Interstitial Oxide Ion Conductivity in the Langasite Structure: Carrier Trapping by Formation of (Ga,Ge)2O8 Units in La3Ga5-x Ge1+x O14+x/2 (0 < x ≤ 1.5).

Framework oxides with the capacity to host mobile interstitial oxide anions are of interest as electrolytes in intermediate temperature solid oxide fuel cells (SOFCs). High performance materials of this type are currently limited to the anisotropic oxyapatite and melilite structure types. The langasite structure is based on a corner-shared tetrahedral network similar to that in melilite but is three-dimensionally connected by additional octahedral sites that bridge the layers by corner sharing. Using low-temperature synthesis, we introduce interstitial oxide charge carriers into the La3Ga5-x Ge1+x O14+x/2 langasites, attaining a higher defect content than reported in the lower dimensional oxyapatite and melilite systems in La3Ga3.5Ge2.5O14.75 (x = 1.5). Neutron diffraction and multinuclear solid state 17O and 71Ga NMR, supported by DFT calculations, show that the excess oxygen is accommodated by the formation of a (Ge,Ga)2O8 structural unit, formed from a pair of edge-sharing five-coordinated Ga/Ge square-based pyramidal sites bridged by the interstitial oxide and a strongly displaced framework oxide. This leads to more substantial local deformations of the structure than observed in the interstitial-doped melilite, enabled by the octahedral site whose primary coordination environment is little changed by formation of the pair of square-based pyramids from the originally tetrahedral sites. AC impedance spectroscopy on spark plasma sintered pellets showed that, despite its higher interstitial oxide content, the ionic conductivity of the La3Ga5-x Ge1+x O14+x/2 langasite family is lower than that of the corresponding melilites La1+y Sr1-y Ga3O7+y/2. The cooperative structural relaxation that forms the interstitial-based (Ga,Ge)2O8 units stabilizes higher defect concentrations than the single-site GaO5 trigonal bipyramids found in melilite but effectively traps the charge carriers. This highlights the importance of controlling local structural relaxation in the design of new framework electrolytes and suggests that the propensity of a framework to form extended units around defects will influence its ability to generate high mobility interstitial carriers.

Introduction

Solid oxide fuel cells (SOFCs) are efficient all-solid-state energy conversion devices that facilitate the use of clean fuels such as hydrogen in place of current fossil-fuel based power generation. The need for sufficient ionic diffusion throughout the device means that current commercial SOFCs operate at elevated temperatures of up to 1000 °C, which creates problems with cell construction and durability. Further expansion of this technology is dependent on reducing their operating temperatures to an intermediate temperature range between 600 and 800 °C, motivating the search for new electrolyte materials that conduct oxide ions efficiently under these conditions.[1−3] The most widely used solid oxide electrolytes are materials based on the fluorite or perovskite structures (e.g., Zr1–YO2– or YSZ,[4] Ce1–GdO2−δ or GDC,[5] and La1–SrGaMg1–O3−δ or LSGM[6]) where cations formally adopt 12-, 8-, or 6-fold coordination and ionic diffusion is mediated by vacancies on the oxide sublattice. A considerable amount of research into intermediate temperature oxide electrolytes has focused on framework materials, where metals with low coordination numbers form networks of tetrahedral MO4 or (bi)pyramidal MO5 units.[7] Cations with low coordination numbers at the centers of the oxide polyhedra defining these networks can offer a different conduction mechanism to the fluorites and perovskites, by expanding their coordination sphere to accommodate interstitial oxide ions as the charge carrying species. It follows that candidate framework oxide electrolytes must have a high concentration of flexible low-coordinate cation environments and a high capacity for anion nonstoichiometry. These conditions are satisfied in the rare-earth oxyapatites such as La10–(MO4)6O3– (where M = Ge[8] or Si[9]) and melilites such as La1+MGa3O7+ (where M = Sr or Ca),[10,11] which are two leading classes of framework oxide electrolytes. These materials contain isolated and two-dimensional interconnected tetrahedra, respectively, and their conductivities are higher than the most commonly used electrolyte materials at ∼600 °C.[12] In apatite, the conduction mechanism relies on exchange between multiple interstitial sites and MO4 tetrahedra to allow oxide ion diffusion in three dimensions,[13] while in melilite the conductivity is thought to be constrained in two dimensions with a cooperative exchange of lattice and interstitial oxygens, facilitated by the formation of 5-coordinate Ga locally.[14,15] Mobile oxide interstitials are found in other structures with isolated MO4 tetrahedra such as scheelite-type La0.2Pb0.8WO4.1[16] and CeNbO4+,[17] while in the cuspidine structure of La4(Ga2–GeO7+)O2 it is the flexibility of Ga to adopt 4- or 5- coordination that permits the incorporation of interstitial oxygen in 1D chains.[18]

Previously, some of us identified local structural motifs in the melilite crystal structure that facilitate the incorporation of interstitial oxide anions.[10] In particular, the melilite framework contains a tetrahedral site which is only connected at three points by bridging oxides to other tetrahedra in the framework and has a nonbridging oxide anion. The gallium cation at the center of these tetrahedra is able to increase its coordination number, allowing the formation of a five-coordinate gallium site incorporating the interstitial oxide anion. By analogy, langasites such as La3Ga5GeO14 were identified as a potential interstitial host due to the presence of a closely related local structural motif.

As with the melilite crystal structure, the langasite crystal structure can be described in terms of a framework of oxide ions and smaller cations with larger A-site cations in the void space of the framework (Figure ). Langasites adopt noncentrosymmetric trigonal structures (space group P321) that have driven interest in their optical, electromechanical and multiferroic properties.[19−21] Similar to the melilite framework (general formula A2CD2O7), the langasite framework (general formula A3BC3D2O14) contains a layer of four-connected tetrahedra (CO4), where all the oxide anions bridge to neighboring units, and three-connected tetrahedra (DO4) which contain nonbridging oxide anions. In contrast, in the langasite framework only 2/5 of the tetrahedral units are three-connected DO4 tetrahedra, compared to 2/3 in the melilite framework, giving rise to a different ordering of the CO4 and DO4 units within the layer, such that each DO4 unit connects only to three CO4 units. Unlike the melilite framework, the langasite framework is three-dimensional, with an additional octahedral BO6 unit bridging the CO4 tetrahedral units in different layers. The CO4 tetrahedra share two corners with BO6 units and two corners with DO4 units. The langasite framework then contains six membered rings bounded by the edges of one BO6 octahedron, three CO4 tetrahedra, and two DO4 tetrahedra, rather than the five membered rings present in the melilite framework. Eight coordinate A-site cations occupy sites in between the tetrahedral layers, in the channels formed by the six-membered rings. This gives rise to the general langasite formula A3BC3D2O14, where A is a rare- or alkaline-earth cation, B is a transition metal or main group cation, and C and D are main group cations. The incorporation of the BO6 octahedra, which connects adjacent layers in langasite, constrains the interlayer separation (A–A and B–B distances, defined by the lattice parameter c = 5.109 Å) to be shorter than those found in melilite (e.g., 5.331 and 5.275 Å in LaSrGa3O7 and LaCaGa3O7, respectively).[22]

Figure 1

Comparison of the (a, b) langasite and (c, d) melilite structures, featuring three-connected D (yellow) and four-connected C (purple) tetrahedra, B octahedra (green), and La3+/Sr2+A cations (blue). (a) The langasite structure A3BC3D2O14 viewed along the stacking axis and (b) perpendicular to the stacking axis, showing a two-dimensional arrangement of three-connected and four-connected vertex-linked tetrahedra, which are connected along the stacking axis by vertex-linked (Ga/Ge)O6 octahedra (green) to produce a three-dimensional network with six-membered ring channels. (c) The layered melilite structure of LaSrGa3O7 (A2CD2O7) viewed along the stacking axis and (d) perpendicular to the stacking axis, showing the two-dimensional connectivity of the framework with five-membered ring channels.

