WorksPhysics and Astronomy
Ni ordering in the van der Waals gap of Ni-intercalated 2H-NbS: from disorder through and order to stripe order
Abstract
We report a synchrotron x-ray diffraction study of Ni-intercalated 2H-NbS over the range , with Ni contents measured by energy-dispersive x-ray spectroscopy. Powder diffraction on the whole series shows that the NbS host is preserved, while the Ni ions in the van der Waals gap evolve from a disordered arrangement to order at and order at , through a range where both motifs coexist, to an orthorhombic stripe-type order at . The in-plane lattice parameter increases with Ni content, whereas the interlayer spacing remains nearly constant except for a pronounced minimum at , which density-functional calculations reproduce for the ordered structure. Single-crystal diffraction on the two commensurate phases shows that the structure, with Ni on a single site and a trimerized Nb lattice, is free of stacking faults, whereas the chiral structure contains Ni on the nominally empty site, inversion twins and stacking faults of the Ni layers. The comparison between powders and crystals shows that grinding mainly degrades the stacking of the Ni layers. These results establish the structural phase diagram of NiNbS and the intercalant disorder to be taken into account in the interpretation of its magnetic properties.
main_v6.pdf
Introduction
Transition metal dichalcogenides (TMDs) of general formula , where is a transition metal and a chalcogen, form one of the most versatile families of layered van der Waals materials. Within each layer, and are strongly covalently bonded, while adjacent layers are held together only by weak van der Waals forces[1]. This anisotropy makes TMDs easy to exfoliate down to the monolayer, but it also opens a second, less exploited route to tailoring their properties: the intercalation of foreign atoms into the van der Waals gap. Because the layers are only weakly coupled, intercalation can proceed over a wide and continuously tunable range of concentrations without destroying the sub-lattice, in contrast to substitution within the layers themselves[2, 3].
Among the group-V TMDs, 2H-NbS is a structurally simple metallic host: it crystallizes in the hexagonal space group (No. 194), remains metallic down to low temperature without the charge-density-wave instability found in several isoelectronic compounds[4], and becomes superconducting below K[5, 6]. This makes it a favorable platform to study how intercalated ions order in the van der Waals gap, and how this ordering feeds back onto the host lattice.
Intercalation of transition metals (, Cr, Mn, Fe, Co, Ni) into NbS and the related NbSe, TaS and TaSe hosts is known to stabilize commensurate superstructures in which the intercalant occupies octahedral sites of the van der Waals gap in an ordered fashion, most commonly at (, chiral space group , stacking of the intercalant layers) and (, centrosymmetric space group , stacking)[2, 7, 8, 9, 10, 11]. These compounds host a remarkably diverse set of magnetic ground states, from Néel and stripe antiferromagnetism to chiral and frustrated triangular-lattice ferromagnetism, strongly coupled to the metallic host bands[12, 13, 14, 15]. Most studies, however, fix the intercalation level at a single commensurate value and do not address how the intercalant order itself develops as is varied. Composition series have been reported for CoNbS[16] and, more recently, for FeNbSe[17], but in the former the Co content was estimated from the nominal synthesis ratio rather than measured. For NiNbS, a recent report further suggests that the host itself may depart from the pure 2H polytype at low intercalation levels, with a mixed 2H/3R stacking for 18.
In this work we present a systematic structural study of NiNbS powders and single crystals whose compositions, , were determined independently by energy-dispersive X-ray spectroscopy (EDX). We first use high-resolution synchrotron powder diffraction on the whole series (§ 4) to identify the compositions at which and Ni superstructures appear, coexist, or give way to disorder and to a stripe phase, to follow the lattice parameters, compared with density-functional calculations of the ordered structures, and to estimate the correlation lengths of the Ni order. Single-crystal diffraction on the two commensurate phases and (§ 5) then provides their full structures, the widths of their superstructure reflections and their diffuse scattering. We finally discuss what the single crystals add to the powder picture and where the two differ (§ 6). Together, these results establish the structural phase diagram of as the basis for a composition-dependent study of its transport and magnetic properties.
Methods
Crystal growth and composition
Single crystals of were grown by chemical vapor transport (CVT) from the elements in the desired stoichiometric ratio, sealed in evacuated quartz ampoules with iodine as transport agent. A temperature gradient of 50 K was applied along the ampoule, the hot end being held at 950°C19. One batch was grown per composition, except for and , for which several batches were grown and the best crystals are reported here. The crystals have a well-defined hexagonal habit with shiny black faces, lateral dimensions from a few micrometers to a centimeter, and thicknesses from a few micrometers to about half a millimeter; they cleave readily perpendicular to .
The composition and morphology of the crystals were examined by scanning electron microscopy (SEM) using a Zeiss GeminiSEM 300 equipped with an Everhart–Thornley secondary-electron detector and an Oxford Instruments energy-dispersive X-ray spectroscopy (EDX) detector. Flat crystals were mounted on aluminum stubs with carbon tape, and EDX maps (typically ) were recorded at 20 kV at several positions on several crystals of each batch; quantification used the Oxford Instruments AZtec software. These measurements yielded the compositions , 0.25, 0.33, 0.37, 0.46 and 0.57 used throughout this work.
Powder X-ray diffraction
Powder X-ray diffraction was measured at room temperature on the two-circle diffractometer of the CRISTAL beamline at Synchrotron SOLEIL, equipped with Mythen2 silicon-strip detectors (Dectris), at a wavelength (18.29 keV) calibrated against a LaB6 standard, over –. The powders were obtained by grinding crystals from the same growth batches as the single crystals. Lattice parameters, Ni occupancies and phase fractions were obtained by Rietveld refinement with Profex/BGMN20, using the parent cell and, where the corresponding superlattice reflections are observed, the () and
() supercells. In these refinements the total Ni occupancy was refined, while the distribution of Ni over the octahedral sites was imposed by the structural model: the powder data determine the Ni content but not the degree of Ni order. All lattice parameters are quoted relative to the parent hexagonal cell.
Single-crystal X-ray diffraction
Single crystals of about , preselected at the Laboratoire de Physique des Solides for their low mosaicity, were measured at K on the four-circle Newport diffractometer of CRISTAL equipped with a Rigaku Oxford Diffraction Atlas CCD detector, at . A total of 9616 and 10361 reflections were collected for and , respectively, reducing to 229 and 468 independent reflections ( and 0.042) up to () for and () for . The data were integrated and scaled with CrysAlisPro (Rigaku Oxford Diffraction), including an empirical spherical-harmonics absorption correction (SCALE3 ABSPACK). The structures were solved with SHELXT21 and refined by full-matrix least squares on with SHELXL22 within Olex223. For all site occupancies except that of Ni were fixed at unity, with Ni on the site only (19 refined parameters: scale, Ni occupancy, positions and anisotropic displacement parameters). For the occupancies of the two Ni sites, and , were refined with their sum constrained to (19 parameters), all other occupancies being fixed at unity. Reciprocal-space planes were reconstructed from the same data sets with CrysAlisPro.
