Preparation and characterization of metastable trigonal layered MSb 2 O 6 phases (M = Co, Ni, Cu, Zn, and Mg) and considerations on FeSb 2 O 6 †

MSb 2 O 6 compounds (M = Mg, Co, Ni, Cu, Zn) are known in the tetragonal trirutile forms, slightly distorted monoclinically with M = Cu due to the Jahn – Teller e ﬀ ect. In this study, using a low-temperature exchange reaction between ilmenite-type NaSbO 3 and molten MSO 4 – KCl (or MgCl 2 – KCl) mixtures, these ﬁ ve compositions were prepared for the ﬁ rst time as trigonal layered rosiaite (PbSb 2 O 6 )-type phases. Upon heating, they irreversibly transform to the known phases via amorphous intermediates, in contrast to previously studied isostructural MnSb 2 O 6 , where the stable phase is structurally related to the metastable phase. The same method was found to be applicable for preparing stable rosiaite-type CdSb 2 O 6 . The formula volumes of the new phases show an excellent correlation with the ionic radii (except for M = Cu, for which a Jahn – Teller distortion is suspected) and are 2 – 3% larger than those for the known forms although all coordination numbers are the same. The crystal structure of CoSb 2 O 6 was re ﬁ ned via the Rietveld method: P 3¯1 m , a = 5.1318(3) Å, and c = 4.5520(3) Å. Compounds with M = Co and Ni antiferromagnetically order at 11 and 15 K, respectively, whereas the copper compound does not show long-range magnetic order down to 1.5 K. A comparison between the magnetic behavior of the metastable and stable polymorphs was carried out. FeSb 2 O 6 could not be prepared because of the 2Fe 2+ + Sb 5+ = 2Fe 3+ + Sb 3+ redox reaction. This electron transfer produces an additional 5s 2 shell for Sb and results in a volume increase. A comparison of the formula volume for the stable mixture FeSbO 4 + 0.5Sb 2 O 4 with that extrapolated for FeSb 2 O 6 predicted that the trirutile-type FeSb 2 O 6 can be stabilized at high pressures.


Introduction
Preparation of a new crystal form for a well-known composition is challenging and very interesting because it can provide new properties and improve our understanding of the composition-structure-property relations. Recently, using a low-temperature ion-exchange reaction, a new form of MnSb 2 O 6 , isostructural with rosiaite PbSb 2 O 6 , was prepared and its magnetic properties and thermal transformations were studied. 1 In this structure, magnetic cations are arranged in trigonal layers that antiferromagnetically order at low temperatures, forming spin-frustrated triangles.
In this study, the same preparative approach was expanded to the entire MSb 2 O 6 family (M = Mg, Fe, Co, Ni, Cu, Zn, and Cd) and the magnetic properties of the new phases with M = Ni, Co, and Cu were studied. A total of five of the seven listed compositions, with M = Mg, Co, Ni, Cu, and Zn, are known to crystallize in the tetragonal trirutile-structure type, 2 which is slightly monoclinically distorted for M = Cu due to the Jahn-Teller effect, whereas CdSb 2 O 6 is found in the rosiaite form. 3 MSb 2 O 6 compounds with a partially filled d subshell (M = Co, Ni, and Cu) have attracted attention due to their magnetic properties, [4][5][6][7][8][9][10][11][12][13] whereas antimonates with nonmagnetic ions (M = Mg, Zn, and Cd) may be interesting as photocatalysts, 14,15 transparent conductors, 16 and/or sensors. [17][18][19] Initially, FeSb 2 O 6 was also reported to exist in the trirutile form; † Electronic supplementary information (ESI) available. See DOI: 10.1039/ c6dt04859e a however, later studies showed that the mineral known as FeSb 2 O 6 is actually FeSbO 4 (Fe 1/2 Sb 1/2 O 2 ) with disordered rutile-type structure. 20 An attempt to prepare FeSb 2 O 6 via solid-state reactions of Sb 2 O 3 and FeSO 4 ·7H 2 O at 1000°C in air was made and a mixture of FeSbO 4 and Sb 2 O 4 was obtained. 20 This result is explained by the oxidation of Fe 2+ in air and does not preclude an attempted preparation in an inert atmosphere at much lower temperatures.

