The substitution of oxygen bonded to Si4+ with other anions is generally difficult due to the strong Si–O bonding, except under harsh conditions20,21,22,23. In this Article, the following synthetic route was employed; the low-temperature (400–700 °C) solid-state reaction of Ba(NH2)2 and SiO2 in a NH3 flow was applied to the one-step synthesis of Ba3SiO5−xNyHz with heavily substituted H− and N3− (Methods and Supplementary Fig. 1). The X-ray diffraction (XRD) pattern for Ba3SiO5−xNyHz synthesized at 600 °C indicates a single Ba3SiO5 phase with a purity of >99% and a large shift to lower diffraction angles from that of Ba3SiO5 (Fig. 1 and Supplementary Fig. 2). The lattice parameters of Ba3SiO5−xNyHz are much larger than those of Ba3SiO5 as determined by Rietveld fitting analysis (∆V/V0 = +9.0%; V and V0 represent volume of sample and reference Ba3SiO5, respectively) (Supplementary Table 1). The H− and N3− contents of Ba3SiO5−xNyHz were determined to be 3.74 and 1.60 mmol g−1, respectively, based on temperature-programmed desorption (TPD) and an acid dissolution method (Methods), which gave the composition of Ba3SiO2.87N0.80H1.86. Diffuse reflectance spectroscopy (DRS) and projected density of states analysis of Ba3SiO5−xNyHz suggest that the N2p bands are located above the H1s and O2p bands and thus contribute to bandgap narrowing compared to white Ba3SiO5 powder (Supplementary Fig. 3).

a, XRD patterns of Ba3SiO5−xNyHz and Ba3SiO5. b, Solid-state 1H MAS NMR spectra of Ba3SiO5-xNyHz, BaH2 and Ba(NH2)2. c, Solid-state 29Si MAS NMR of Ba3SiO5 and Ba3SiO5−xNyHz. d, Calculated crystal structure of Ba3SiO2.5NH2 (left) with SiO2NH and Ba6H2 blocks (right). e, Lattice H− and N3− contents in Ba3SiO5−xNyHz before (fresh) and after heating in Ar at various temperatures. The used sample was obtained by heating the sample (Ar/650 °C) under ammonia synthesis conditions (400°C, 0.1 MPa) for 2 h. Insets: photographs of the corresponding Ba3SiO5−xNyHz powders. The error bars represent the standard deviation of the mean based on n = 3 independent measurements. f, X-band EPR spectra of Ba3SiO5−xNyHz before and after heating in Ar at various temperatures. a.u., arbitrary units.

Source data

The location of N and H in Ba3SiO5−xNyHz has been confirmed by 1H and 29Si solid magic-angle spinning (MAS) NMR spectroscopy analysis. In the 1H MAS NMR spectrum (Fig. 1b), the Ba3SiO5-xNyHz sample shows a main sharp signal at 8.3 ppm. This is assigned to Ba–H species as its chemical shift is very similar to that of BaH2 (10.3 ppm). The peaks at 3.1 and 0.8 ppm are attributed to Si–H and N–H species, respectively. The former is well consistent with H-terminated Si nanocrystals and molecular silicon hydrides (3–6 ppm)24 and the latter is close to those of Ba(NH2)2 (0.2 ppm) and Ca(NH2)2 (0.7 ppm)25. The N–H peak is derived from surface NHx species on the Ba3SiO5−xNyHz (Supplementary Fig. 4). In the 29Si MAS NMR spectrum (Fig. 1c), both Ba3SiO5−xNyHz and Ba3SiO5 have signals at −66.8 ppm, which are attributed to the isolated SiO4 tetrahedra (orthosilicate anion) unit (Q0 site). As well as this signal, new signals at −53.6, −41.3 and −29.7 ppm appeared for Ba3SiO5−xNyHz. These peaks are assignable to the Q0 site with different anion coordinations since the 29Si chemical shift tends to be more shielded with increasing group electronegativity sums of ligands bonded to Si26,27. Possible species are SiO2NH, SiONH2 and SiOxNHx with surface OH and/or NH species, respectively. The tetragonal Ba3SiO5 crystal has two types of oxygen site, that is four OI sites in SiO4 and one OII site surrounded by six barium atoms (Ba6O). Density functional theory (DFT) calculations suggest that the Ba3SiO2.5N1.0H2.0 unit cell is mainly comprised of N–H pairs that substitute two OI sites (2O2−⇒ N3− + H−) to form SiO2NH blocks and extra H–H pairs that substitute one OII site (O2−⇒ 2H−) to form Ba6H2 blocks (Fig. 1d and Supplementary Figs. 5 and 6). In the Ba6H2 block, one H− ion is located at an OII site and another H− ion occupies an interstitial site adjacent to the Ba6H1 unit, resulting in lattice expansion (∆V/V0 = +9.4%, Supplementary Tables 1–4). The Ba3SiO5, which is regarded as an antiperovskite, can accommodate a variety of combinations of elements because of its sufficient lattice space28, enabling a unique mixed-anion structure.

