We performed TGA and HT PXRD analysis of samples (Fig. S2, S3, Table S2). TGA showed all samples to be stable up to around 650°C (Fig. S2). Sample EA0 showed the smallest volume thermal expansion but the with largest anisotropy (strong expansion along [001] but very small along [100]. The other samples had a relatively small anisotropy of expansion, EA40 and EA60 larger along [001], EA80 and EA100 along [100]. EA20 was later shown to be a mixture of two phases and therefore irrelevant for comparison (Fig. S1 and Table S1).
We continued the heat treatment of the same samples for one month further at 300°C and at 550°C for additional 200 hours to obtain the better crystallinity. We made the final Rietveld analyses on samples after this treatment and the main Rietveld results, we present in Table 1 and Fig. 1. The structure details of Rietveld refinements (CIF files) are in supplementary information (Table S1) as well as the lists of bond lengths and angles (Table S3).
All samples contained major phases with crystal structures of the same general type. As an example, the crystal structure of EA60 is shown in Fig. 2. The crystal structures of all synthesized EA phases are built upon the same structural principles as a material group called NASICON (NA Super Ionic CONductor) (Boilot et al. 1988), more specifically they have the same architecture as the pure phosphorus end-member of this group, NaZr2(PO4)3. This structure type contains columns of Na coordinations (in our case also K) sandwiched between two Mg or Al coordinations. The columns are connected through tetrahedral groups (SiO4 and PO4 in the original NASICON and SO4 in EA phases).
Among EA phases, NaMgAl(SO4)3 exhibits R-3c symmetry, which implies it only has one type of alkali-atom site and a disordered distribution of Mg and Al among the octahedral coordinated sites which “sandwich” it. EA phases with K replacing Na partially (more than 35 mol%) or completely have R-3 symmetry, which implies two symmetrically independent alkali-atom sites (AI at 3a; 0,0,0 and AII at 3b: 1/3,2/3,1/6) and two symmetrically independent octahedral sites (MI and MII both at 6c: 2/3,1/3, z). The refinement of the NaMgAl(SO4)3 was also attempted in the lower symmetry group, but R factors were very similar and, even more important, the atomic coordinates were practically the same, suggesting that all octahedral coordinations are of the same size due of the high disorder of Mg and Al. The alkali sites have a coordination number (CN) of six and form of a much-distorted octahedron that can be more accurately described as a trigonal antiprism. It was possible to refine directly the occupancies of AI and AII by Na and K due to a large difference in atomic numbers. The results can be seen in Fig. 3. K exhibits a large preference for AII which is also supported by the significantly higher cation-oxygen bond lengths in AII compared to AII observed for all EA phases (see Table S3). For compositions with K > 50 mol%, this site is completely occupied by K. This is confirmed by NMR (see below).
The occupancy of Mg and Al at the M sites cannot be directly refined due to their similar atomic numbers but can be inferred from bond lengths. The results can be seen in Fig. 4. The bond lengths suggest a large preference of Mg for the MI site (sandwiching AI) and Al for the MII site (sandwiching AII) in all R-3 phases. In Figs. 3 and 4 the data for R-3 phase in EA20 are not included, because the diffraction overlaps of the two EA phases present in this sample do not allow the same accuracy of results as in the other samples.
To gain further insight into the local structure around Al and Na, 27Al and 23Na MAS NMR spectra were recorded for all samples. The 23Na MAS NMR spectra (Fig. 5) show a gradual evolution. EA0 (one alkali site) is very complex and contains several 23Na NMR resonances, which reflects a high degree of local disorder, which might be due to displacement of some of the Na atoms from the 6b position to the 18e position, which is occupied in NASICON phases with “excess” Na coupled with Si for P substitution. In contrast, the 23Na MAS NMR spectrum e.g., of EA80 can be simulated with a single 23Na resonance confirming the preference of Na for one of the two crystallographic positions (AI). The 23Na MAS NMR spectrum of EA40 clearly contains two distinct 23Na resonances with well-defined second-order quadrupolar line shapes. From the intensity, there is a strong preference for Na to be located on one of these sites, which supports the observations based on PXRD (Fig. 3).
The 27Al NMR spectra show a similar trend. EA0 shows a single asymmetric line shape, which is characteristic for a 27Al site in a disordered local environment with a maximum intensity at δ(27Al) = -27 ppm, whereas the major resonance is at δ(27Al) = -21 to -22 ppm for the members of the solid solution series from EA40 to EA100. Thus, for Al chemical shift significantly changes from EA0 through EA20 (mixture of phases) to EA40-EA100. This is in accordance with a change from a disordered to a largely ordered distribution in the crystal structure. Again, 27Al NMR confirms the miscibility gap. Moreover, a minor resonance is seen just below δ(27Al) ≈ -20 ppm. This is either from a small impurity phase (KAl(SO4)2) or from a small amount of Al located on Mg sites.