Here we focus on the parent compound La3Ga5GeO14 where the octahedral B and tetrahedral C sites are reported to be occupied by Ga3+, with a 50/50 mixture of Ga3+ and Ge4+ occupying the smaller tetrahedral D sites.[23] The Ga3+ and Ge4+ cations in the DO4 tetrahedra should be able to accommodate coordination numbers above 4 and are therefore expected to play the role of the Ga3+ tetrahedra in the melilite. We present the structural and physical characterization of the La3Ga5–Ge1+O14+ (0 ≤ x ≤ 1.5) solid solution, using neutron diffraction, solid state 71Ga and 17O nuclear magnetic resonance (NMR), density functional theory (DFT) structural calculations, and AC impedance spectroscopy. This doped langasite family shows a high capacity for the accommodation of interstitial oxygen when synthesized under sufficiently mild conditions. We show that, despite the structural similarities between the two parent materials, the interstitial oxide anion in this langasite is associated with a local structural motif different from that observed in the melilite, which binds the interstitial anion strongly and prevents high ionic conductivity even at relatively high interstitial anion contents.

Results

Synthesis and Structure Solution of La3Ga5–Ge1+O14+

Initially, the syntheses of La3Ga5–Ge1+O14 were carried out by a conventional ceramic method at a temperature of 1300 °C. This produced single-phase materials with the langasite structure for x ≤ 0.30, while samples with x > 0.30 contained additional Bragg peaks originating from an unknown secondary phase (confirmed by TEM-EDX, see Figure S1). The limit of dopant incorporation into La3Ga5–Ge1+O14+ was extended from x = 0.30 to x = 1.5 by synthesis at lower temperatures using a sol–gel method. The compositions 0.4 ≤ x ≤ 1.5 were found to have characteristic decomposition temperatures, which decrease asymptotically with x from 1300 °C (x = 0.4) to 850 °C (x = 1.5), above which they undergo a phase separation (see Figures S1 and S2). Consequently, the synthesis temperature for each composition in this range was optimized to 50 °C below the decomposition temperature (Figure S2). The refined cell parameters for La3Ga5–Ge1+O14+ vary smoothly over the 0 < x ≤ 1.5 range showing near-linear trends that are consistent with the formation of a solid solution (Figure a). Upon doping of La3Ga5–Ge1+O14+ up to x = 1.5, there is an expansion of the a and b lattice parameters by approximately 0.08 Å accompanied by a 0.12 Å contraction along the c axis. A similar trend in c/a is observed on the inclusion of interstitial oxide in melilite.[41] As a result, there is a decrease in the volume of the cell by ∼0.5 Å3, accompanied by a large decrease in the c/a ratio (Figure b) which supports the substitution of Ga3+ by smaller Ge4+ cations (4-coordinate ionic radii of 0.47 and 0.39 Å, respectively).[24]

Figure 2

Lattice parameter variation in the La3Ga5–Ge1+O14+ series. (a) Normalized a (= b) parameter (black points), c parameter (red points), and unit cell volume (blue points) and (b) variation in the c/a ratio. (c) Laboratory PXRD patterns across the full range of compositions accessed (0 ≤ x ≤ 1.5), showing that the langasite structure is retained as a single phase, with a clear trend in Bragg reflection positions (e.g., 122 and 131 highlighted in pink). (d) The 122 and 131 Bragg reflections, which cross over as the series is traversed, reflecting the large change in c/a.

High-resolution neutron powder diffraction data were collected from La3Ga3.5Ge2.5O14.75 (x = 1.5, the most oxygen-rich member of the series) at 9 K in order to localize scattering from interstitial oxygen as effectively as possible. A nuclear scattering density map was generated from this data set by the maximum entropy method[25] (Figure ), revealing two symmetrically inequivalent peaks (12 peaks per unit cell when all symmetry equivalents are included) that do not correspond to atomic positions in the parent langasite structure. This contrasts with doped melilites, where a single position close to the center of the channels (associated with a displacement toward a single Ga atom) was identified that corresponded to the presence of interstitial oxide.[10] In langasites, the sites of unassigned nuclear scattering density attributed to interstitial oxides are located in the vicinity of the 6-membered rings, displaced away from the center of the channel, within the layers of interconnected tetrahedra. Note the additional scattering density around other atomic sites in the map, especially for O2 and O3 bridging the GaC and GaD and the GaB and GaC cation sites, respectively, which is suggestive of local structural disorder expected from the structural relaxation of neighboring atoms in the framework to accommodate the interstitial.

Figure 3

Nuclear scattering density map of La3Ga3.5Ge2.5O14.75 derived by the maximum entropy method from neutron diffraction data collected at 9 K, plotted as an isosurface map (yellow surface), with the structure of La3Ga5GeO14 overlaid (red = oxide, green = Ga/Ge, blue = La). The atom labels shown are those corresponding to the parent structure, where GaB, GaC, and GaD correspond to B, C, and D cation sites, respectively. Two new symmetrically inequivalent peaks that are displaced from the atom positions of the parent framework are circled. The relatively large displacements of the O2, O3, and GaC sites are also visible, indicating that these are the principal atoms involved in the relaxation of the framework.

Given the extensive relaxation of the interstitial-containing langasite indicated by the maximum entropy map, DFT calculations were performed to identify likely local structural motifs in doped langasites. First, the scattering density associated with the oxide interstitial was added into the langasite model and the subsequent relaxation of the framework was identified by DFT. Three 2 × 2 × 2 supercells were constructed, each containing four interstitial oxide ions, giving a composition of La3Ga4Ge2O14.5 (Figure S3). Note that each 2 × 2 × 2 supercell contains 96 sites (12 per subcell) where interstitial oxide can potentially be accommodated, as indicated by the unassigned peaks in the nuclear scattering density map. Ga and Ge were fully ordered, with Ga on GaB and GaC sites and Ge on GeD sites, in accordance with Ga/Ge ordering proposed in the previously published structure of La3Ga5GeO14.[23] The additional oxide ions were placed at four of the 96 identified locations; each 2 × 2 × 2 supercell was assigned a different subset of four occupied sites, and cell parameters and atomic positions were optimized. The lowest energy 2 × 2 × 2 DFT calculated supercell is shown in Figures a and S4, where the extra oxide ions are found in new (Ga/Ge)2O8 units, in which the GaC and GaD sites are bridged by two oxide ions, one interstitial (O4), and one lattice oxygen (O2b) that has been displaced by over 1 Å from its original position in O2, rather than the single bridging oxide ion present in the undoped parent La3Ga5GeO14 structure. This forms a pair of distorted edge-sharing square-based pyramids, whose apexes O1 and O3b (see Figures e–g) point in opposite directions. Due to the chosen ordering of Ga and Ge atoms, each (Ga/Ge)2O8 unit in the original DFT models contained one Ga and one Ge atom. Two alternative models were constructed by taking the lowest energy model (Figure S3) and reordering the Ga and Ge atoms to make fully Ga2O8 and fully Ge2O8 units (Figure S4). The model with Ge2O8 units is 0.05 eV/FU more stable than the original model with GaGeO8 units, and the model with Ga2O8 units is 0.05 eV/FU less stable, suggesting that smaller, higher charged Ge4+ ions are preferred in these sites close to the additional O2– ions. NMR parameters were calculated for the undoped parent model and the three interstitial doped models (Figure S4) using the GIPAW method[25_1,26] and CASTEP code[27] and used to validate the experimental NMR spectral assignments.

Figure 4

DFT and Rietveld structures of La3Ga3.5Ge2.5O14.75, where blue = La, red = O, green = Ga, and purple = Ge. (a) DFT relaxed 2 × 2 × 2 superstructure of La3Ga4Ge2O14.5, shown for clarity as a single two-dimensional sheet (i.e., a 2 × 2 × 1 fragment), which includes three (Ga,Ge)2O8 units incorporated at random within the corner-sharing network. The unit cell boundary of the crystallographic unit cell is overlaid in black. (b) Average structure of La3Ga3.5Ge2.5O14.75 from Rietveld refinement at 9 K. (c) One possible local configuration of an interstitial-containing unit cell of La3Ga3.5Ge2.5O14.75, constructed from the Rietveld model. (d) The configuration of a noninterstitial-containing unit cell, constructed from the Rietveld model. (e, f) Two representative local geometries of GaGeO8 units extracted from the DFT-relaxed 2 × 2 × 2 superstructure. (g) Local geometry of the (Ga,Ge)O8 units extracted from the Rietveld model. (h) Local geometry of the corner-sharing GaD-O4 and GaC-O4 tetrahedra in the parent phase obtained by Rietveld refinement. The corresponding bond angles and distances are listed in Tables S2 and S3.