Sample preparation and analysis
Only reagent-grade starting materials were used. NaSbO 3 was prepared at 550-850°C from a stoichiometric mixture of dried Sb 2 O 3 , Na 2 CO 3 , and NaNO 3 , as previously reported, 1,21 and its phase purity was confirmed via powder X-ray diffraction (XRD). MSO 4 ·xH 2 O (M = Fe, Co, Ni, Cu, Zn, and Cd) was dried at 400°C (in flowing argon atmosphere for FeSO 4 ). Subsequently, NaSbO 3 , MSO 4 , and KCl were carefully mixed using a mortar and pestle at the molar ratio of approximately 1 : 2.5 : 5, providing a five-fold excess of MSO 4 against the stoichiometric amount. For M = Cd, a double excess was also tested, with a 1 : 1 : 2 ratio of NaSbO 3 , CdSO 4 , and KCl. In some instances, half of the KCl was substituted with KBr to further lower the solidus temperature.
For M = Mg, this approach was inapplicable due to the excessively high eutectic temperature of the MgSO 4 -KCl join, 665°C. 22 Therefore, a MgCl 2 -KCl mixture was used instead.
The mixtures were heated for various periods at various temperatures in air (except for M = Fe, for which a CO 2 atmosphere was used), cooled, washed with warm water, dried, and examined via XRD.
The elemental composition of the samples was estimated using an electron microprobe (INCA ENERGY 450/XT) equipped with an X-Act ADD detector based on an electron microscope VEGA II LMU (Tescan) operated at the accelerating voltage of 20 kV. No potassium was detected; however, some samples showed a non-zero sodium content. The molar ratios M/Sb and Na/Sb were averaged on 8-11 points in each sample.

Diffraction studies
XRD was performed using Cu Kα radiation by an ARL X'TRA diffractometer equipped with a solid-state Si(Li) detector eliminating Co and Fe fluorescence and other undesired wavelengths (although sometimes very weak reflections from CuK β and WL α radiations were observed). Rietveld refinements were performed using the GSAS + EXPGUI suite. 23,24 Neutron diffraction experiments were carried out via the cold neutron two-axis diffractometer G4.1 located at ORPHEE reactor, Laboratory Léon Brillouin, Saclay, France. The wavelength of the incident monochromatic neutrons was 2.426 Å. Neutron diffraction data were obtained in the 2θ range from 6.0 to 85.9°with the step of 0.1°. The measurements were carried out using the special Orange cryostat for neutron research. The sample was placed in a cylindrical vanadium container having a diameter of 6 mm. The diffraction pattern was treated by the Rietveld method using the FullProf suite. 25

Magnetic measurements
The magnetic properties (temperature dependencies of the magnetic susceptibility and magnetization isotherms) of the MSb 2 O 6 (M = Co, Ni, and Cu) samples in the temperature range from 2 to 350 K under a magnetic field up to 9 T were measured using Quantum Design MPMS-7T and VSM option of the Quantum Design PPMS-9T devices. The electron spin resonance spectra at room temperature were obtained using an X-band ESR spectrometer CMS 8400 (ADANI) ( f ≈ 9.4 GHz, B ≤ 0.7 T). The effective g-factors of the samples were calculated with respect to an external reference for the resonance field. BDPA (a,g-bisdiphenyline-b-phenylallyl), g ref = 2.00359, was used as the reference material.