Peak widths were extracted from one-dimensional cuts through the reconstructed planes, integrated over r.l.u. in the perpendicular direction and fitted with a Gaussian plus a constant background. All host and superstructure reflections with a signal-to-noise ratio above 10 were analyzed (141 host and 61 superstructure reflections for , 55 and 181 for ). The full widths at half maximum (FWHM) were converted into using along , and for cuts along () or for cuts along ().
Density-functional calculations
Density-functional theory (DFT) calculations were performed with Quantum ESPRESSO24,25 using the PBEsol exchange-correlation functional26, ultrasoft pseudopotentials from pslibrary 1.0.027, Marzari–Vanderbilt smearing of 0.01 Ry28 and the simplified rotationally invariant DFT+ scheme29. A Hubbard parameter eV was applied to the Ni states for all compositions, close to the value eV obtained for from linear-response density-functional perturbation theory30–32. Plane-wave cutoffs of 100–110 Ry for the wave functions and 900 Ry for the charge density were used, with -point meshes of (3 × 3 supercell, ), () and (); the stripe order at was modeled in the orthohexagonal cell, with Ni rows along stacked AA along . Lattice parameters and atomic positions of the ordered supercells were fully relaxed.
Powder diffraction across the series
Composition
Figure 1 compares the Ni content obtained by EDX with the total Ni occupancy refined from the powder diffraction data. The two determinations agree within the uncertainty of the refinement (±0.03, dominated by the correlation between Ni occupancy and displacement parameters) for , the refined value being systematically slightly larger than the EDX one at low ; the largest deviation, 0.045, occurs for . In the following, compositions are labelled by their EDX values.

Figure 1. Ni content of NiNbS obtained from Rietveld refinement of the synchrotron powder diffraction data versus determined by EDX. The dashed line indicates equality. Error bars (±0.03) reflect the correlation between Ni occupancy and displacement parameters in the refinement.
Ni superstructures
The diffraction patterns of the whole series are shown in Figure 2. All of them are indexed by the 2H-NbS host cell, supplemented by superlattice reflections arising

Figure 2. Synchrotron powder diffraction patterns of NiNbS at room temperature ( Å), from (bottom) to (top). A smooth background was subtracted from each pattern, which was then normalized to the intensity of the 002 reflection; the vertical offset is arbitrary. The shaded band indicates the range magnified in the inset. Inset: 6–13° range with the intensity multiplied by three, the strong 002 and 00 reflections being truncated. Vertical lines mark the calculated positions of the host 002 reflection and of the superlattice reflections, indexed in the parent hexagonal cell: dotted lines for the 2 × 2 superstructure (half-integer indices) and dashed lines for the R30° superstructure (indices in thirds), computed with the mean cell of the series ( Å, Å); the corresponding reflections shift by at most 0.07° across the series.
from the arrangement of Ni in the van der Waals gaps. Although such superstructures are usually resolved by single-crystal diffraction, the brilliance and angular resolution of CRISTAL allow the weak superlattice reflections to be identified unambiguously in the powder patterns, so that both the 2 × 2 (half-integer indices) and the R30° (indices in thirds) motifs can be followed across the series (inset of Figure 2 and Table 1). A small 3R-NbS fraction is detected at only, and traces of elemental sulfur (α-S, < 0.1 wt%) in most samples.
At , no superlattice reflection is observed, consistent with a disordered Ni occupancy of the van der Waals gap. At , only the half-integer reflections of the 2 × 2 order are present, and at only the reflections in thirds of the order, with both even and odd . At and 0.46 the two families coexist: the order is then seen through its odd- reflections only, and the 2 × 2 reflections are weak. The strong preferred orientation inherent to the platelet morphology prevented a reliable determination of the volume fractions of the two orders in this coexistence range. The coexistence of the two motifs away from their ideal stoichiometries is consistent with lattice-model predictions for the superlattice phase diagram of intercalated TaS [11], and dual magnetic transitions suggesting a similar coexistence have been reported for FeNbSe [17]. At , finally, strong half-integer reflections reappear, more intense and narrower than at , while the host reflections with in-plane components split into doublets (§ 4 E).
Lattice parameters and comparison with DFT
The refined lattice parameters are listed in Table I and shown in Figure 3. The in-plane parameter increases with Ni content, from 3.3225 Å at to 3.3344 Å at , reflecting a progressive expansion of the Nb–S framework, as reported for CoNbS [16]; it is however constant between and 0.25. The parameter varies by less than 0.035 Å over the whole series. It is nearly constant from upward, at 11.894 ± 0.003 Å, but shows a pronounced minimum at , 0.020 Å below the value at and 0.028 Å below that at . A similar but stronger -axis response, saturating near with a slope about an order of magnitude larger, has been reported for CoNbS [16]; a comparison with other intercalated TMDs is given in the SM33. Whereas grows steadily upon Fe or Cu intercalation, the Ni and Co compounds of NbS remain close to the pristine value.

Figure 3. Lattice parameters of NiNbS in the parent hexagonal cell from Rietveld refinement of the powder data (Table I). (a) In-plane parameter . (b) Out-of-plane parameter , with a pronounced minimum at .
Density-functional calculations of the relaxed ordered structures help to interpret these trends. Their lattice parameters are compared with experiment in tbl:2. As usual for PBEsol, and are slightly underestimated, by about 0.3–0.4% and 0.1–0.45%, respectively. The calculations reproduce the increase of with Ni content, up to the stripe structure at . More remarkably, they also yield a local minimum of at for the perfectly ordered supercells: decreases by 0.008 Å from to and increases by 0.003 Å from to
. The sign of these variations is that of the experiment, but their amplitude is several times smaller than the measured −0.020 and +0.028 Å (between , 0.25 and 0.33). For the fully ordered stripe structure at , the calculated is even shorter, 11.842 Å: as at , a perfectly ordered 2 × 2-derived Ni arrangement contracts the interlayer spacing. This contraction is not observed at , whose lies on the plateau of the higher compositions; we come back to this difference in § 6 C. The relaxed cell is orthorhombic, with (§ 6 D).