An initial Rietveld model for La3Ga3.5Ge2.5O14.75 was generated by collapsing the lowest energy DFT supercell (Figure S5) into an averaged 1 × 1 × 1 cell in space group P321 (Figure a), with Ga and Ge modeled as being fully disordered over the three cation sites GaB, GaC, and GaD. Note that, in the averaged cell, the framework atoms with additional scattering density identified by the maximum entropy map (GaC, O3, and O2, see Figure ) are split between their original framework positions and a new set of positions that are associated with the local defects (GaCb, O2b, and O3b). The occupancies of GaCb, O2b, and O3b are defined by the defect concentration. This model was refined simultaneously against three TOF NPD databanks, collected at 9 K. Peaks arising from the vanadium sample holder were excluded from the refinement. Lattice parameters and atomic coordinates were refined freely. Isotropic displacement parameters were refined freely for framework atoms, while O2b and O4 which form the shared edge in the (Ga/Ge)2O8 unit with low occupancies were constrained to have the same isotropic displacement parameters. The occupancies of all atoms, including new sites arising from the interstitial O4 and displaced framework O2b oxygens, were fixed to values defined by composition. In the final stage of the refinement, the Ga/Ge ratio at each site was allowed to refine to take advantage of the moderate difference in the neutron coherent scattering length (7.3 and 8.2 fm, respectively), with a constraint on the global composition: this confirmed that these cations are partially ordered, with a preference for tetrahedral coordination of Ge (which occupies 53(3)% of tetrahedral C and D sites, in comparison to 32(4)% of the octahedral B sites). This refinement produced a good fit to all of the features of the diffraction patterns (Rwp = 2.08%, Rexp = 0.73%, fits shown in Figure ). For comparison, the interstitial sites O2b and O4 and split Ga/Ge tetrahedral GaC and O3 sites (namely, Ga3b and O3b in Figure ) were excluded from the model (i.e., so that it corresponds to the parent structure with no interstitials), and the refinement was repeated. This produced a much poorer fit to the data (Rwp = 3.27%). Finally, the defect-containing model derived at 9 K was refined against a data set collected at room temperature and found to provide a similarly good fit (Figure S6).

Figure 5

Rietveld refinement of La3Ga3.5Ge2.5O14.75 against high resolution NPD data collected at 9 K. (a) Fit to 2θ = 169° bank, with small d-spacing range inset, (b) fit to 2θ = 90° bank, and (c) fit to 2θ = 30° bank. Black crosses = observed data (yobs), red line = calculated intensities (ycalc), gray line = yobs – ycalc, and black markers = allowed hkl reflections.

The refined model of La3Ga3.5Ge2.5O14.75 in Figure c,d is based upon the same framework exhibited by the undoped parent structure La3Ga5GeO14. At this composition, 75% of the unit cells must contain an interstitial oxygen ion. This has a major impact on the coordination environments found in the corner-linked tetrahedral network, because each interstitial oxygen is accommodated by bridging two adjacent network cations in the GaC and GaD positions. To achieve this, the framework oxide O2 is displaced by 1.21(1) Å (becoming O2b) but retains its bridging contacts with the GaC and GaD sites, with a corresponding displacement of the GaC cation (by 0.28(1) Å becoming GaCb) and of the framework oxide O3 (by 0.47(1) Å, becoming O3b) that connects GaC to the (GaB)O6 octahedra (Figure ). This increases the coordination number of the adjacent cations in GaCb and GaD to five, producing a pair of edge-sharing square-based pyramids to form a new (Ga,Ge)2O8 structural unit analogous to that found by DFT (Figure e–g). In contrast to the two tetrahedral sites, the coordination number at the neighboring BO6 octahedral site does not change, although its geometry is modified by O3 displacement to O3b. This increases the O3(b)–GaB distance from 1.928(2) Å to 2.02(1) Å, reducing the axial O3(b)–GaB–O3 angle from 173.0(1)° to 159.7(3)°, and distorts all four equatorial O3(b)–GaB–O3 angles further away from regularity (see Table S1). The (Ga,Ge)2O8 units connect to the extended two-dimensional network by corner sharing with adjacent GaC and GaD tetrahedra in a disordered fashion (Figure a,b), so that the long-range connectivity of the network is retained. This mode of interstitial oxide coordination, which directly changes the coordination environments of two adjacent originally tetrahedral cations and distorts the next-nearest neighbor octahedron, contrasts with the structural response of the melilite structure where excess oxygen joins the coordination sphere of a single Ga3+ ion, which causes a less extensive structural perturbation with maximum atomic shifts of 0.44 Å.[10,11]

The local geometry of the (Ga,Ge)2O8 units in La3Ga3.5Ge2.5O14.75 consists of two square-based pyramids that share a basal edge as shown in Figure f, where the apical oxygens O1 and O3b are pointing in opposite directions. Unlike the Ge2O8 units found in La2Ge3O9,[28] the pyramids are distorted from ideal geometry and are not related by inversion symmetry. Formation of these units in La3Ga3.5Ge2.5O14.75 is coupled with a splitting of the GaC and O3 sites, corresponding to features at these positions in the nuclear scattering density map (Figure ). This maintains physically reasonable coordination environments that are consistent with the DFT model, with all (Ga,Ge)–O distances in the range 1.767(3)–2.098(15) Å (compared to 1.789–2.066 Å by DFT) as shown in Figure e–g and Table S2. The refined O–M–O bond angles of the (Ga,Ge)2O8 units are similarly consistent with the ranges predicted by DFT (Table S3), despite the pronounced distortion away from ideal pyramidal geometry; the angles between the cations and the bridging oxygens (see Figure ) are among the most acute at 76.6(3)° and 72.4(2)°, respectively, which is characteristic of electrostatic repulsion between highly charged cations in edge-sharing tetrahedra.[29] While Ga and Ge are fully ordered in the DFT model, they are only partially ordered in the average (refined) structure, with Ga showing a preference for the GaC site (63(2)% Ga) over the GaD site (47(4)% Ga). This is consistent with the different sizes of the two sites in the (Ga,Ge)2O8 unit and with the spread of DFT-calculated formation enthalpies presented above (0.10 eV/FU between Ga2O8 and Ge2O8 configurations, which lies in the range of kBT at the synthesis temperature). Bond valence sums suggest that the average GaD site is well suited to occupancy by Ge4+ (BVS = 4.08(3), where the calculated error is derived from refined bond lengths) but requires Ga3+ to be overbonded (BVS = 3.88(3)), while the larger GaCb site favors Ga3+ occupancy (BVS = 2.39(8)) with Ge4+ under-bonded (BVS = 2.57(8)).

The formation of (Ga,Ge)2O8 units in La3Ga5–Ge1+O14+ (x ≤ 1.5) is consistent with the observation of a Ge2O8 motif in the (nonlangasite) structure of La3GaGe5O16[29] (x = 4), which can be considered as an oxygen-saturated end-member of the La3Ga5–Ge1+O14+ series (Figure ). Here, the additional 2 oxides per formula unit are accommodated by fundamental changes to the langasite framework, with fully ordered Ge2O8 units interrupting the 2D connectivity of C and D tetrahedra, and driving a rearrangement of the La3+ and GaO6 columns away from their trigonal conformation in langasite to produce a mixture of an expanded 10-membered channel with two La3+ sites and langasite-like 6-membered channels. Importantly, the dimer unit is not directly bound to the BO6 octahedra, unlike in the doped langasite. Because of the different framework connectivities, a continuous langasite-type solid solution is only obtained in the region x ≤ 1.5. In La3GaGe5O16, the Ge2O8 units are more regular than the (Ga,Ge)2O8 units formed in the langasite structure (Figure S7) because the 10-membered channels are large enough to accommodate them within a highly ordered framework. However, no further oxygen can be accommodated by the mechanism of M2O8 formation, because there are no adjacent corner-sharing tetrahedra.

