice, that is, it is directed along the Fe–Fe square diagonal and equal to , where a is the Fe–Fe distance.

Figure 2: Quasiparticle interference around the single impurity.
figure 2

a, The schematic Fermi surface of (Li1−xFex)OHFe1−yZnySe and three scattering processes in k space. The eccentricity of the ellipses is about 0.55. The first Brillouin zone is shown by the solid black line, that is, there are two overlapping pockets per zone. b, A sketch showing the QPI structure expected for this Fermi surface; note that q2 and q3 differ from q1 by the first and second reciprocal lattice vector, respectively. Although within the theory employed in this paper they are all equivalent, in reality we expect the signal for larger reciprocal lattice vectors to be weaker, as reflected in this sketch. The sketch is drawn by the self-correlation using the Fermi surface in a. c, A typical dI/dV map g(r, E = 8.5 meV) measured at the smaller superconducting gap energy (Δ2 = 8.5 meV) around the single impurity (T = 1.5 K). d, FT image of c, which is qualitatively similar to the sketch shown in b.

This Fermi surface geometry generates a rather simple pattern in the QPI image. Since all double pockets around the M-points are equivalent by symmetry (both in the normal and in the superconducting state), there is one roughly circular spot in QPI around the zone centre q = 0, which extends up to approximately 2kF (where kF is the Fermi vector of the outer pocket). This spot is periodically repeated at each reciprocal lattice vector G. Due to tunnelling matrix elements, one expects the spot around G = 0 to be the strongest, the one around G = (1,0) weaker (labelled as q2 in Fig. 2b), and the one around G = (1,1) even weaker (q3). We will see that this is exactly the picture we observe. The q2 and q3 spots do not carry any additional information, being symmetry-equivalent to q1, so in our analysis we concentrated on the latter. The measured QPI results in a large area can be found in Supplementary Note 3. Note that if the spin–orbit coupling induced gap is sufficiently large, one may, in principle, be able to resolve the q1 spot into three concentric rings, corresponding to the thresholds of the scattering inside the inner barrel, inside the outer barrel, and the interband scattering. Currently this is beyond attainable resolution, but if in the future, were it to become possible, in the spirit of the discussion below one would be able to distinguish between the bonding–antibonding s± and quasi-nodeless d states.

In Fig. 2c, we show a spatial map of the measured differential conductivity g(r, E = 8.5 meV) = [dI/dV ](r, E = 8.5 meV) at 1.5 K around the single impurity shown in Fig. 1a. One can clearly see patterns generated by the QPI. A Fourier transformation (FT) of Fig. 2c is shown in Fig. 2d, which can be directly compared with the cartoon in Fig. 2b. This validates the Fermi surface topology assumed in Fig. 2a, although our experimental resolution cannot assess the amplitude of the spin–orbital-coupling-induced splitting. We emphasize that to obtain dI/dV on a fine energy mesh, we have used 64 × 64 grids in the FOV with a single impurity in the centre32, and measured the STS (− 25 mV to 25 mV) at each grid point, masking the remainder of the observation window. Then we can rearrange the g(r, E) data for all the points to create maps with 64 × 64 pixels at each energy. This new method ensures stability in the following phase-related analysis on the QPI intensity. For details of the measurements and spatial position corrections we refer the reader to Methods and to Supplementary Note 4.

As discussed, in principle we should have three scattering channels and thus three circles around q = 0, as marked in Fig. 3d. Taking into account the orbital character variation around the Fermi surfaces, we observe that these three rings would roughly correspond to the orbital channels dxydxz/yz, dxz/yzdxz/yz, dxydxy for the interband, inner-intraband and outer-intraband scattering, respectively. Near q = 0 we see a bright spot (Fig. 2b) that arises from the very short q-scattering within one single Fermi surface and between the two Fermi surfaces. Due to the existence of significant background QPI intensity in the region close to q = 0 we shall not consider this part as relevant for the analysis.

Figure 3: Determination of sign reversal of the superconducting order parameter.
figure 3

a, The real-part difference of the FT-QPI δρ (q, E = 8.5 meV) = Re[ρ(q, E = 8.5 meV) − ρ(q, E = 8.5 meV)]. b, Integral of δρ(q, E) versus E. The integral range of |q| is illustrated by the ring in a with specified |q| range, 0.13π/a < |q| < 0.55π/a, with a the Fe–Fe bond length. The strong peak at approximately 4 meV is due to an impurity bound state. The calculated δρ(E) is shown in the inset as a blue solid line for s± and a red dashed line for s++. c, Integral of δρ(q, E) as a function of E with the bound state peak masked out as discussed in the main text. The theoretical results are again presented as an inset, with the same convention. d, Sketch of the Fermi surface electronic pockets, illustrating two possible sign-reversal scenarios; the blue and red colours represent the opposite signs of the order parameter. The red and blue scattering vectors span same-sign order parameters, and the green ones the opposite-sign ones, in both cases.