Correlation lengths of the Ni order
The superlattice reflections are broader than the host reflections throughout the series. A multi-reflection analysis of the excess widths [Supplemental Material (SM)33] gives in-plane correlation lengths of the Ni order of several hundred Å at all ordered compositions, but correlation lengths along only ≈ 100–200 Å, i.e. about ten unit cells, for , while the host is correlated over ≈ 500–650 Å in plane and ≈ 1000–1500 Å along except 0.37, for which only two superlattice reflections are observed, so the widths cannot be decomposed independently. The stacking of the Ni layers from one van der Waals gap to the next is therefore the least correlated element of the structure in the powders, and, as shown by the comparison with single crystals in § 6 A, the one most affected by grinding. The exception is , for which the superlattice reflections show no measurable excess broadening along .
Orthorhombic distortion at
At the host reflections with in-plane components are split. The 110 and 220 reflections, single and narrow at , are each described by two components separated by and 0.100°, respectively (Figure S1 of the SM33). Both splittings correspond to the same relative difference of interplanar spacings, , i.e. the splitting grows with the scattering angle, while the 00 reflections remain single. This is most naturally explained by an orthorhombic distortion of the hexagonal lattice. Indexing the hexagonal reflections in the orthohexagonal cell (, ), this corresponds to an orthorhombic strain . The 2 × 2-type order at is therefore not hexagonal, which points to a stripe arrangement of the Ni ions (§ 6 D).

Figure S1. Splitting of the host reflections of . (a) 110 and (b) 220 reflections, indexed in the parent hexagonal cell, of (open circles) compared with the single, narrow reflections of (blue line). Both patterns are background-subtracted and normalized to their maximum. The black line is a fit with two pseudo-Voigt components (shaded), whose centers (dashed lines) are separated by in (a) and in (b), corresponding in both cases to . The splitting grows with the scattering angle, as expected for an orthorhombic distortion of the hexagonal lattice. The lower strips show the fit residuals.
Structural phase diagram
The ordering of Ni in the van der Waals gap evolves non-monotonically with composition: disordered at , 2 × 2 at , at , coexistence of both motifs at –, and a -type stripe order at . This sequence is qualitatively consistent with cluster-expansion and Monte Carlo modeling of intercalant ordering in TaS¹¹, in which the and motifs compete and coexist away from their stoichiometric compositions as a result of intercalant–intercalant interactions acting on different length scales.
| (Å) | (Å) | Superstructure | Impurities | |
| 0.18 | 3.3225 | 11.8850 | none | 3R-NbS, -S |
| 0.25 | 3.3226 | 11.8648 | -S (weak), unidentified lines | |
| 0.33 | 3.3255 | 11.8924 | (even and odd ) | -S (weakest) |
| 0.37 | 3.3276 | 11.8939 | (odd only) + weak | -S |
| 0.46 | 3.3276 | 11.8940 | (odd only) + very weak | -S |
| 0.57 | 3.3344 | 11.8965 | , orthorhombic (stripes) | none detected |
Table I. Lattice parameters of the host (parent hexagonal cell) and observed superstructures of the NiNbS powders, with determined by EDX. The statistical uncertainty of the lattice parameters is about Å; the relative uncertainty between samples, used for the error bars of Figure 3, is estimated at Å, and the common wavelength-calibration uncertainty is 0.1–0.3%. For , is the hexagonal average of the orthorhombic cell.
In the powders, the Ni order is well correlated in plane but much less along , where its correlation length is about ten unit cells for : the weakest link of the structure, and the one most sensitive to mechanical damage, is the registry of the Ni layers across the NbS sheets. The NbS host itself is pure 2H at all compositions except for a small 3R fraction at , the lowest composition studied. This is compatible with the mixed 2H/3R stacking reported for ¹⁸ and suggests that Ni intercalation stabilizes the 2H polytype.
The powder data thus establish the sequence of Ni orders across the series, the composition dependence of the lattice parameters with its minimum of at , and the short correlation length of the Ni order along . They leave three questions open. Since the distribution of Ni over the octahedral sites is imposed by the structural model in the Rietveld refinements, they do not tell how complete the Ni order is at the commensurate compositions. They cannot decide whether the broadening of the superstructure reflections is intrinsic or introduced by grinding. Finally, they give no access to the local structure, the Ni–S and Nb–Nb distances, needed to understand the minimum of . These questions are addressed by single-crystal diffraction on the two commensurate phases.
Single-crystal diffraction of the commensurate phases
Refined structures
Given the singular behavior of the composition, single-crystal diffraction was performed on and on , the two compositions for which the superstructure is best defined. The reciprocal-space reconstructions in the plane are shown in Figure 4, indexed in the reciprocal lattice of the parent hexagonal cell. Superstructure reflections appear at half-integer and for , and at for . The data are described by the space groups (No. 194, ) and (No. 182, ), respectively. The refined structures are given in Tables III and IV. They are isostructural with the corresponding members of the FeTaS family, in which the $2a \times 2a$ order at is centrosymmetric with stacking of the intercalant layers and the order at is chiral with stacking$¹⁰.

Figure 4. Reciprocal-space reconstructions of the plane of (a) NiNbS and (b) NiNbS, indexed in the reciprocal lattice of the parent hexagonal cell ( Å). Superstructure reflections appear at half-integer for and at for .
For , the structure was refined with Ni on the octahedral site of the quadrupled cell only, the three other octahedral positions per van der Waals gap (6g) being left empty. All occupancies were fixed at unity except that of Ni, which refines to 0.944(8), i.e. , in agreement with the EDX value and with the total Ni content refined from the powder data, 0.28(3) (Figure 1). This single-site model describes the data with , and the largest residual peak of the difference Fourier map, 0.99 e Å, lies close to the Nb1 column and not on the empty 6g sites; a small fraction of Ni on these sites, not included in the model, cannot be excluded. The absence of diffuse scattering at this composition (§ 5 C) further shows that the stacking of the Ni layers is well correlated in this crystal. This order distorts the Nb sublattice: the Nb atoms split into a 6b site and an undistorted 2b site lying in the same column as the Ni ions. The 6h Nb atoms form equilateral Nb triangles with a short Nb–Nb distance of 3.2005(6) Å, against 3.4285 Å between triangles and Å for the undistorted lattice, i.e. a trimerization of the Nb triangular net of about 3.4%. The same analysis applied to the published structure of FeTaS [9] gives short and long Ta–Ta distances of 3.209 and 3.405 Å for Å, so that this trimerization is a generic response of the host layer to intercalant order rather than a peculiarity of NiNbS.