Figure 6

Contrasting structures of La3GaGe5O16,[29] which istriclinic with ordered 6- and 10- membered rings, and La3Ga4Ge2O14.5 which is trigonal with only 6-membered rings and extensive disorder associated with relaxation around the interstitial oxide defect. (a, b) La3GaGe5O16[29] structure, viewed along the a and b axes, respectively, where octahedrally coordinated ions (green) are linked by three tetrahedra (three-connected in yellow and four-connected in purple) to form columns along a, which are connected by ordered Ge2O8 (gray) units separating the langasite-like 6-membered and expanded 10-membered channels. (c) DFT-relaxed structure of a 2 × 2 × 2 supercell of La3Ga4Ge2O14.5, viewed along the stacking axis, showing (Ga,Ge)2O8 units (gray) incorporated into the trigonal langasite framework in a disordered way. La3+ ions are shown in blue.

MAS NMR Spectroscopy

17O (spin I = 5/2) NMR spectra were obtained to investigate the presence of interstitial oxygen sites in the La3Ga5–Ge1+O14+ (with 0 ≤ x ≤ 1.5) langasite structures (Figures , 8, S8, and S9). The detection of 17O spins is challenging due to the very low natural abundance of 17O (0.037%). Here the 17O enrichment strategy relies on post-synthesis oxygen exchange[30] at elevated temperature (750 °C over 24 h) between 17O-enriched O2 gas and the as-prepared langasite (Figure S10). The 17O MAS NMR spectra of 17O enriched La3Ga5GeO14 obtained at magnetic fields of 9.4 and 20 T are shown in Figure and display a series of resonances in the 50–220 ppm region. The broadening of the NMR lines arises from the presence of second-order quadrupolar interactions, which are not averaged to zero by MAS but which can be removed by two-dimensional (2D) triple-quantum magic angle spinning (3QMAS) experiments (Figures b, S11a at 20 T, and S12a at 9.4 T).[31−33] These corresponding 2D spectra could be simultaneously fitted with at least 4 signals at isotropic chemical shifts δiso,cs of 122, 135, 175, and 199 ppm (full NMR parameters are given in Table ) and reproduce accurately the 1D 17O MAS NMR spectra of La3Ga5GeO14 (Figure ).

Figure 7

17O MAS NMR spectra of 17O enriched La3Ga5GeO14 obtained at 9.4 and 20 T (full line). The one-dimensional simulated spectra (dashed black lines) and individual NMR sites (full black lines) obtained from line shape analysis of the two-dimensional sheared 17O triple-quantum (3Q) MAS NMR are given below the experimental data at each field. Simulation of the one-dimensional spectra (dashed red lines) and individual NMR sites (red full lines) based on the GIPAW NMR calculations are also given. The sharp lines around 70 ppm (marked with #) correspond to absorbed H2O.

Figure 8

(a) 17O MAS NMR spectra of 17O enriched La3Ga5–Ge1+O14+ (with 0 ≤ x ≤ 1.5) as a function of Ge doping level obtained at 20 T and under MAS rates of 22 kHz. Simulation of the one-dimensional spectra (dashed red lines) and individual NMR sites (red full lines) based on the GIPAW NMR calculations are given below the experimental data. The peak positions for the O2b and O4 sites obtained from the calculations are highlighted in a blue box. The sharp lines around 70 ppm (marked with #) correspond to absorbed H2O. (b) Two-dimensional sheared 17O triple-quantum (3Q) MAS NMR spectrum of 17O enriched La3Ga4Ge2O14.5. Left: isotropic projection of the 2D 3Q MAS spectrum. Top: 17O MAS NMR spectrum. Asterisk (*) denotes spinning side bands.

Table 1

Experimental and Calculated 17O, 71Ga, and 73Ge NMR Parameters for La3Ga5GeO14 and La3Ga4Ge2O14.5a

site	environment	coordination	δiso,cs, ppmb	|CQ|, MHz	ηQ
O	O1	GaIV–O	175(10)	4.5(0.4)	0.5(0.2)
			180.7c	4.3	0.0
			189.0 ± 22.2d	4.1 ± 0.2	0.1
		GeIV–O	122(20)	3.9(0.3)	0.6(0.2)
			153.9c	5.7	0.0
			164.5 ± 17.2d	5.7 ± 0.5	0.1 ± 0.1
	O2	GaIV–O–MIV (M = Ga, Ge)	135(10)	3.5(0.4)	0.1(0.2)
		GaIV–O–GaIV	138.8c	3.0	0.9
			143.9 ± 13.4d	3.4 ± 0.4	0.8 ± 0.2
		GaIV–O–GeIV	140.0c	5.0	0.4
			145.6 ± 22.5d	5.0 ± 0.4	0.6 ± 0.2
	O3/O3b	GaIV–O–GaVI	199(10)	3.8(0.2)	0.4(0.2)
			204.7 ± 0.3c	4.0 ± 0.3	0.7 ± 0.1
			195.3 ± 15.5d	4.2 ± 0.6	0.7 ± 0.2
		GaIV–O–GaV	135.6 ± 15.4d	2.6 ± 0.3	0.4 ± 0.2
		GaIV–O–GeV	189.0 ± 15.3d	6.3 ± 0.3	0.4 ± 0.1
		GeVI–O–GaV	212.4 ± 19.5d	5.1 ± 0.3	0.7 ± 0.1
		GeVI–O-GaIV	209.6 ± 12.8d	4.9 ± 0.6	0.8 ± 0.2
	O2b and O4 in La3Ga4Ge2O14.5	GeV–O–GaV	270(10)	3.0(0.2)	0.4(0.2)
			271.6 ± 12.5d	2.7 ± 0.3	0.5 ± 0.1
Ga	GaC	GaO4	-e	-e	-e
			140.7c	25.3	0.8
			135.6 ± 18.8d	26.5 ± 5.4	0.5 ± 0.2
		GaO5	-e	-e	-e
			109.5 ± 10.2d	23.3 ± 5.7	0.5 ± 0.2
	GaD	GaO4	250(10)	16(1)	-e
			276.5c	13.9	0.0
			263.9 ± 11.1d	13.2 ± 4.2	0.2 ± 0.1
		GaO5	140(10)	13(2)	-e
			177.3 ± 4.5d	10.0 ± 2.6	0.6 ± 0.4
	GaB	GaO6	15(10)	6(1)	-e
			32.2 ± 2.0c	5.0 ± 1.7	0.0
			44.5 ± 7.9d	8.1 ± 4.8	0.5 ± 0.3
Gef	GeD	GeO6	210.9 ± 0.8c	15.5	0.0
			164.8 ± 3.2d	16.6 ± 5.7	0.5 ± 0.2
		GeO5	55.2 ± 12.5d	18.2 ± 3.6	0.6 ± 0.2
	GaB	GaO6	–101.5 ± 3.2d	16.6 ± 5.7	0.6 ± 0.1

Experimental and calculated values are given in bold and plain text, respectively. 17O and 71Ga experimental values were obtained from the 2D 3QMAS (at 9.4 and 20 T) and 1D MAS spectra at 20 T, respectively (see Methods section for further details). Experimental 73Ge NMR spectra were not obtained due to its low sensitivity. The calculated 17O, 71Ga, and 73Ge isotropic shieldings, σiso, were converted into isotropic chemical shifts δiso,cs following an expression of the form δiso,cs = σref + mσiso with (σref, m) = (223.70 ppm, −0.888) for 17O,[32] (1502.63 ppm, −0.867) for 71Ga,[32] and (1424.24 ppm, −1) for 73Ge.[34] Standard deviations of the calculated data are given in the table (unless the values are less than 0.1). Only the absolute values of CQ are reported. Superscripts IV, V, and VI indicate 4-, 5-, and 6-coordinated cations, respectively.

The isotropic chemical shift δiso,cs is reported, except for 71Ga where only the experimental shift δ is given (obtained from 1D MAS spectra) and is compared to the 71Ga calculated isotropic chemical shift δiso,cs. This is a fair assumption since the 71Ga quadrupolar induced shift (determined as δiso,Q = −PQ2/40νO(71Ga)2 with PQ and νO(71Ga) the quadrupolar product and the 71Ga Larmor frequency), from which δ is shifted from δiso,cs and is found lower than 0.1 ppm at 20 T.