Preparation, identification, and phase transformations
As expected, reactions of NaSbO 3 with low-melting mixtures of MSO 4 (M = Co, Ni, Cu, and Zn) or MgCl 2 with KCl (or KCl-KBr) resulted in formation of five new trigonal layered phases of the rosiaite type (Table 1, Fig. 1). However, this required careful selection of the temperature and duration, typically in the ranges of 420-500°C and 40-80 minutes. The lower limits were determined by the solidus temperature of the salt mixture and incomplete reaction, whereas the higher limits were determined by the MSb 2 O 6 transformation to the stable form. Moreover, the results were also dependent on the total mass of the reaction mixture. Because the furnace temperature was set only slightly higher than the solidus temperature, complete melting of the salt mixture required considerable time, especially long with large batches. To prepare 1-2 g of the desired product in one stage, a 5-10 g batch was required, which usually resulted in either incomplete exchange or (at least partial) conversion to the stable trirutile-type polymorph.
The same approach was applied to the preparation of CdSb 2 O 6 , for which a rosiate-type structure is the stable form. Thus, the limitations of reaction temperature and time were unnecessary, and the exchange was performed at 500°C for 2 h. This resulted in the formation of single-phase products with both 2-and 5-fold excess of CdSO 4 , marked I and II, respectively. Both have much better crystallinity than others (Fig. 1g) and the lattice parameters are in good agreement with literature data (Table 1).
In contrast to the samples prepared from sulfates, our MgSb 2 O 6 contained admixtures of foreign crystalline phases including MgO (Fig. 1c). This is because we failed to avoid hydrolysis while drying the starting hydrous MgCl 2 .
The analytical data, as shown in Table 1, despite considerable scattering from point to point, confirm that obtained NiSb 2 O 6 and CuSb 2 O 6 are sufficiently pure: the M/Sb ratio is stoichiometric within the experimental accuracy and the Na content is essentially zero. On the other hand, the MgSb 2 O 6 contained foreign phase(s) even after the removal of MgO via acid leaching. In addition to XRD, this is evident from the microprobe data showing the elevated average Mg/Sb ratio and especially high scatter.
With other three compounds, the situation was less obvious. The single analyzed sample of ZnSb 2 O 6 and one of the two CoSb 2 O 6 samples (labelled II) showed somewhat reduced M/Sb ratio (although within the experimental accuracy) associated with a non-zero sodium content in both Co samples (in the presence of Zn, determination of Na was unreliable). This suggests the possibility of sodium substi-tution for M 2+ and correlates with variations in the XRD data. A comparison of the XRD patterns of CoSb 2 O 6 I ( Fig. 1e) and CoSb 2 O 6 II ( Fig. 1f ) shows that the less pure sample CoSb 2 O 6 II, in contrast to CoSb 2 O 6 I, does not show splitting of the pairs 001 + 100 and 002 + 111, indicating a considerably lower c/a ratio. Similar variations were observed with the ZnSb 2 O 6 samples, but those with unresolved 001 + 100 reflections have not been reported herein. The different behaviours of Zn and Co compounds compared to those of Ni and Cu might be tentatively attributed to the somewhat larger ionic radii of the former pair of cations, which is favourable for sodium substitution. However, ionic radius of Cd 2+ is much larger and closer to that of Na + , but no Cd deficiency indicating sodium substitution can be seen in Table 1, especially with sample I, where a lower excess of the Cd salt might favour incomplete ion exchange. Therefore, we hypothesized that the elevated sodium content in the CdSb 2 O 6 samples might be due to the adsorbed or occluded salts rather than ionic substitutions. Note that the sodium content in CdSb 2 O 6 (I) diminishes on calcination, probably due to volatilization of the salt impurity. The role of Na in MSb 2 O 6 has been further discussed in Section 3.2.