| Atom | Wyckoff | occ. | (Å) | |||
| Ni | 0 | 0 | 0 | 0.944(8) | 0.0094(3) | |
| Nb1 | 0.49427(3) | 0.98853(6) | 1 | 0.0064(3) | ||
| Nb2 | 0 | 0 | 1 | 0.0065(3) | ||
| S1 | 0.16764(4) | 0.83236(4) | 0.11897(7) | 1 | 0.0077(3) | |
| S2 | 0.11436(8) | 1 | 0.0078(4) |
Table III. Refined atomic positions of NiNbS at 294 K in space group (No. 194) with Å, Å and . Refinement on 229 independent reflections (226 with , ) and 19 parameters converged to , , GooF = 1.35. All occupancies were fixed at unity (not refined) except that of Ni; is the equivalent isotropic displacement parameter.
| Atom | Wyckoff | occ. | (Å) | |||
| Ni1 | 0.910(2) | 0.00740(15) | ||||
| Ni2 | 0 | 0 | 0.090(2) | 0.0057(12)* | ||
| Nb1 | 0.50216(2) | 1 | 0.00436(15) | |||
| Nb2 | 0 | 0 | 1 | 0.00431(5) | ||
| S | 0.33276(7) | 0.33180(7) | 0.63183(4) | 1 | 0.00553(16) |
Table IV. Refined atomic positions of NiNbS at 294 K in the chiral space group (No. 182) with Å, Å and . Refinement on 468 independent reflections (465 with , ) and 19 parameters converged to , , GooF = 1.324, with a Flack parameter of 0.39(7). The Ni occupancies of the and sites were constrained to sum to , so that the refinement determines their partition and not the total Ni content.
For , in contrast, the Ni distribution is not confined to a single site. Besides the site of the cell, which refines to an occupancy of 0.910(2), a minority Ni population of 0.090(2) is found on the site (the sum of the two occupancies being constrained to the nominal content, so that only their partition is determined) , i.e. on the octahedral position that the ideal order leaves vacant. About 9% of the Ni therefore sits on the “wrong” site of the triangular Ni superlattice. The Ni–S distances of the two sites are close but not identical, 2.3796(4) Å for and 2.3689(4) Å for , both corresponding to regular NiS octahedra. The Nb sublattice remains essentially undistorted (Nb–Nb = 3.31796 and 3.31767 Å, against Å), in marked contrast with the trimerized Nb net at . The Flack parameter refines to 0.39(7) [0.35(6) from Parsons’ quotients[34]: the measured crystal is not a single enantiomorph but contains an inversion-twin fraction of about one third.
Widths of the host and superstructure reflections
The widths of the host and superstructure reflections of the two commensurate compositions are compared in Table V for the powders and for the single crystals, and representative single-crystal cuts are shown in Figure 5. Rather than converting them into correlation lengths, which would require a deconvolution of the very different resolution functions of the two setups, we compare the widths and use the ratio of superstructure to host widths within each technique as the relevant quantity.

Figure 5. Representative single-crystal cuts through host (black) and superstructure (red) reflections of (a,b) and (c,d) , along the in-plane direction (a,c) and along (b,d), centered on the fitted peak positions. The reflections, whose indices are given in each panel, were chosen among those whose widths are closest to the family averages of Table V. In-plane cuts are along for and along for . Lines are fits with a Gaussian plus a constant.
In the single crystals, the superstructure reflections are as narrow as those of the host in both directions: the ratio is 0.85(3) along and 0.87(7) along $L for , and 0.98(6) and 1.08(6) for . The Ni sublattice is therefore ordered over the same length scale as the host, within the resolution of the measurement, which is set by the pixel size of the reconstructions and by the mosaicity of the crystals.
Diffuse scattering
Superimposed on the Bragg reflections, the reconstructions of in the plane reveal continuous diffuse scattering Figure 6. This diffuse intensity has two distinctive features: it forms rods along rather than isotropic clouds, and it is present only at the superstructure positions , not at the host positions. Both point to a stacking disorder that affects the intercalant sublattice alone: the layers remain well stacked, while the registry of the Ni order from one gap to the next is only partially correlated. The coexistence of Bragg peaks and diffuse rods indicates that regions in which the stacking of the Ni layers is maintained over many unit cells coexist with faulted regions. This is consistent with the short correlation length of the Ni order along found in the powders.

Figure 6. (a) Reciprocal-space reconstruction of the plane of NiNbS, showing rods of diffuse scattering along at the superstructure positions , while the host reflections at integer (allowed for even only) show no diffuse component. (b) Reconstruction of the plane of NiNbS: neither the host reflections at integer nor the superstructure reflections at half-integer show a diffuse component along .
No such diffuse scattering is observed for [Fig. 6(b)]. Between the Bragg reflections, the intensity along the superstructure rows of the plane exceeds the background measured halfway between rows by at most 20%, as along the host rows, whereas along the superstructure rows of it reaches 1.7 to 2.3 times the background. The Ni layers of the crystal are therefore stacked without detectable faults, in line with the description of its structure by a single Ni site, whereas at the stacking faults accompany the partial occupation of the site and the inversion twinning (Sec. V B).
| powder | crystal | powder | crystal | |
| 5.64(1) | 48(1) | 7.03(4) | 56(3) | |
| 17.70(5) | 41(1) | 12.74(8) | 55(1) | |
| 3.14(1) | 0.85(3) | 1.81(2) | 0.98(6) | |
| 4.61(1) | 78(3) | 4.67(1) | 56(3) | |
| 53.3(5) | 68(4) | 16.49(4) | 61(1) | |
| 11.5(1) | 0.87(7) | 3.53(1) | 1.08(6) |
Table 5. Widths (in ) of the host and superstructure (SS) reflections along the in-plane () and out-of-plane () directions for and powders and single crystals. Powder: Scherrer widths, with , and the FWHM in radians, of the and 002 (host) and and (SS) reflections for , and of the and 002 (host) and and (SS) reflections for . Single crystal: average over all well-resolved reflections of each family (Sec. II C). Uncertainties are the fit errors for the powder and the standard errors of the mean for the crystals; the errors on the ratios are propagated from those of the widths. Absolute widths cannot be compared between the two setups.