Calculated for the La24Ga40Ge8O112 supercell corresponding to the La3Ga5GeO14 langasite structure.

Calculated for the La24Ga32Ge16O116 supercell corresponding to the La3Ga4Ge2O14.5 langasite structure.

This 71Ga NMR signal is not observed experimentally under the conditions (20 T and MAS frequency of 65 kHz) used here (see Figure S13).

Experimental 73Ge NMR spectra were not collected (see Supporting Information for discussions of the calculated data).

Attribution of these resonances is not trivial. There has recently been a growing interest in combining experimental NMR data collection with computational prediction of NMR parameters. Using this strategy, we have computed the site-averaged NMR parameters (using GIPAW[25_1−27]) for the undoped La3Ga5GeO14 parent model for spectral assignment. This approach is validated by the accurate reproduction of the experimental spectrum by the simulated one (Figure ), reinforcing confidence in both the DFT local model of the structure and the accuracy of the GIPAW calculations, which has been extensively illustrated previously.[35−37] The two main NMR signals of equal intensity observed at isotropic chemical shifts δiso,cs of 135 and 199 ppm are assigned to O2 bridging the two tetrahedral sites GaC and GaD and O3 bridging tetrahedral GaC and octahedral GaB sites, respectively. GIPAW calculations predict very similar shifts δiso,cs for O2 when bridging two Ga (δiso,cs = 138.8 ppm) or two Ge (δiso,cs = 140.0 ppm) and are in excellent agreement with the experimental data (135 ppm). The experimental resonance observed at δiso,cs of 175 ppm is assigned to the apical oxygen O1 bound to GaD based on the calculated δiso,cs of 180.7 ppm. Although quantification of NMR signal intensities of quadrupolar nuclei such as 17O should be interpreted with caution,[38] it is experimentally found that the intensity of this O1 resonance corresponds to approximately 20% of the O2 (or O3) site, in fair agreement with the anticipated O1:O2 (or O1:O3) site ratio of 17% based on the averaged model. The fourth 17O NMR signal is observed at δiso,cs of 122 ppm and attributed to the apical oxygen O1 connected to Ge cation occupying the GaD site. This is a reasonable assignment due to the large experimental error in this shift (20 ppm), which arises from the overlap of its resonance with the one of the 135 ppm of O2 signal and the sizable standard deviation of the shielding reference used to determine the calculated 17O shifts (12.1 ppm).[39]

The 17O MAS NMR spectra of 17O enriched La3Ga5–Ge1+O14+ (with 0 ≤ x ≤ 1.5) obtained at a magnetic field of 20 T are shown in Figure (and at 9.4 T in Figures S8 and S9) and reveal a line shape similar to that of La3Ga5GeO14 at shifts lower than 220 ppm. Very importantly, an additional downfield resonance centered at 270 ppm (Table ), not seen in La3Ga5GeO14, is now also observed in La3Ga5–Ge1+O14+ (with 0.5 ≤ x ≤ 1.5), the intensity of this signal becoming greater with increasing Ge concentration. A similar observation was made on Y-doped La7.5Ca2.5Ge6O25.75 apatite (La8CaYGe6O26.5 and La8Y2Ge2O27) for which the interstitial oxygens around Ge, yielding a GeO5 unit, were observed at a shift of around 280 ppm, higher than the Ge tetrahedral 17O NMR signal observed at 170 ppm.[39]

Site-averaged 17O GIPAW NMR parameters of La3Ga4Ge2O14.5 were calculated for the three 2 × 2 × 1 supercells of the three interstitial doped models discussed above (Figure S4) and used for experimental spectral assignment (Table ) following the validation of this approach on La3Ga5GeO14. First, the calculated 17O shifts support the experimental assignment of the peaks at 122, 135, 175, and 199 ppm to the O1 apical in the square-based pyramids Ge cation on the GaD site, O2 and O1 apical in the square-based pyramids Ga cation on the GaD site, and O3/O3b oxygen sites, respectively. The large standard deviation obtained for these resonances probably reflects the structural disorder in these materials; this is in agreement with the broadening of the 17O signals with increasing Ge content resulting from the chemical shift distribution as evidenced by the elongation of the signals along the +1 diagonal in the 17O TQMAS spectra (Figures b, S11, and S12). More importantly, the downfield resonance experimentally observed at 270 ppm is found to correspond to the O2b and O4 positions for which the isotropic chemical shift δiso,cs is predicted at 271.6 ppm. Estimation of the experimental quadrupolar coupling constant CQ (∼3 MHz) and asymmetry parameter ηQ (∼0.4) are also supported by the calculated quadrupolar coupling constant CQ (2.7 MHz) and asymmetry parameter ηQ (0.5) for this interstitial oxygen. The relative content of interstitial oxygen as determined by integration of the 17O MAS NMR signal of La3Ga3.5Ge2.5O14.75 yields an estimated value of 6 ± 2%, in agreement with the accommodation of 5.4% of extra oxygen.

The introduction of interstitial oxygens in the La3Ga5GeO14 langasite structure must be accompanied by an increase in coordination number of the Ge and/or Ga atoms which can be probed by 73Ge[34,40,41] and 69Ga/71Ga[35,42,43] NMR, respectively. Although 73Ge has a spin of I = 9/2 and is potentially informative, 73Ge NMR spectra have not been obtained experimentally due to its very low sensitivity arising from a combination of low gyromagnetic ratio, large quadrupolar moment (Q = −19.6 × 10–30 m2), low natural abundance (7.76%), and the featureless nature of the NMR spectra of Ge containing disordered materials.[34]69Ga (spin I = 3/2, natural abundance of 60.4%) and 71Ga (spin I = 3/2, natural abundance of 39.6%) are more informative,[42] and 69Ga is usually the NMR nucleus of choice for Ga because of both being more sensitive than 71Ga and yielding sharper NMR lines due to its smaller quadrupolar moment. The 71Ga MAS NMR spectra of La3Ga5–Ge1+O14+ obtained at 20 T and under very fast MAS at 65 kHz are given in Figure and showed a line shape constituted of multiple resonances. The 71Ga spectrum of La3Ga5GeO14 presents one signal at 15 ppm corresponding to the central transition of the Ga site in octahedral geometry as well as a broad signal (with a full width at half-maximum of approximately 45 kHz) centered at around 240 ppm and corresponding to Ga in tetrahedral geometry, with the increase in shift with decreasing coordination number being typical of 71Ga (and other nuclei).[35,39,42−44] Two tetrahedral GaC and GaD sites are present in La3Ga5GeO14, and the site-averaged GIPAW NMR calculations reveal that both sites have large quadrupolar coupling constants CQ with values of 25.3 and 13.9 MHz, respectively (Table ). While the 13.9 MHz value obtained for the GaD site is on the upper limit of CQ values usually obtained for Ga in tetrahedral geometry,[35,39,42] the much larger value of 25.3 MHz (and ηQ = 0.8) obtained for a tetrahedral GaC site is unusual and reflects the calculated distorted Ga geometry of this site with two short Ga–O distances (of 1.838 and 1.845 Å from DFT, and 2 × 1.807(1) Å from ambient temperature neutron refinement) and two elongated Ga–O bonds (of 1.895 and 1.992 Å by DFT, and 2 × 1.882(1) Å by neutron refinement). The predicted calculated MAS line width of the GaC site is approximately 600 kHz at 20 T (Figure S13) and is therefore significantly larger than the 65 kHz MAS frequency used here (Figure ), yielding a broad and complex MAS line shape[45] and preventing its observation even under the very high MAS and ultrahigh magnetic field experimental conditions used here. In that case, NMR spectra under static conditions usually enable the observation of sites with very large CQ values, and the corresponding static 71Ga NMR spectrum of LaGa5GeO14 at 20 T (Figure S13) indeed displays a broad component spanning more than 350 kHz (from 600 to −800 ppm) which is tentatively assigned to the GaC site.

Figure 9

71Ga MAS NMR spectra of La3Ga5–Ge1+O14+ (with 0 ≤ x ≤ 1.5) obtained at 20 T and under MAS rates of 65 kHz. The spectra were recorded with a 2 ms DFS enhancement pulse.[32] The one-dimensional simulated spectrum (dashed lines) and deconvoluted spectra (dotted lines) were obtained from best fit simulations (see Table ).