It was noticed that the transformation to the stable polymorph was markedly enhanced in the presence of a salt melt, obviously due to the dissolution-precipitation mechanism. As illustrated in the XRD patterns in Fig. 2, for ZnSb 2 O 6 , taken as a typical example, prolonged treatment with the molten salt at 420-500°C resulted in the conversion of the metastable trigonal layered rosiaite-type phase into the stable trirutile-type polymorph. On the other hand, this conversion was not observed at considerably higher temperature of 750°C if ZnSb 2 O 6 was free of the salt mixture. Note that exchange reaction of NaSbO 3 with ZnCl 2 under hydrothermal conditions yielded stable ZnSb 2 O 6 at even lower temperature of 230°C (for 4 days), whereas phase-pure products with M = Pb, Cu, Ni, and Mn could not be obtained via this method. 14 A comparison of the patterns 1 and 4 depicted in Fig. 2 shows that annealing of ZnSb 2 O 6 at 750°C resulted in an a Treated with aqueous acetic acid to remove free MgO before analysis. b Sodium content could not be estimated due to overlap of its strongest line with the line from Zn. c Extremely inhomogeneous distribution of Na: only found in 2-4 points of 10 or 11.  Fig. 2). No additional phases were found within the sensitivity of the XRD. This means that the amorphous phases have the same composition as the crystalline products, being intermediates in the transformation between the trigonal and tetragonal phases. This is different from the previously studied case of MnSb 2 O 6 , where the stable phase was structurally related to the metastable phase and the transformation might have proceeded within an intact crystal. 1

Structural characterization
As shown in Table 1 and Fig. 4, the formula volumes of trigonal MSb 2 O 6 (except for M = Cu) have an excellent correlation with the ionic radii of M 2+ , and for the five new phases, they are 2-3% larger than those for the stable trirutile-type phases although all coordination numbers are the same. Fig. 4 presents an additional example of a morphotropic series where substitution of a larger cation (herein, Co 2+ in CoSb 2 O 6 ) by an even larger cation (herein, Mn 2+ or Cd 2+ ) leads to less dense structure types, 28 opposite to the widely assumed homology rule.
As discussed earlier, 1 the ilmenite-type structure of NaSbO 3 and rosiaite-type structure of MSb 2 O 6 (M = Ca, Cd, Sr, Pb, and Ba) are based on the same octahedral SbO 6/2 layers. Therefore, M 2+ substitution for 2Na + may occur by mere gliding of the layers to avoid the face sharing of MO 6 and SbO 6 octahedra. This gliding may result in stacking faults. Therefore, it is not surprising that XRD peaks in the powder patterns ( Fig. 1) are considerably broader than those of starting NaSbO 3 . Largest broadening is observed with M = Cu. This, together with the deviation seen in Fig. 4, may be an indication of lowered symmetry due to the Jahn-Teller effect characteristic of Cu 2+ . Inset in Fig. 1 shows that the reflection 100 that may split upon transition from the trigonal to any lower crystal system is markedly broader than 001, which does not split in any case. However, the quality of the powder pattern precludes a detailed structural investigation.    Only the crystal structure of CoSb 2 O 6 (the sample I with minimum sodium impurity, if any, see Table 1) was refined using the Rietveld method. The main results are listed in Tables 2-4 and in the ESI † (cif ). Fig. 5 compares the experimental and calculated XRD profiles, whereas Table 4 and Fig. 6 compare the new and previously known structures of CoSb 2 O 6 .