Discussion
Single crystals versus powders
In contrast with the single crystals, the superstructure reflections of the powders, obtained by grinding crystals of the same batches, are markedly broader than those of the host: the ratio reaches 3.1 along and 11.5 along for , and 1.8 and 3.5 for (Table V), while the host reflections themselves are quite narrow (–). Grinding therefore degrades the Ni order much more strongly than the host lattice, and does so predominantly along , i.e. by disrupting the stacking correlations of the intercalant layers rather than their in-plane order. This is consistent with the layered morphology of the crystals, which cleave along : the mechanical damage introduced by grinding acts primarily on the van der Waals gaps, which is precisely where the Ni sublattice resides. The effect is strongest at , which is also the composition at which the lattice parameters of the two kinds of samples differ most.
Expressed in the parent hexagonal cell, the single crystals give and for , and and for . The lattice expands between the two compositions, as in the powders ( and Å for the single crystals, against and Å for the powders). The single-crystal values of at both compositions and of at are smaller than the powder values by the same amount, 0.22–0.24%, which we attribute to the wavelength calibration of the two setups. The value of at is instead 0.34% smaller than in the powder, i.e. about 0.013 Å below what the calibration alone would give. The single crystal, whose superstructure reflections are as narrow as the host ones (§ 5 B) and which shows no diffuse scattering (§ 5 C), therefore has a markedly shorter interlayer spacing than the powder, in which the Ni order is degraded by grinding: the minimum of at is even deeper in the best-ordered material.
Intrinsic intercalant disorder and chirality at
The single-crystal analysis reveals an intrinsic intercalant disorder that the powder data cannot detect: at , 9.0(2)% of the Ni sits on the site that the ideal order leaves empty, whereas the data are described by a model with Ni on the site only and show no diffuse scattering. Even the best-ordered crystals therefore carry a few percent of quenched intercalant disorder, which acts directly on the magnetic lattice of these frustrated triangular systems[13, 15].
The stacking disorder of the Ni layers revealed by the diffuse rods is not independent of the chirality of the structure. In it is precisely the stacking of the Ni layers along that defines the handedness of the structure, so that a stacking fault is a natural boundary between domains of opposite chirality. The Flack parameter of 0.39(7) shows that such inverted domains are present in the very crystal that produces the diffuse rods, and the two observations are best regarded as two views of the same defect: a network of -axis stacking faults separating chiral domains, of the kind imaged by transmission electron microscopy in FeTaS[10] and consistent with the streaking of superlattice reflections along reported for FeTaS[9]. Since the magnetism of NiNbS and of its Co and Cr analogues is discussed in terms of chiral and non-coplanar spin textures, the inversion-twin fraction, the stacking faults and the 9% of Ni on the site set a concrete structural bound on how clean a chiral magnetic ground state can be in such crystals.
The -axis minimum at
The minimum of the interlayer spacing at the commensurate composition cannot be accounted for by a simple steric (Vegard-type) effect, which would make grow monotonically with Ni content. Three observations bear on its origin.
First, the minimum is an intrinsic property of the ordered structure. The relaxed DFT+ supercells, which are perfectly ordered at every composition, reproduce a local minimum of at , between and (Table II). The 9% of misplaced Ni found at is therefore not required to explain why is larger at than at .
| Cell (atoms) | (56) | (26) | (20) | (14) |
| , DFT | 3.309 | 3.313 | 3.316 | 3.322† |
| , DFT | 11.858 | 11.850 | 11.853 | 11.842 |
| , exp. | 3.3225* | 3.3226 | 3.3255 | 3.3344‡ |
| , exp. | 11.8850* | 11.8648 | 11.8924 | 11.8965‡ |
Table II. Lattice parameters (parent hexagonal cell, in Å) of the relaxed ordered DFT+ supercells compared with the powder values of Table I at the closest compositions. * , disordered. † Mean of Å and Å of the orthorhombic cell. ‡ ; hexagonal average of the orthorhombic cell.
Second, the comparison of the two refined structures (Figure S4, Table S2 of the SM33) locates the contraction of between and , −20.4(1) mÅ. At the S layers determining the gap split into two sites: S1 (, three S atoms in four), bonded to Ni, and S2 (, one in four), which has no Ni neighbor and caps the contracted Nb triangle. S2 sits 0.050 Å closer to the Ni plane than the S atoms at , whereas S1 stays within 5 mÅ of its height, with a Ni–S bond of 2.3841(6) Å, slightly longer than the 2.3796(4) Å of the majority site at . The S2 displacement alone contributes mÅ to , through the van der Waals gap, while the small outward shift of S1 offsets mÅ; the gap width, averaged over the S sites, thus decreases by mÅ, whereas the S–Nb–S slab thickness changes by only mÅ. The contraction of at is therefore not due to a shorter Ni–S bond but to the S atoms that are not bonded to Ni, which relax into the empty part of the gap above the Nb trimers. DFT+ reproduces the trimerization and the S2 shift ( mÅ), but moves S1 outward by mÅ, which cancels the effect on ; this explains why the minimum of is weaker in the calculation than in experiment.
| S site | ||||
| Exp. | S (3/4) | 0.11897(7) | 1.549 | 1.407 |
| S (1/4) | 0.11436(8) | 1.604 | 1.352 | |
| Exp. | S (all) | 0.13183(4) | 1.564 | 1.402 |
| DFT | S (3/4) | 0.11912 | 1.551 | 1.412 |
| S (1/4) | 0.11426 | 1.609 | 1.354 | |
| DFT | S (all) | 0.13212 | 1.566 | 1.397 |
Table S2. Heights ( in fractional units; in Å) of the S atoms above the Ni plane () and the Nb plane ().

Figure S4. Origin of the contraction of at relative to . (a) Nb plane of the single-crystal structure, viewed along . The Nb triangles contracted to 3.200 Å (red) are capped by the S atoms (diamonds); the other Nb–Nb distances are 3.315 and 3.429 Å (trimerization 3.4%). Ni atoms (green) sit in the van der Waals gap. (b) To-scale side views of the refined structures, with heights measured from the Ni plane. At all S atoms are equivalent; at they split into S (three in four, bonded to Ni) and S (one in four, no Ni neighbor), and the lower S reduces the mean gap from 2.804 to 2.786 Å. (c) Contributions to , in mÅ: the S and S height changes, each weighted by twice its fraction of the S atoms, and the S–Nb–S slab thickness . In the single crystals the S shift (−24.9) outweighs the outward S shift (+6.9) and the slab change (−2.4), giving −20.4 mÅ. In the relaxed DFT+U supercells the S shift (+21.6) cancels that of S (−21.6), leaving only −1.4 mÅ.