The 71Ga MAS NMR spectra of La3Ga5–Ge1+O14+ (with 0.5 ≤ x ≤ 1.5) reveal the presence of a third resonance in addition to the two resonances previously observed in La3Ga5GeO14. This additional resonance increases in intensity with Ge content and appears at a shift of around 140 ppm, intermediate between those observed at 15 and 240 ppm for Ga in 6- and 4-fold coordination, respectively. Additionally, simulation of this resonance yields a quadrupolar constant CQ on the order of 13 MHz. Both shift and CQ values of this new 71Ga resonance point toward its assignment to a 5 coordinated Ga site, based on the similar NMR parameters obtained for Ga in trigonal bipyramidal geometry in LaGaGe2O7 (δ = 89.6 ppm, CQ = 11.6 MHz),[44] and for Ga  in square pyramidal geometry in Sr- and Mg-doped LaGaO3 (δ = 150 ppm, CQ = 10.1 MHz).[35] The present site-averaged GIPAW NMR calculations on La3Ga4Ge2O14.5 predict a 5-fold coordinated square pyramidal GaD site appearing at a shift of 177.3 ppm with CQ of 10 MHz (Table ), in fair agreement with the experimental data. The calculations also predict a very large CQ (>23 MHz) for a 5 coordinated GaC site which is not resolved or observed in the 71Ga MAS NMR spectra even under the optimal experimental conditions of very fast MAS (65 kHz) and ultrahigh magnetic field (20 T) used in this work, which demonstrates the extensive disorder in the system in good agreement with neutron diffraction data. These 71Ga NMR results highlight a change of coordination number of Ga atoms from four to five, needed to accommodate interstitial oxygens upon doping of Ge.

In summary, there is an excellent agreement between the experimental 17O and 71Ga NMR spectra and the ones simulated from the site-averaged GIPAW NMR calculations performed on DFT in the refinement of neutron diffraction data. The data show that interstitial oxygen is incorporated into the langasite framework by coordination to two adjacent tetrahedra to form a new bipyramidal (Ga,Ge)2O8 unit. The distorted pyramidal geometry of the (Ga,Ge)2O8 units in the average structure results from the high degree of disorder within the framework, both in terms of interstitial oxygen incorporation and the disordering of Ga3+ and Ge4+ over multiple sites. The combination of structural refinements and multinuclear NMR spectroscopy aided by DFT calculations enables a full understanding of the local defect chemistry in the doped langasites investigated.

AC Impedance Spectroscopy

Arrhenius plots of the total and bulk conductivity of La3Ga5–Ge1+O14+ for 0 ≤ x ≤ 0.5 are shown in Figure . For these compositions, the total conductivity values were extracted from the intercept of the arcs in the complex impedance plane and the type of response (bulk, grain boundary, or electrode) was assigned based on the capacitance values.[46] The parent material La3Ga5GeO14 was found to have a conductivity of 4 × 10–6 S cm–1 at 700 °C, which increased to a plateau of 3.6 × 10–4 S cm–1 with doping up to x = 0.3 (La3Ga4.7Ge1.3O14.15), with the appearance of the Warburg electrode response for the doped compounds which is characteristic of oxide ion conduction. Figure a shows the evolution of the complex impedance plots for La3Ga4.7Ge1.3O14.15 as a function of temperature. As the temperature increases, the electrode response becomes more dominant until at 800 °C the arcs are no longer resolved. At 800 °C and higher temperatures the total conductivity was extracted from the intercept.

Figure 10

(a) Complex impedance plots for La3Ga4.7Ge1.3O14.15 at 600, 700, and 800 °C showing the Warburg arc characteristic of oxide-ion conductors. (b) Complex impedance plot for the insulating La3Ga5GeO14 parent material at 700 °C. In both panels, the numbered red points denote the logarithm of the frequency at those points. (c) Arrhenius plots of the total conductivity (empty symbols) and bulk conductivity (filled symbols) of La3Ga5–Ge1+O14+ for x = 0, 0.1, 0.2, 0.3, and 0.5.

For all compositions, the total conductivity at temperatures above 525 °C is dictated by the grain boundary contribution and was found to be constant for samples where x ≥ 0.3 (see the overlap of the conductivity values for x = 0.3 and 0.5 in Figure c). Thus, in order to properly assess the change in conductivity in La3Ga5–Ge1+O14+ as a function of x we investigated the bulk conductivity in the lower temperature range of 350–525 °C. In this temperature range, the M″ vs frequency plot allows us to discriminate the bulk and grain boundary contribution to the total conductivity[47−49] under the conditions measured up to 2 MHz. Typical M″ plots for the composition La3Ga4.7Ge1.3O14.15 are shown in Figure S14. The maximum conductivity values were obtained from the compositions x = 0.4, 0.45, and 0.5 (see Figure a). In order to confirm the nature of the charge carriers, the conductivity of La3Ga4.5Ge1.5O14.25 (which lies within the most highly conducting 0.35 ≤ x ≤ 0.5 range) was measured as a function of oxygen partial pressure in the range 1 to 10–15 atm (Figure b). The conductivity was found to be constant over the entire oxygen partial pressure range, which is consistent with ionic conductivity as the dominant transport process. Furthermore, the temperature dependence of the conductivity in dry and wet air was also tested from 350 to 950 °C for x = 0 and 0.5 langasites and was found to be identical under the two different conditions, which suggests that there is no proton conduction in these materials (see Figure S15).

Figure 11

(a) Variation of the bulk conductivity for La3Ga5–Ge1+O14+ with 0 ≤ x ≤ 1.5. The corresponding activation energies are shown in Table S4. (b) p(O2) dependence of the bulk conductivity for La3Ga4.5Ge1.5O14.25. (c) Bulk conductivity at 500 °C for the melilite La1+Sr1–Ga3O7+[50] (black circles) and the langasite La3Ga5–Ge1+O14+ (blue squares) as a function of the concentration of extra oxygen atoms per channel, defined as x per A-site cation. The solid lines are a guide to the eye.

The oxide ion conductivities in the langasite and melilite structures are compared in Figure c, where the variation of the conductivity at 500 °C is plotted as a function of the concentration of extra oxygen atoms per channel (defined as x per A-site cation: i.e., x/2 for La1+Sr1–Ga3O7+ and x/3 for La3Ga5–Ge1+O14+). This shows that the increase in conductivity per oxygen interstitial is less pronounced in langasites. Moreover, the conductivity reaches a maximum value of 2.4 × 10–5 S·cm–1 for the La3Ga5–Ge1+O14+ langasite at a concentration of ∼0.15 interstitial oxygen per A-site, whereas in the La1+Sr1–Ga3O7+ melilite the conductivity continues to increase with oxygen content to reach a conductivity of 5 × 10–3 S·cm–1 at 0.3 interstitial oxygen per A site, 2 orders of magnitude higher than the most conductive langasite.

The difference in the response of conductivity to the concentration of oxide interstitials between the langasite and melilite structures reflects the different structural responses to doping in the two frameworks. In melilite, the interstitial oxide O4 is accommodated close to the center of the 5-membered channels where it is coordinated by two La3+/Sr2+ and one Ga3+ nearest-neighbor cation. This satisfies the local bonding requirements of O4, for example, producing a bond valence sum (BVS) of 1.90 in La1.54Sr0.46Ga3O7.27,[10] without requiring a substantial structural rearrangement: atom displacements of up to 0.44 Å permit the accommodation of O4 in a local GaO5 unit (Figure ), with 50% of the BVS for O4 contributed by bonding to La3+/Sr2+ in adjacent layers. In the langasite structure, it is not possible to accommodate the interstitial oxide O4 in an analogous site directly between La3+ cations from adjacent layers, close to the center of the 6-membered channels: while this would produce a pair of chemically plausible La–O4 distances similar to those found in the doped melilite, the expansion of the channel by the incorporation of the BO6 octahedra means that the resulting (Ga,Ge)–O4 distances (minimum 2.495 Å) would be substantially longer than their melilite counterparts (1.81 Å). A BVS calculation corresponding to such a model confirms that O4 would be heavily under-bonded in such a position (BVS = 1.17). Thus, in the langasite structure, the inclusion of the BO6 octahedra forces the interstitial oxide O4 to be accommodated further away from the center of the channel.