A comparison of the bond lengths and bond angles in the two polymorphs (Table 4) shows that both the CoO 6 and SbO 6 octahedra, especially the former, are much more regular in the new phase than in the stable trirutile-type counterpart. The same was noticed for MnSb 2 O 6 1 although its stable form is not of the trirutile type. The trigonal structure of CoSb 2 O 6 is based on the distorted hexagonal close packing (better to say, 2H eutaxy) of oxygen anions, where cations fill half octahedral voids. Each oxygen atom has 10 oxygen neighbours at the distances of 2.61-2.98 Å plus two oxygen neighbours more distant at 3.31 Å. Although the rutile and trirutile structure types are also often described as 2H derivatives, it is actually incorrect. 2 Indeed, the two independent anions in tetragonal CoSb 2 O 6 5 have oxygen environments of 11 + 2 and 10 + 1 + 2; thus, the tetragonal close packing with 11 (rather than 12) neighbours is a more adequate term in this case. 2 The cation arrangements in the two polymorphs also significantly differ. In the trirutile-structure type, each metal-oxygen octahedron shares two edges with other octahedra, and higher-valence cations have only one common edge. In the rosiaite-structure type, the number of shared edges is the same only as an arithmetic mean. Octahedra populated with lower-valence cations have no shared edges at all, whereas those populated with higher-valence cations share three edges each (Fig. 6). In addition, the trigonal form, despite the larger formula volume, exhibits shorter Sb-Sb distances across the shared edges: 2.96 Å in contrast to 3.05 and 3.11 Å for Sb-Sb and Sb-Co, respectively, in the tetragonal form. All this leads to stronger cation-cation repulsion in the rosiaite type and explains its instability with respect to the trirutile type. For large M 2+ (Cd⋯Ba), size disparity with Sb 5+ precludes the formation of the trirutile type; then, the rosiate type becomes stable. The crystal-chemistry role of the sodium impurity remains unclear. Previously, 1 we suggested the existence of small nonexchanged ilmenite-type fragments as stacking faults in the main MnSb 2 O 6 phase. However, this disagrees with the lattice parameter changes. Pure NaSbO 3 has a c/a ratio of 3.011 for the three-layered rhombohedral packing. 29 For the primitive   one-layered packing, it should be 3.011/3 = 1.004. These admixtures must increase the c/a ratio of our products. However, a comparison of the samples I and II of CoSb 2 O 6 in Table 1 shows the opposite. As discussed in Section 3.1, sodium substitution for M 2+ was most probable with M = Cd but could not be supported by the analytical data and lattice parameters (Table 1). In addition, the introduction of large Na + cations in the interlayer is also expected to increase the c/a. Moreover, to provide charge balance, there should be either additional Na + ions (for which we do not see available sites) or holes that can be easily interpreted as Co 3+ ; however, this is hardly applicable for zinc or cadmium. The problem could not be solved by powder XRD owing to the very small content of sodium, its relatively small X-ray scattering factor, and diffuse nature of the diffraction patterns. The crystal structure refinement reported herein is based on the ideal composition CoSb 2 O 6 .

Special case of FeSb 2 O 6
Our low-temperature preparation of FeSb 2 O 6 in an inert atmosphere differs from the preceding high-temperature preparation in air: 20 instead of the mixture FeSbO 4 + 0.5Sb 2 O 4 , we obtained FeSbO 4 + 0.5Sb 2 O 3 (Fig. 7), due to the reducing power of the excess FeSO 4 . In any case, FeSb 2 O 6 could not be prepared, due to incompatible oxidation states, reacting according to the following scheme: 2Fe 2+ + Sb 5+ → 2Fe 3+ + Sb 3+ . This electron transfer from Fe 2+ (3d 6 ) to Sb 5+ (4d 10 ) creates an additional 5s 2 shell for Sb and thus should result in an increase in the total volume. The formula volume of the hypothetical trirutile-type FeSb 2 O 6 was estimated to be approximately 101.4 Å 3 by extrapolation using the plot in Fig. 4 and Shannon's ionic radius for high-spin Fe 2+ . The volume of the equivalent stable mixture FeSbO 4 + 0.5Sb 2 O 4 is considerably higher, 103.9 or 104.4 Å 3 , obtained using the data for monoclinic or orthorhombic Sb 2 O 4 , 30 respectively. Therefore, trirutile-type FeSb 2 O 6 , which is unstable at normal pressure, may be stabilized at high pressures although this may be complicated by high-spin-to-low-spin transitions.