Third, the amplitude of the minimum is several times larger in experiment than in DFT (see above), and larger still in the best-ordered single crystal than in the powder (Sec. V A). The difference may come from the intercalant disorder that the calculation ignores: at the data are described by a single Ni site repeated in every gap, without detectable stacking faults, whereas at the partial occupation of the $2b$ site and the stacking faults of the Ni layers (Sec. IV C) make the bonding across the gaps less uniform; at itself, the degradation of the Ni stacking by grinding reduces the contraction of the powder with respect to the single crystal. A contraction of the stack is a collective response that requires every gap to be bonded in the same way, so that such disorder can only reduce it. The same argument explains why shows no contraction of , whereas the perfectly ordered stripe structure at has, in DFT, an even shorter than at $x = 1/4 ({xref}Table II`). First, the measured composition departs from , so that at least Ni per formula unit must occupy defect positions. Second, the stripe order itself need not be perfect in our sample. In both cases the bonding across the gaps is no longer uniform and the contraction is lost. The overall picture is thus that increases weakly with the Ni content, and that a minimum appears on top of this trend only where the Ni ions form a well-developed long-range order, as at . X-ray absorption spectroscopy at the Ni L and S K edges at and would determine how the Ni –S hybridization and the charge transfer to the host differ between the two structures, and in particular at the S2 sites that are not bonded to Ni.
Stripe order at
The single-crystal structure of also provides the key to the reappearance of reflections at . In NiNbS, Ni fills only the column of the quadrupled cell, leaving three octahedral positions per van der Waals gap empty. Filling one of these columns, at and , brings the composition to without changing the periodicity. The Ni ions then form chains of nearest neighbors along one of the three equivalent directions of the triangular lattice, separated by empty rows: the order becomes a stripe order, and selecting one direction lowers the symmetry from hexagonal to orthorhombic. The stripe order itself only requires the orthohexagonal cell used in the DFT calculation (Table II). Keeping in addition the Nb trimerization of the structure, the corresponding structure has space group (No. 63) with Å, Å and , i.e. the orthohexagonal setting of the same lattice (Table VI); a lower symmetry such as its subgroup cannot be excluded from the present powder data. This model accounts naturally for the orthorhombic splitting of the host reflections (Sec. III E): the relaxed DFT structure of the stripe order is indeed orthorhombic, with , four times the strain measured at , , as expected for a calculation that assumes a fully ordered, single-domain stripe phase, and since its lattice coincides with the one, it produces superstructure reflections at the same half-integer positions of the parent hexagonal cell. The model corresponds to , slightly less than the measured composition; the remaining Ni per formula unit is presumably accommodated in a disordered way, much like the of misplaced Ni at .
| Atom | Wyckoff | |||
| Ni1 | 4a | 0 | 0 | 0 |
| Ni2 | 4b | 0 | 0 | |
| Nb1 | 4c | 0 | 0 | |
| Nb2 | 4c | 0.99426 | ||
| Nb3 | 8g | 0.25861 | 0.24713 | |
| S1 | 16h | 0.25146 | 0.91618 | 0.11897 |
| S2 | 8f | 0.16667 | 0.11436 | |
| S3 | 8f | 0.66764 | 0.11897 |
Table VI. Model (No. 63) structure of the stripe order, obtained from the refined NiNbS structure by additionally filling the octahedral column, with Å, Å, Å and , in the standard setting of the C-centered cell. The refined Ni occupancy 0.944 is kept on both Ni columns.
The intensities calculated for this model are compared with those of the refined structure in Table VII. Filling the column doubles the structure factor of the reflections of a single domain, multiplying their intensity by four, while it extinguishes the and reflections, which the structure produces with comparable intensity. The three orientation variants of the stripes are equally probable, and averaging over them restores intensity at all half-integer positions. What survives the domain average is an overall enhancement of the superlattice reflections relative to the 002 host reflection, by a factor 2.2 for and 2.1 for . Part of this enhancement comes from the reference itself, since the 002 intensity drops by 38% when the additional Ni is inserted, the Ni and host contributions to this reflection being in antiphase; at a common scale factor the two superlattice reflections gain 37 % and 31 %, respectively. In either normalization the model predicts the stronger superlattice reflections observed at . The reflection provides a useful internal reference: the two Ni layers of the cell are related by the screw axis, so that their contributions cancel exactly for odd in both models, and this reflection measures only the distortion of the Nb–S framework.
Calculated intensities of the 002 host reflection and of three superlattice reflections, indexed in the parent hexagonal cell, for the refined NiNbS structure (, Table III) and for the stripe model (, Table VI). Structure factors include Cromer–Mann atomic form factors and the refined displacement parameters; powder intensities are calculated for Å and normalized to for each model. For , the value in parentheses is that of a single orthorhombic domain, the other one being the average over the three orientation variants.
| (Å) | ||||
| 002 | 5.912 | 2 | 55700 | 34800 |
| 5.741 | 6 | 2300 | 3160 (0) | |
| 5.164 | 12 | 87 | 87 (87) | |
| 3.314 | 6 | 2050 | 2680 (8040) | |
| Structure factors (arb. units) | ||||
| Powder intensities, | ||||
| 002 | 5.912 | 2 | 100 | 100 |
| 5.741 | 6 | 11.7 | 25.7 (0) | |
| 5.164 | 12 | 0.71 | 1.14 (1.14) | |
| 3.314 | 6 | 3.43 | 7.19 (21.6) |
Table VII.
The model also agrees with the correlation lengths: at the superstructure shows no measurable excess width along (SM33), so that the stripes are correlated in three dimensions, in contrast with the short-range stacking of the Ni layers at lower . The calculation of Table VII keeps the Nb and S positions of the structure, whose Nb trimerization will reorganize once the second Ni column is filled; the predicted intensities are therefore robust for the reflections dominated by the Ni sublattice but only indicative for those dominated by the host. A Rietveld refinement of the superlattice intensities of the pattern and a single-crystal study at this composition, which would also allow the selective extinctions to be observed within single domains, are the natural next steps.
Conclusion
We have carried out a composition-resolved structural study of NiNbS for , with Ni contents determined by EDX, combining synchrotron powder and single-crystal X-ray diffraction with DFT+ calculations. All samples retain the 2H-NbS host, with a small 3R fraction at only, while the ordering of Ni in the van der Waals gap follows a non-monotonic sequence: disorder at , order at , R30° order at , coexistence of the two at –, and an orthorhombic stripe order at . In all ordered phases except the stripe phase, the Ni order is much shorter-ranged along than in plane. The in-plane lattice parameter increases monotonically with , while the interlayer spacing responds weakly, saturates above , and shows a pronounced minimum at .