Figure 12

Comparison of the local structural relaxation of langasite and melilite frameworks upon incorporation of interstitial oxygen. (a) Refined crystal structure of parent undoped La3Ga5GeO14, shown in color, with the refined structure of doped oxygen interstitial containing La3Ga3.5Ge2.5O14.75 overlaid in gray, showing the large atom displacements associated with relaxation of the framework around interstitial O4 by the formation of a (Ga,Ge)2O8 unit. (b) Structure of the melilite LaSrGa3O7, in color, with the refined structure of La1.54Sr0.46Ga3O7.2710 overlaid in gray, showing a less dramatic relaxation of the structure around the interstitial oxide O4 by formation of a single trigonal bipyramidal GaO5 unit. Arrows indicate approximate displacement directions. Red = oxygen, green = Ga/Ge, blue = La/Sr, gray = atoms belonging to doped framework.

The formation of the (Ga,Ge)2O8 unit in langasite allows the bonding requirements of both the interstitial oxide O4, and the displaced framework oxide O2b, to be met simultaneously (BVS of 1.7(1) and 2.1(1) respectively) by coordination to 2× (Ga,Ge) and 1× La3+ nearest neighbors, in contrast to melilite. This means that the majority of the BVS contribution for interstitial oxide comes from bonding to the framework (Ga,Ge) cations (60% and 65%, respectively, for O2b and O4), rather than the La3+ cations from the adjacent layers (which contribute 40% and 35%, respectively), indicating an increased reliance on the framework bonding to stabilize the interstitial oxide in langasite. The doubly bridged (Ga,Ge)2O8 unit is formed from one CO4 four-connected and one DO4 three-connected octahedron and requires displacement of the O2 oxygen that bridges them. In melilite, such a displacement would disrupt the local coordination at both of the distinct framework forming sites by changing both their coordination numbers, favoring coordination solely to the three-connected tetrahedron. Langasite, however, features a third polyhedron that forms the framework, the octahedral BO6 site, which does not increase its primary coordination sphere when the dimer of the bridged tetrahedra forms. The BO6 octahedron is thus able to buffer the relaxation at the two tetrahedral sites through changes in bond angles described by the O3 anion, which it shares with the four-connected GaD site. The extra octahedral site in langasite thus both enlarges the channel to favor oxide displacement and enables that displacement to change the coordination environment of both the three- and the four-connected tetrahedra.

The associated local atom displacements, including a displacement of 1.2 Å from O2 to O2b to form a (Ga,Ge)2O8 unit (Figure a), represents a deep thermodynamic trap for the interstitial oxides because of the extensive relaxation that would have to accompany any movement of the interstitial defect. This is reflected in the higher activation energies in La3Ga5–Ge1+O14+ (∼1.1 eV) compared to La1–Sr1+Ga3O7+ (∼0.4 eV). Consistent with this, the activation energies tend to increase with interstitial content in the langasites (Table S4), which may be associated with increasing structural relaxation around the greater defect content. Displacement of the interstitial in langasite requires associated movement of the extensive framework relaxation around it. Additionally, it is clear that the three-dimensional framework connectivity introduced by the BO6 octahedra does not result in a net improvement to the conductivity of the langasite framework compared to that of melilite. On the contrary, by allowing larger structural relaxation around the interstitial oxygen by ensuring the maintenance of framework connectivity despite coordination changes at the two tetrahedral units, the octahedra make defect transport more difficult than in melilite. The octahedra increase the size of the channels to favor the accommodation of interstitial oxide far from the channel centers as part of a heavily relaxed dimer formed by the two distinct originally tetrahedral sites. The plateau in the conductivity plot at 0.15 oxygen atoms per channel (Figure c) implies that, in addition to defect concentration, competing structural and chemical factors (e.g., increasing local strain) may have a strong influence on the conductivity of the system. Further work is required on other families of langasite ionic conductors to distinguish the impact of such structural factors, which are likely to be common to all langasite compositions, from other chemical factors that may be influential (e.g., the increasing concentration of Ge4+ upon doping, which tends to be less amenable to 5-coordinate geometry than Ga3+).[10,11,18]

Conclusions

We have combined neutron diffraction, multinuclear NMR spectroscopy, and DFT calculations to show that the langasite structure of La3Ga5GeO14 can accommodate a substantial concentration of interstitial oxide ions by tuning the Ga/Ge ratio according to La3Ga5–Ge1+O14+, with the solid solution extending to x = 1.5. The interstitial oxides are accommodated by formation of (Ga,Ge)2O8 units within the corner-linked tetrahedral network. This corresponds to a 5.4 mol % excess of oxide ions with respect to the parent framework, which is higher than that of the established interstitial oxide ion conductors such as the oxyapatites, which can accommodate a theoretical maximum of 3.8 mol % excess oxide,[7] and melilites which accommodate up to 4.6 mol %.[50] To access such high doping levels in La3Ga5–Ge1+O14+ requires a sol–gel synthesis that allows the phase to form at relatively low temperatures, suggesting that the materials are metastable.

AC impedance spectroscopy at 500 °C showed that the ionic conductivity of La3Ga5–Ge1+O14+ increases from 2.2 × 10–7 S cm–1 at x = 0 to a maximum of 2.4 × 10–5 S cm–1 in the range 0.4 < x < 0.5 and then decreases at higher doping levels to 6.9 × 10–6 S cm–1 at x = 1.5. The existence of a maximum in conductivity with x and the lower conductivities obtained at all x compared to those of the melilite La1+Sr1–Ga3O7+ are consistent with the large structural rearrangement required to accommodate the (Ga,Ge)2O8 units in langasite. This acts to trap the interstitial oxide, reducing its mobility relative to that of melilite, where the structural perturbation associated with interstitial formation is smaller. The incorporation of corner sharing BO6 octahedra into the langasite framework enables the formation of the complex dimer structure around the defect by allowing both tetrahedral sites, four- and three-connected, to relax heavily while maintaining its own geometry and retaining network connectivity. Although the octahedra increase the dimensionality of the framework, they allow it to relax much more extensively in response to interstitial defects and thus reduce their mobility. The complexity of the relationship between conductivity, composition, and structure in langasite motivates further study to resolve the structural and chemical contributions to ionic trapping in this system and, particularly, to identify distributions of cations at the three structure-forming sites that reduce the complexity and thus intertia of the defect formed around the interstitials that are readily introduced into the structure.

While the oxide ion conductivity of the La3Ga5–Ge1+O14+ langasite is not competitive with existing oxide electrolytes, the langasite structure is chemically diverse, accommodating a wide range of cations on all of the sites in the structure: examples from the ICSD[51] with the langasite prototype finds six different A-site cations (including representatives from the alkali metals, alkaline earths, rare earths, and main group) and 21 different B site cations (including alkaline earths, diverse transition metals, and main group elements). There is therefore great scope to tune the chemistry of the langasites to control the defect trapping identified here, for example, by identifying combinations of cations on the three distinct polyhedral sites that stabilize interstitials with less pronounced framework relaxation, which may lead to the synthesis of oxide ion conductors with superior properties.

Methods

Synthesis

The syntheses of La3Ga5–Ge1+O14+ with 0 ≤ x ≤ 0.3 were carried out by a conventional ceramic method by mixing stoichiometric amounts of binary oxide starting materials La2O3 (Alfa Aesar, Reacton 99.99%, dried at 950 °C for 12 h), Ga2O3 (Alfa Aesar, 99.99%, dried at 220 °C), and GeO2 (Sigma-Aldrich, ≥ 99.99%, dried at 220 °C). The precursors were mixed by hand in an agate mortar and then annealed in alumina crucibles at 1300 °C using heating and cooling rates of 3 °C·min–1 and dwell times of 12 h.