Some of us have recently proposed24 a new methodology for robust determination of possible order parameter sign reversal, using QPI from a single non-magnetic impurity. The central prediction is that the anti-symmetrized intensity of the real part of the FT-QPI is proportional to

for a weak scatterer, and the structure of this result is universal in the sense that it does not depend qualitatively on the strength of the scattering and the details of the electronic structure24. Here Δ1 and Δ2 are the two gaps associated with the two bands, U is a scalar scattering potential, A represents an area in q-space containing gap-sign-changing scattering vectors, and ρ(q, E) represents the FT-QPI, which is the FT of the spatial map of the differential conductivity g(r, E). The prediction for an s± pairing state is that this quantity, δρ(E), at an energy between the two gaps will be coherently enhanced and hence does not change sign with E > 0, while for an s++ case, this quantity is generally small, with an alternating sign between the two gaps. We have calculated δρ(E) from our measured data and show it in Fig. 3a. The phase shift has been corrected by taking the impurity as the origin point of the QPI image before the FT, as demonstrated in ref. 32. The details can be found in Supplementary Note 4. To get an enhanced signal/noise ratio, we have integrated the data within the ring defined by 0.13π/a < |q| < 0.55π/a, with a being the Fe–Fe bond length. This area is determined by simulation, with the requirement that the small-q scattering near q = 0 can be effectively neglected. This is explained in Supplementary Note 5. The energy dependence of δρ(E) is presented in the main panel of Fig. 3b. One can see a sharp peak at about 4.0 mV, which corresponds to an impurity resonance (see Fig. 1b). This peak is unrelated to the phase-dependent analysis of QPI and needs to be removed prior to further analysis. To illustrate the energy dependence of the δρ image, we present a movie with a series of δρ images as a function of energy (see Supplementary Movie 1). As one can see, at some low energies, an artificial asymmetry of δρ(q, E) is observed. Above the smaller gap, this asymmetry becomes weaker and finally vanishes; the energies between the two gaps, namely between 8.5 to 14 meV, which are most relevant to our analysis, are very little affected by this asymmetry.

To eliminate the effect of the bound state, we have masked the central part of the two-dimensional (2D) map of dI/dV with a circle of R = 3 pixels radius (29 pixels around the defect) by assuming a parabolic relation δρ(E) = AE2 + B|E| in the energy window from 3 to 5.5 meV. The details of this masking can be found in Supplementary Note 6. The calculated experimental data of δρ(E) after this masking are shown in the main panel of Fig. 3c. The masking procedures for experimental data and theoretical calculations are given in Supplementary Notes 6 and 7.

We are now ready to compare the experimental data with the theoretical predictions of ref. 24. To begin with, we have simulated the results of the unmasked processing presented in Fig. 3b, by integrating the simulated δρ(q, E) over the same ring in q-space. For parameters we used Eimp = 4.0 meV, Δ2 = 8.5 meV, and Δ1 = 14 meV. The calculated result is shown in the inset of Fig. 3b (more details of the calculations are presented in Methods and in Supplementary Note 6). We immediately observe that the s± model reproduces the essential features of the experimental data. One can, however, argue that in the s± case the spectral weight is dominated by the impurity resonance. To address this point, we present in Fig. 3c the δρ(E) obtained after the removal of the resonance, as discussed above, and, in the inset, the calculated δρ(E) subjected to the same masking procedure. Again, we see a qualitative difference between the two cases, with only the s± calculation reproducing the experimental spectrum (and, in fact, reproducing it very well). This gives us strong confidence that gaps change sign between two or more Fermi surface sheets, as shown in Fig. 3d. Note that our analysis is not based on a detailed, model-dependent comparison between the measurements and simulations, but upon a very qualitative analysis, and the observed agreement hinges exclusively on the fact that the assumed pairing state has a sign-changing order parameter. A new round of measurements has been repeated on the same impurity later, and the resulting energy dependence of the anti-symmetrized FT-QPI spectra is quite similar to the data presented here. In addition, a new impurity in the same sample was chosen to perform the control experiment, with a very similar result. All these results are described in Supplementary Note 8 and Supplementary Figs 7 and 8. The similar results from the control experiments lend strong support to the analysis and conclusions presented in the paper.