Single-crystal refinements show that NiNbS is described by a centrosymmetric structure () with Ni on a single site and no diffuse scattering, accompanied by a 3.4% trimerization of the Nb triangular net, whereas the chiral order of NiNbS () leaves the Nb sublattice undistorted but is intrinsically imperfect, with 9.0(2)% of the Ni on the nominally vacant site, a substantial inversion-twin fraction and stacking faults of the Ni layers. The contraction of at is carried by the S atoms that are not bonded to Ni and relax into the empty part of the van der Waals gap above the Nb trimers, the Ni–S bond itself being slightly longer than at . The relaxed DFT+ structures reproduce a local minimum of at , and an even shorter for the ordered stripe structure at : a well-developed long-range Ni order contracts the interlayer spacing, whereas intercalant disorder suppresses this contraction.
Two continuations follow naturally. Magnetic and transport measurements on the same composition series will show whether the structural singularities identified here, the single-site order and contracted stacking at , the intrinsic disorder at and the stripe order at , have counterparts in the magnetic and electronic response. On the structural side, single-crystal data at , which would determine the stripe structure, and an analysis of the diffuse scattering of the disordered sample, which would reveal the short-range Ni correlations that precede the commensurate superstructures, would complete the picture.
Acknowledgments
We acknowledge SOLEIL for providing synchrotron beamtime on CRISTAL (proposal 20251118) and the Paris Île-de-France Region in the framework of DIM MaTerRE (project DAC-VX). This work has been supported by the Croatian Science Foundation under the project numbers IP-2020-02-9666, IP-2025-02-7384. G. P. and P. P. acknowledge support of the project Cryogenic Centre at the Institute of Physics – KaCIF co-financed by the Croatian Government and the European Union through the European Regional Development Fund-Competitiveness and Cohesion Operational Programme (Grant No. KK.01.1.1.02.0012) and the project Ground states in competition – strong correlations, frustration and disorder — FrustKor financed by the Croatian Government and the European Union through the National Recovery and Resilience Plan 2021-2026 (NRRP).
Supplemental Material for “Ni ordering in the van der Waals gap of Ni-intercalated 2H-NbS: from disorder through and order to stripe order”
Naveen Singh Dhami; Ricardo Guillermo Torres Ramirez; Gaurav Pransu; Ivo Batistić; Pierre Fertey; Erik Elkaim; El-Eulmi Bendeif; Bruno Gudac; Nikolina Penić; Naveen Kumar Chogondahalli Muniraju; Neven Barišić; Petar Popčević; Victor Balédent
SM_v6.pdf
Orthorhombic splitting at
Correlation lengths in the powders
Figure S2 and Table S1 give the correlation lengths of the host and of the Ni superstructure in the powders, in plane and along , obtained from all resolved reflections. The host is well correlated in every sample: – Å with no significant dependence on , and – Å for –. The only host anomaly is a reduction of to about Å at , where the lattice becomes orthorhombic. The superstructure correlation lengths are defined differently: they are obtained after removing the host width at the same , so that they measure the additional disorder of the Ni sublattice with respect to the framework, and should not be compared directly with the host values.

Figure S2. Correlation lengths of the NbS2 host (circles) and of the Ni superstructure (squares) in NixNbS2 powders, (a) in plane and (b) along c, versus Ni content (EDX). Host: mean standard error over the accepted reflections. Superstructure: anisotropic fit to all resolved superlattice reflections after removal of the host width, with error bars from leave-one-reflection-out fits. Dotted lines indicate approximate boundaries between ordering regimes.
| Host | Host | SS | SS | ||
| 0.18 | (5) | (5) | – | – | – |
| 0.25 | (5) | (3) | 9 | ||
| 0.33 | (5) | (3) | 8 | ||
| 0.37 | (5) | (4) | – | – | – |
| 0.46 | (6) | (4) | 7 | ||
| 0.57 | – | (3) | 7 |
Table S1. Correlation lengths (Å) shown in Fig. S2. Host: mean over instrument-corrected (in-plane) or (along ) reflections, error = standard error of the mean ( in parentheses). Superstructure (SS): anisotropic fit of the excess width of superstructure reflections, error = jackknife standard error. *Upper error bar set at the largest finite leave-one-out value.
In plane, the Ni order is correlated over several hundred Å at all compositions where a superstructure is observed ( Å at , 420 Å at , and 730 Å at ; the value at is poorly constrained). Along the picture is very different. For – the superstructure is correlated over only – Å, i.e. about ten unit cells, five to fourteen times less than the host in the same direction. The stacking of the Ni layers from one van der Waals gap to the next is therefore the weakest correlation in these samples, consistent with the diffuse rods along observed in the single-crystal data at (§ 5 C of the main text). The exception is , where the superstructure shows no measurable excess width along ( Å), so that the Ni stripes are correlated in three dimensions. This agrees with the better-ordered state inferred from the intensities and widths of the half-integer reflections.
At the host reflections split into orthorhombic doublets (Figure S1), so that no in-plane host width could be extracted, and the host reference used to subtract the instrumental and strain broadening at this composition is built from the reflections alone; the superstructure values at should therefore be regarded as provisional. The widths quoted in Table V of the main text are Scherrer widths of single reflections, without subtraction of the host width; they lead to the same conclusion, namely superstructure reflections much broader than the host reflections along .
Comparison of with other intercalated dichalcogenides
Figure S3 compares the evolution of the lattice parameter with intercalant content in 3d-intercalated NbSe2, TaS2 and NbS2.

Figure S3. Evolution of the lattice parameter with intercalant content in (a) NbSe, (b) TaS and (c) NbS ( = Fe, Cu, Co, Ni); panel (d) enlarges the –11.98 Å region of (c). Symbol colors denote the intercalant (Fe red, Cu teal, Ni blue, Co purple). Filled symbols are powder diffraction, open symbols single-crystal diffraction, and half-filled symbols data for which the sample type is not stated. Stars mark the powder values of van den Berg and Cossee[1] (as tabulated in Ref.[2]), and open squares the single-crystal values reported for NiNbS[3] and CoNbS[4]. Our NiNbS data (blue circles) use the EDX composition and show powder (filled) and single-crystal (open) values. The gray band in (a) marks the superstructure crossover. Data are from Ref.[5] (FeNbSe),[6] (CuNbSe),[7] (NiNbSe),[8] (CoNbSe),[9] (CuTaS),[10] (NiTaS),[11] (CoTaS),[12] (FeNbS),[13] (CuNbS) and[14, 15] (CoNbS). Values from Refs.[6, 9, 10, 13, 14, 15] were digitized from published figures, so that their uncertainty is about the symbol size. Whereas grows monotonically upon Fe and Cu intercalation, the Ni and Co compounds of NbS remain close to the pristine value, and our Ni data show a minimum at followed by a plateau for .