Powders of La3Ga5–Ge1+O14+ with x > 0.3 were prepared by a sol–gel method. Stoichiometric amounts of La(NO3)3·6H2O (Alfa Aesar, 99.99%), Ga(NO3)3·6H2O (Alfa Aesar Puratronic, 99.999%), and GeO2 (Sigma-Aldrich, ≥99.99%) were dissolved in deionized water to make an aqueous solution of concentration of langasite of 10–3 mol dm–3. One molar equivalent of citric acid and ethylene glycol was then added, and the solution was heated and stirred for 14 h until a dry gel was obtained. The gel was then calcined at 600 °C for 60 h to remove the organic components, to produce an amorphous carbon-free powder (confirmed by CHN analysis). The amorphous powder was pressed into pellets and fired at 800–1200 °C for 12 h (see Figure S2 for relationship between x and firing temperature). The langasite phase was found by powder X-ray diffraction (PXRD) to start forming at 700 °C, and the synthetic temperature was optimized for every composition studied. All the compositions studied were examined by energy dispersive X-ray (EDX) analysis in the transmission electron microscope (Figure S1). The synthesis of 17O enriched samples for 17O NMR was performed by heating the samples in a sealed Pyrex tube under 60% 17O enriched 17O2 gas (Isotec, used as received) at 750 °C for 24 h using heating and cooling rates of 5 °C·min–1, following a standard procedure.[21,35]

Powder Diffraction

Routine analysis of phase purity and lattice parameters were performed with a Bruker D8 Advance diffractometer with a monochromated Cu source (Kα1, λ = 1.54418 Å) in Bragg–Brentano mode with sample rotation. Time of flight neutron diffraction data were collected using the HRPD instrument at ISIS (U.K.). Samples were contained in thin-walled vanadium cans. Nonambient temperatures were accessed using either an in situ closed-cycle refrigerator (e.g., for data at 9 K) or an in situ furnace (for T > 300 K). Rietveld refinements were performed using TOPAS Academic version 5.[52] For the backscattering bank, instrument parameters were fixed to calibrated values, while for the 90° and 30° banks, a sample displacement parameter was refined. A time-of-flight peak shape and a Chebyschev background function was refined for all banks. Maximum entropy analysis was performed using JANA2006[53] with BayMEM.[25_1]

DFT Calculations

Periodic plane-wave DFT calculations were performed using the VASP[54] and CASTEP[27] packages. Initial structural relaxations with VASP were run in a 2 × 2 × 2 supercell of the parent La3Ga5GeO14 cell, with a 2 × 2 × 3 k-point grid. Cell parameters and atomic positions were optimized until all atomic forces were below 0.01 eV/Å, using the PBE functional,[55] projector augmented-wave potentials,[56] and a plane-wave cutoff energy of 600 eV. Calculation of NMR parameters was carried out using GIPAW[25_1,26] as implemented in CASTEP, after structures had been reoptimized with the same functional and cutoffs as those used with VASP, and using pseudopotentials automatically generated on-the-fly by CASTEP. The calculated isotropic magnetic shieldings (σiso) were converted to the isotropic chemical shifts (δiso,cs) to allow comparison with the experimental values as detailed in Table . The quadrupole coupling constants are obtained as CQ = eQV/h and the asymmetry parameter as ηQ = (V – V)/V, where an ordering |V| ≥ |V| ≥ |V| of the principal components of the traceless electric field gradient tensor is assumed.

Solid State NMR

71Ga NMR experiments were performed on a 20 T Bruker Avance II 850 MHz spectrometer using a Bruker triple resonance 1.3 mm HXY (in double resonance mode) for fast magic angle spinning (MAS) experiments at rotational rates νr = 65 kHz and using a Bruker triple resonance 3.2 mm HXY (in double resonance mode) for static experiments, both probes tuned to X = 71Ga at ν0 = 259.3 MHz. One-dimensional MAS NMR spectra were recorded using a rotor-synchronized (1 period) Hahn echo sequence with selective pulses (π/2 pulse length of 4 μs) at a radio frequency (rf) field amplitude of ν1 = 16 kHz. A double-frequency sweep (DFS)[57] pulse of 2 ms from 800 to 200 kHz at a rf field amplitude of 15 kHz giving an optimum signal enhancement of approximately 2 was used for all 71Ga MAS experiments. One-dimensional static NMR spectra were obtained with QCPMG acquisition with 3000 echos and 204 points per echo. A recycle delay of 2 s, sufficient to obtain quantitative data, was used for all 71Ga experiments.

17O NMR experiments were carried out on 9.4 T Bruker Avance III 400 MHz spectrometer using a Bruker triple resonance 4 mm HXY (in double resonance mode) tuned to X = 17O at ν0 = 54.25 MHz and spinning the samples at a MAS rate of νr = 14 kHz, and on a 20 T Bruker Avance II 850 MHz spectrometer using a Bruker triple resonance 3.2 mm HXY (in double resonance mode) tuned to X = 17O at ν0 = 115.28 MHz and spinning the samples at a MAS rate of νr = 22 kHz. Rotor synchronized Hahn echo experiments were carried out with one rotor period using a pulse length π/2 = 1 μs at a rf field amplitude of ν1 = 83 kHz and a recycle delay of 1 s. Two-dimensional triple-quantum MAS experiments[31] were performed at 9.4 T using a z-filter pulse sequence.[32] Hard and soft pulses were performed at rf field amplitudes of 83 kHz and 10 kHz, respectively. The calculation of the 17O isotropic chemical shifts δiso,cs from the 3QMAS z-filter data are described in the Supplementary Methods.

The 71Ga and 17O chemical shifts were externally referenced to a 1M solution of Ga(NO3)3 in water and to water, respectively, all at 0.0 ppm. NMR data were processed and simulated using the Brucker Topspin package.

Ceramic Processing

Three different sintering protocols were necessary to produce dense ceramics suitable for conductivity measurements across the full compositional range, which are summarized in Tables S5 and S6. For compositions 0 ≤ x ≤ 0.20, pellets with >86% of the crystallographic density were prepared by ball-milling the powders, pelletizing, and then cold isostatic pressing before annealing in air at 1275 °C for 24 h. For compositions x ≥ 0.30, large (5 g) batches of presynthesized powders were ball-milled, before preparation of dense pellets by spark plasma sintering (SPS) under vacuum in a Sumitomo Coal SPS-1050. For compositions 0.3 ≤ x ≤ 0.7, a 20 mm diameter graphite die was used, with an applied pressure of 50 MPa and sintering temperatures of 1050–1150 °C. This produced ceramics with up to 97% of the crystallographic density, which were then subjected to an additional ex situ annealing step (700 °C in flowing oxygen for 60 h) to remove residual graphite, which was confirmed by CHN analysis. For the sintering of the most heavily substituted compositions (x = 1.0, 1.5), the lower decomposition temperatures necessitated the use of a double-acting WC die of 5 mm diameter[58] to apply a pressure of 550 MPa with sintering temperatures of 800–850 °C. This produced ceramics with ∼73% of the crystallographic density. A heating/cooling rate of 200 °C min–1, and dwell times of 5–10 min, were used for all samples processed in this way. Rapid pulses (duration 3 ms) were used with a pattern of 12:2 on and off pulses. In all cases, the phase purity of the densified ceramics was confirmed by PXRD prior to electroding for conductivity measurements (Figure S16).

Conductivities were determined by AC impedance spectroscopy using an Agilent E4980 LCR meter over the 2 MHz–20 Hz frequency range and a Solartron 1255B frequency response analyzer coupled to a 1296 dielectric interface over the 1 MHz–100 mHz frequency range, using 300 mV perturbation voltage for both instruments. The measurements were carried under a flow of dry air at temperatures of 350–1000 °C. The samples were electroded with gold paste, which was cured at 625 °C prior to data collection. For comparison of conductivities of samples with different densities (e.g., the data presented in Figures c and 12a), the measured dielectric permittivity values were corrected for differences in porosity by the Heidinger model.[59−61] The p(O2)-dependence of the conductivity in the range 10–15–1 atm pO2 was evaluated at 400, 500, 600, and 800 °C in a gas flow by a controlled mixing of Ar, O2, CO, and CO2 in different ratios. The partial oxygen pressure was monitored by an YSZ potentiometric sensor, and equilibration of the samples with the gas environment was ensured at each data point collected by the dwelling at each temperature for several hours until a stable signal was obtained.