Magnetic properties
3.4.1. NiSb 2 O 6 . The temperature-dependence of the magnetic susceptibility χ = M/B of NiSb 2 O 6 , measured at B = 0.1 T, is shown in Fig. 8(a). In both the field-cooled (FC) and zerofield cooled (ZFC) measurement protocols, it passes through smooth maximum at ∼19.5 K, indicating an onset of antiferromagnetic long-range ordering. The Néel temperature, as estimated from the first derivative dχ/dT (T ), is noticeably lower, T N = 15 K, than the maximum on χ(T ), which is characteristic of low-dimensional magnets with strong short-range correlations. Below T N , the χ ZFC (T ) and χ FC (T ) dependencies demonstrate clear divergence and additional growth of the magnetic susceptibility, signalling the presence of non-negligible spinglass and/or impurity effects. This is also confirmed from the ESR data, which reveal the superposition of two resonance modes, which are most probably related to the main signal (broad signal) and impurity line (narrow signal) as shown in the left inset in Fig. 8(a). The fit of the experimental χ(T ) data in the high-temperature range with the Curie-Weiss law χ = χ 0 + C/(T − Θ) allows an estimation of the temperature-independent term as χ 0 = 4.87 × 10 −5 emu mol −1 , Curie constant C = 1.3 emu K mol −1 , and Weiss temperature Θ = −33 K. The temperature-independent term χ 0 takes a positive value, indicating the predominance of the van Vleck contribution of Ni 2+ ions   over the sum of the Pascal constant of the individual ions χ dia = −1.12 × 10 −4 emu mol −1 . 31 The effective magnetic moment calculated from the obtained Curie constant takes μ eff = 3.2μ B /Ni 2+ . This value is in reasonable agreement with the theoretical estimation according to μ ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi where the average effective g-factor was directly measured via electron spin resonance (ESR) g = 2.21, which is well consistent with typical for Ni 2+ values.
The negative sign of the Weiss temperature indicates the predominance of antiferromagnetic exchange interactions in the system, in agreement with the long-range antiferromagnetic ordering. The field dependence of the magnetization of NiSb 2 O 6 ( Fig. 8b) demonstrates characteristic S-type behaviour at low temperatures, which indicates the competing exchange interactions of both the ferro-and antiferromagnetic nature.
Interestingly, note that the magnetic properties observed herein for this novel phase possessing the trigonal layered rosiaite-type structure are essentially different from those reported for its stable polymorph with the tetragonal trirutiletype structure (space group P4 2 /mnm). The magnetic behaviour of all the trirutile-type compounds MSb 2 O 6 (M = Ni, Co, and Cu) is quite similar. Their low-dimensional magnetic behaviour was confirmed from the magnetic susceptibility, specific heat and thermal expansion measurements, ESR and NMR/ NQR techniques, etc. as well as by electronic structure calculations. A characteristic feature is the presence of a wide temperature range, where all MSb 2 O 6 with a trirutile structure exhibit short-range antiferromagnetic order. For the case of trirutile NiSb 2 O 6 , it was found that the χ(T ) demonstrates a clear low-dimensional broad maximum at T max ∼ 40 K followed by long-range antiferromagnetic order below T N = 2.5 K. 4 Despite the almost perfect 2D square Ni 2+ sublattice, the data were well described within the formalism of antiferromagnetic S = 1 1D spin chain via Ni-O-O-Ni superexchange pathways with the main exchange parameter J/k B ∼ −45 K.