Structural response to the order at
The -axis dip at is accompanied by a distinct distortion of the host. In the cell the Nb2 atoms (2b) lie directly beneath the Ni atoms and remain on the ideal lattice, whereas the Nb1 atoms (6h) are displaced by 0.066 Å to form Nb3 trimers with a short bond of 3.2005(6) Å (Figure S4). Defining the trimerization as , where Å is the in-plane lattice parameter of the cell, so that Å is the Nb–Nb distance of the undistorted triangular lattice, we obtain 3.4% at , compared with at . A similar trimerization is found in FeTaS (short Ta–Ta bond 3.209 Å against the undistorted Å, i.e. 3.0%)[16], whose is also smaller than at (−0.13 Å). There the gap shrinks ( versus 3.071 Å) while the slab thickens ( versus Å)[17], whereas in the slab changes by only mÅ.
Defining the interlayer period as , with and averaged over the S sites, the contraction mÅ ( relative to ) comes almost entirely from the gap, mÅ ( versus Å), the slab thickness changing by only mÅ ( versus Å) (Figure S4). This contraction is not carried by the Ni–S bond, which is slightly longer at , against Å ( mÅ). At the S layer bordering the gap splits into S1 (12k, bonded to Ni) and S2 (4f, bonded to no Ni), which caps the Nb3 trimer. S1 stays within 5 mÅ of the height of the S atoms at , whereas S2 lies 0.050 Å closer to the Ni plane (Table S2); the S2 displacement alone contributes mÅ to , partly compensated by mÅ from the outward shift of S1. The trimerization and the gap contraction thus occur at the same S2 sites. Relaxed DFT+ supercells reproduce both the trimerization and the S2 shift ( mÅ), supporting a link between the two, but they also move S1 outward by 14 mÅ, which cancels the effect on and leaves a much weaker minimum of ; a direct test of the causal relation would require a bonding analysis of the distorted and undistorted structures.
Rietveld refinement for selected compositions
The Rietveld refinement for selected compositions are presented here.

Figure S5. Rietveld refinements of synchrotron powder XRD patterns (0.6779 Å) for (a) NiNbS, with the 2H host refined in P/mmc and 3R-NbS and α-S included as secondary phases, and (b) NiNbS, with the 2H host refined in the ordered P22 setting. Open circles: observed; red line: calculated; blue line: difference; vertical ticks: Bragg positions.
References
- [1]J. M. van den Berg and P. Cossee, Inorg. Chim. Acta 2, 143 (1968).
- [2]K. Anzenhofer, J. M. van den Berg, P. Cossee, and J. N. Helle, J. Phys. Chem. Solids 31, 1057 (1970).
- [3]Y. Utsumi Boucher, I. Bialo, M. A. Gala, W. Tabiś, M. Rosmus, N. Olszowska, J. J. Kolodziej, B. Gudac, M. Novak, N. K. Chogondahalli Muniraju, I. Batistić, N. Barišić, P. Popčević, and E. Tutiš, Phys. Rev. B 109, 085135 (2024).
- [4]P. Popčević, Y. Utsumi, I. Bialo, W. Tabiś, M. A. Gala, M. Rosmus, J. J. Kolodziej, N. Tomaszewska, M. Garb, H. Berger, I. Batistić, N. Barišić, L. Forró, and E. Tutiš, Phys. Rev. B 105, 155114 (2022).
- [5]X. Xu, B. Jiang, R. Wang, Z. Qiu, S. Guo, B. Lv, and R. Zhong, Phys. Rev. B 113, 134420 (2026).
- [6]H. Luo, J. Strychalska-Nowak, J. Li, J. Tao, T. Klimczuk, and R. J. Cava, Chem. Mater. 29, 3704 (2017).
- [7]A. S. Wadge, A. Kazakov, X. Gong, D. Jastrzebski, B. J. Kowalski, A. Lynnýk, L. Plucinski, A. Fak h redine, R. Diduszko, M. Aleszkiewicz, J. Korczak, D. Wutke, M. Rosmus, R. Kurleto, N. Olszowska, C. Autieri, and A. Wisniewski, Phys. Rev. B 112, 235129 (2025).
- [8]H. C. Mandujano, P. Y. Zavalij, A. Manjón-Sanz, H. Cao, and E. E. Rodriguez, Phys. Rev. Mater. 9, 014403 (2025).
- [9]K. E. Wagner, E. Morosan, Y. S. Hor, J. Tao, Y. Zhu, T. Sanders, T. M. McQueen, H. W. Zandbergen, A. J. Williams, D. V. West, and R. J. Cava, Phys. Rev. B 78, 104520 (2008).
- [10]X. D. Zhu, Y. P. Sun, S. B. Zhang, H. C. Lei, L. J. Li, X. B. Zhu, Z. R. Yang, W. H. Song, and J. M. Dai, Solid State Commun. 149, 1296 (2009).
- [11]P. Park, W. Cho, C. Kim, Y. An, M. Avdeev, K. Iida, R. Kajimoto, and J.-G. Park, Phys. Rev. B 109, L060403 (2024).
- [12]S. Wu, Z. Xu, S. C. Haley, S. F. Weber, A. Acharya, E. Maniv, Y. Qiu, A. A. Aczel, N. S. Settineri, J. B. Neaton, J. G. Analytis, and R. J. Birgeneau, Phys. Rev. X 12, 021003 (2022).
- [13]G. Wu, T. Wu, Z. Li, L. Zhao, R. H. Liu, H. Chen, D. F. Fang, J. L. Luo, and X. H. Chen, J. Phys.: Condens. Matter 21, 275601 (2009).
- [14]M. Nakayama, K. Miwa, H. Ikuta, H. Hinode, and M. Wakihara, Chem. Mater. 18, 4996 (2006).
- [15]W. Cho, K. Nam, Y. An, Y. Y. Kim, M.-H. Jung, K. H. Kim, and J.-G. Park, arXiv:2603.26069 (2026).arxiv.org/abs/2603.26069
- [16]E. Morosan, H. W. Zandbergen, L. Li, M. Lee, J. G. Checkelsky, M. Heinrich, T. Siegrist, N. P. Ong, and R. J. Cava, Phys. Rev. B 75, 104401 (2007).
- [17]J. L. Musfeldt, J. Yang, R. E. Putnam, S. Zhang, J. T. Haraldsen, S. W. Cheong, and Z. Liu, npj 2D Mater. Appl. 10, 3 (2025).