To summarize, we have shown, based on both the observation of an in-gap impurity state for the non-magnetic impurities, and a novel theoretical analysis of quasiparticle interference data, that the order parameter in intercalated FeSe, specifically in (Li1−xFex)OHFe1−yZnySe, alternates sign, either between the Fermi surface sheets, or within individual sheets, as illustrated in Fig. 3d. Furthermore, the order parameter sign must affect a considerable fraction of the scattering processes, meaning that the two opposite signs are roughly balanced. This puts severe restrictions on the available scenarios. Of those proposed so far, two satisfy this experiment: the bonding–antibonding s-wave state or a nearly nodeless d-wave state, as discussed, for instance, in ref. 5. At present, our results do not allow us to distinguish between the two. The latter has an advantage of having a clear candidate for pairing glue, namely spin fluctuations resulting from inter-pocket nesting in the unfolded zone (note that such fluctuations were observed, at q = {π, π/2} (refs 33,34,35) and calculated, at q = {π, 0.625π} (ref. 14) in KxFe2−ySe2 or other intercalated system, which would have naturally led to a state depicted in Fig. 3d(right). A principal problem with this option, however, is that in the related KxFe2−ySe2 it is incompatible with the nodeless Γ-pocket gap17, so one has to assume different pairing symmetry in these otherwise fairly similar materials, and the symmetry-required gap nodes at the points where the red and blue Fermi lines intersect in Fig. 3d(right), must be very steep, that is, dΔ/dk Δ/kF. The bonding–antibonding s± scenario, on the other hand, has an additional advantage in the sense that in this case the sign of the order parameter naturally follows the orbital character of the bands, with the xy orbitals carrying one sign, and the xz/yz orbitals the other (Fig. 3d). This greatly narrows the selection from various possible scenarios in the related systems36. We must emphasize that our present work is based on a novel theoretical proposal24. It would be interesting and important to do the same experiment and analysis on a system with different gaps having the same sign. A good candidate for this control experiment would be LiFeAs, with its good surface quality and well-separated hole pockets. But in that case it is not entirely clear whether all hole pockets have the same sign of the order parameter. The optimally doped Ba1−xKxFe2As2 may be another good system to try this idea, supposing that the surface is clean enough. Our observation of sign-reversing gaps in this electron-Fermi-pocket-only system suggests a universality of the pairing mechanism in the Fe-based superconductors.


Sample synthesis.

(Li1−xFex)OHFe1−yZnySe single crystals were synthesized by the hydrothermal ion-exchange method. Firstly, single crystals with the nominal composition K0.8(Fe0.9Zn0.1)2−xSe2 were manufactured in advance by the same self-flux method that had been used to synthesize pristine K0.8Fe2−xSe2 single crystals. Afterwards, 10 ml deionized water, 5 g LiOH (J&K, 99% purity), 0.6 g iron powder (Aladdin Industrial, 99% purity), and 0.3 g selenourea (Alfa Aesar, 99% purity) were added to a 50 ml Teflon-lined stainless-steel autoclave. After complete mixing, some pieces of K0.8(Fe0.9Zn0.1)2−xSe2 single crystals were added into the mixture. Then, the autoclave was sealed and heated up to 120 °C, and the temperature was maintained for 40 to 50 h. Finally, the (Li1−xFex)OHFe1−yZnySe single crystals were obtained by cooling the autoclave to room temperature. An X-ray energy dispersive spectrum (EDS) analysis using scanning electronic microscopy suggests that the composition ratio of Fe:Se is about 1.2:1, and the ratio of Zn:Se is about 2%, indicating partial substitution of Zn at the Fe sites in the Fe layer. The characteristics of the sample are presented as Supplementary Note 1.

STM/STS measurements.

The STM/STS measurements were carried out in a scanning tunnelling microscope (USM-1300, Unisoku) with an ultrahigh vacuum, low temperature and high magnetic field. The samples were cleaved in an ultrahigh vacuum with a base pressure of about 1 × 10−10 torr. Tungsten tips were used during all the STM/STS measurements. To raise the signal-to-noise ratio, a typical lock-in technique was used with an a.c. modulation of 0.4 mV at 987.5 Hz.

Differential conductivity measurements on a dense energy mesh in real space.

To eliminate uncertainties during the QPI analysis, we have measured the full spectrum using a dense energy mesh at each point in the real space. Firstly, we choose an area of 6 nm × 6 nm with a single impurity sitting at the centre of the image. Then, we divide the scanning area into 64 × 64 pixels and measure tunnelling spectra in the voltage window from − 25 mV to 25 mV with the same set-point at each position. Then the 2D mapping images of the differential conductivity at different energies are derived from the tunnelling spectrum at each point.

Theoretical calculations.

To describe the FT-QPI result, we employ a two-band tight-binding parametrization of the two elliptic electron pockets of the Fermi surface, rotated with respect to each other by 90° on the Fermi surface, as described in Supplementary Note 7. The scattering of quasiparticles by a non-magnetic impurity, measured in the FT-QPI, was calculated as a correction to the local density of states (LDOS), using the standard T-matrix approach describing multiple scattering by a single impurity. In particular, we compute the anti-symmetrized correction to the LDOS

Here, Gμ0(k, E) = − (i0 + ɛμ(k)τ3 + Δμτ1)/(E2 + Δμ2 + ɛμ2(k)) refers to the Nambu–Gor’kov Green’s function for the band μ, and τi is the ith Pauli matrix. ɛμ(k) and Δμ refer to the quasiparticle energy and superconducting gap, respectively, of the corresponding band. The T-matrix for the multiple scattering by a single impurity in the band and Nambu–Gor’kov space is defined as

where Uμμ = Uintraτ0 and Uμν = Uinterτ0 are the intra- and interband impurity scattering strengths, respectively.

Data availability.

The data that support the plots within this paper and other findings of this study are available from the corresponding author upon reasonable request.

Additional Information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.