3.4.2. CoSb 2 O 6 . Similar to NiSb 2 O 6 , both CoSb 2 O 6 samples revealed an onset of antiferromagnetic long-range order at low temperatures with approximately the same Néel temperature, as high as 11 K (Fig. 9), although sample II shows a somewhat larger discrepancy between the ZFC and FC results, which is in agreement with the higher Na impurity level. Once again, similar to the nickel sample, the new rosiaite-type CoSb 2 O 6 polymorph does not show any sign of a low-dimensional maximum on χ(T ) in contrast to the tetragonal trirutile phase with short-range ordering at T max ∼ 35 K, preceding the antiferromagnetic ordering at T N = 13.5 K. [5][6][7] Opposite to the known trirutile phase, which was treated in the frame of the Ising dimers model with J/k B ∼ −10 K and spin-gap Δ ≈ 34 K, 7 the magnetic susceptibility of the new rosiaite-type CoSb 2 O 6 nicely follows the Curie-Weiss law over the wide temperature range above T N (Fig. 9). On the other hand, the main parameters of the magnetic subsystem, obtained from the experimental data, are comparable for both structure types. In particular, the temperature-independent term χ 0 = 7.3 × 10 −5 emu  mol −1 for the new sample indicates a prevailing van Vleck contribution of Co 2+ over the diamagnetic counterpart of ions constituting CoSb 2 O 6 ; the Weiss temperature Θ ≈ −38 K is consistent with the predominance of antiferromagnetic exchange interactions and very close to the value ∼−32 K reported for the trirutile sample. 7 The effective magnetic moment μ eff ≈ 4.9μ B /Co 2+ is in reasonable agreement with the theoretical estimations using the average effective g-factor g av = 2.3 ± 0.1 obtained from the anisotropic ESR absorption line (left inset of Fig. 9). Detailed investigations of the magnetic properties of the new rosiaite-type CoSb 2 O 6 polymorph have been planned to be reported later together with its magnetic structure, which is now under study via low-temperature neutron diffraction.
3.4.3. CuSb 2 O 6 . On comparing the magnetic behaviour of the trirutile-and rosiaite-type MSb 2 O 6 families, the most drastic difference was observed for M = Cu. The trirutile counterpart undergoes a structural phase transition from the high-temperature tetragonal phase (space group P4 2 /mnm) to the low-temperature monoclinic phase (space group P2 1 /n) at T ph = 380 K. 8,9 However, it has close similarity to its trirutiletype relatives, NiSb 2 O 6 and CoSb 2 O 6 : 1D short-range order was observed with T max = 60 K followed by the long-range ordering at T N = 8.5 K. 9,10 Susceptibility measurements performed on both the powder and single crystal samples of CuSb 2 O 6 fit extremely well over a wide temperature range to a nearestneighbour-only S = 1/2 uniform Heisenberg 1D chain model with an exchange constant ranging from −86 K to −98 K (see ref. 11 and references therein).
Moreover, the latter low-temperature phase transition has attracted significant interest because it was suggested to be possible spin-Peierls-like phase transition. 9,12 The octahedral oxygen environment of the magnetic Cu 2+ ions gives rise to the dominant Cu-O-O-Cu superexchange pathway, which forms magnetic chains along the (a + b) direction at z = 0, and (a − b) direction at z = 0.5 13 with the main intrachain exchange integral J ∥ /k B ∼ −94 K (ref. 13) and remarkably small ratio J ∥ /J ⊥ = 120. 6 The temperature-dependence of the magnetic susceptibility χ = M/B of new rosiaite-type CuSb 2 O 6 , measured at B = 0.1 T, is shown in Fig. 10(a). This follows the Curie-Weiss behaviour at elevated temperatures and does not show clear signature of any long-range order transition down to 2 K. However, left-side bending at low temperatures and a clear deviation from the Curie-Weiss law at T < 100 K can be observed, indicating rather strong short-range exchange interactions in this compound. The dependencies of magnetic susceptibility measured in the field-cooled (FC) and zero-field cooled (ZFC) regimes do not demonstrate any divergence, confirming the absence of any spin disorder effects. The fit of the experimental data in the high-temperature range with the function χ = χ 0 + C/(T − Θ) allows an estimation of the temperature-independent term χ 0 = 1.46 × 10 −5 emu mol −1 , Curie constant C = 0.447 emu K mol −1 , and Weiss temperature Θ = −4 K.
Similar to the NiSb 2 O 6 and CoSb 2 O 6 samples, the temperature-independent term χ 0 comprises diamagnetic Pascal and paramagnetic van Vleck contributions. The positive value of χ 0 indicates the predominance of the van Vleck term χ vV = 7.18 × 10 −4 emu mol −1 over the sum of the Pascal constant of individual ions χ dia = −1.12 × 10 −4 emu mol −1 . 31 The van Vleck contribution is defined by the splitting of the d-shell of Cu 2+ ions in a distorted octahedral environment.
The effective magnetic moment calculated from the obtained Curie constant takes μ eff = 1.9μ B /Cu 2+ . This value is in excellent agreement with the theoretical estimation according to μ ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi where the average effective g-factor was directly measured by electron spin resonance (ESR). As depicted in the left inset in Fig. 10(a), the ESR spectrum shows relatively broad anisotropic absorption line, which is typical of the powder pattern of the Cu 2+ compounds. The estimated values of the principal components of g-tensor are g ∥ = 2.20 and g ⊥ = 2.19, providing an average g av = 2.196.
The negative sign of the Weiss temperature indicated the predominance of antiferromagnetic exchange interactions in the system. This is also confirmed by the temperature-depen- dence of the scaled product (χ − χ 0 )(T − Θ) shown in the right inset of Fig. 10(a). At high temperatures, this product follows a horizontal line, obeying the Curie-Weiss law, whereas below approximately 175 K, it deviates downward, indicating the predominance of the antiferromagnetic short-range correlations. Despite the wide range of correlations, the system does not reach ordered state, which probably relates to frustration and competing exchange interactions. The presence of exchange interactions of both the ferro-and antiferromagnetic nature follows from the field dependence of the magnetization (Fig. 10(b)), which demonstrates characteristic S-type behaviour at 2 K, which is similar to those for the isostructural compound with Ni 2+ . However, at higher temperature, magnetization shows ordinary linear trend, typical of antiferro-or paramagnets. This behaviour is completely different from the trirutile-type CuSb 2 O 6 , which demonstrated clear spin-flop transition with H SF ∼ 1.3 T at 5 K below the Néel temperature T N = 8.5 K. 6,9 Interestingly, the relatively high Néel temperature for trirutile-type CuSb 2 O 6 has not been confirmed from theoretical calculations due to the tiny balance in the energy levels of d x 2 −y 2 and d 3z 2 −r 2 electronic shells of Cu 2+ ions in the octahedral oxygen environments and hence is related to large intrinsic magnetic anisotropy. 11 The value predicted from this theoretical analysis is less than 1.5 K. 11 Consequently, the structural changes for the new polymorph lead to more regular octahedral parameters, reducing the magnetic anisotropy for rosiaite-type compound, which in turn results in an essential decrease in the ordering temperature.
The parallel and independent low-temperature neutron diffraction study of CuSb 2 O 6 confirmed the absence of longrange magnetic order down to 1.5 K. Fig. 11 only shows the same reflections that were observed by XRD at room temperature although their relative intensities are entirely different. In particular, the first pair of reflections 001 + 100 is practically invisible. Good agreement between the experimental and calculated profiles confirmed the rosiaite-type structure; however, full-scale structure refinement was not attempted due to the scarce experimental data.
The diffuse maximum between 33 and 39°, as shown in Fig. 11, coincides, in terms of the interplanar spacings d, with that observed in the XRD pattern of Fig. 1b between 21 and 24°. Therefore, it is not related to short-range magnetic ordering.

Conclusions
Herein, five new metastable trigonal layered MSb 2 O 6 phases (M = Co, Ni, Cu, Zn, and Mg) were prepared by the low-temperature ion-exchange reactions, together with previously known CdSb 2 O 6 . FeSb 2 O 6 could not be prepared due to the incompatible oxidation states but is predicted to stabilize at high pressures. CoSb 2 O 6 and NiSb 2 O 6 antiferromagnetically order at 11 and 15 K, respectively. In contrast to their stable trirutile-type polymorphs, they do not show any signs of low-dimensional magnetic behaviour at higher temperatures. CuSb 2 O 6 does not exhibit long-range magnetic order down to 1.5 K but shows antiferromagnetic short-range correlations.