The yellow-coloured Ba3SiO5−xNyHz powder was heated under Ar gas flow to monitor the thermal stability of lattice H− and N3−. Upon increasing the temperature from 400 to 650 °C, the sample colour changed from light yellow to light green and finally to dark green (Fig. 1e). Most of the lattice H− and a part of the N3− (from the subsurface region) can be removed by heating at 650 °C in Ar flow (Fig. 1e, Supplementary Fig. 7 and Supplementary Table 5), leading to introduction of a high density of Va sites into the crystal without destruction of the tetragonal Ba3SiO5 framework (Supplementary Figs. 8–11). The colour change due to the introduction of Va sites is well understood by electron paramagnetic resonance (EPR) and DRS analysis (Fig. 1f and Supplementary Figs. 12–14). The as-prepared Ba3SiO5−xNyHz exhibits weak EPR signals at g of approximately 2.004, 2.003 and 2.002 (Fig. 1f and Supplementary Fig. 12). The intensity of these signals was enhanced with the heating temperature and a new EPR signal at g = 1.967 with strong intensity appeared above 600 °C. These two sets of peaks could be attributed to the unpaired electron trapped at the Va of the SiO2NH and Ba6H2 sites. The unpaired electron density of the sample heated at 650 °C was estimated to be 1.1 × 1020 cm−3 (Supplementary Fig. 13), which is smaller than the total electron density (approximately 7.4 × 1020 cm−3) that was evaluated by the iodometric titration method29. This indicates that the sample has both unpaired electrons and paired electrons in the lattice. Moreover, when the dark green Ba3SiO5−xNyHz powder (Ar/650 °C) was heated under ammonia synthesis conditions, the lattice H− and N3− were regenerated and the dark green colour turned to the original yellow (used) (Fig. 1e).

The fresh Ba3SiO5−xNyHz without any TM site shows continuous ammonia production at approximately 0.37 mmol gcat−1 h−1 without degradation in activity for 150 h at 400 °C and 0.9 MPa (Fig. 2a and Supplementary Table 6). The total amount of ammonia was estimated to be approximately 5.25 mmol (per 0.1 g of catalyst), which is much higher than that of the amount of lattice N3− (0.16 mmol) and H− (0.37 mmol). This indicates that the ammonia produced originates from the activation of molecular N2 and H2 but not from the decomposition of Ba3SiO5−xNyHz (Supplementary Fig. 15). There was no colour change or alteration of the crystal structure after the 150 h reaction (Fig. 2b and Supplementary Figs. 8 and 9), which suggests that the Ba3SiO5−xNyHz is chemically stable under the ammonia synthesis conditions. When the Ba3SiO5−xNyHz was pretreated in Ar flow at 650 °C for 2 h, XRD peaks shifted towards higher angles because of removal of most lattice H− and surface N3− ions (Fig. 2b). The Va-introduced Ba3SiO5−xNyHz exhibited more than three times the NH3 synthesis rate (approximately 1.20 mmol g−1 h−1) than the original one, probably due to the increase of specific surface area from 8.0 to 19.5 m2 g−1. After the ammonia synthesis reaction, the Va sites were re-occupied by H− and N3− (Supplementary Table 5), leading to a shift of XRD peaks towards the original positions and the recovery of the 1H NMR signal (Fig. 2b,c and Supplementary Fig 11a). Neither white Ba3SiO5, Ba3Si6O9N4, BaSi2O2N2 or Ba(NH2)2 nor SiO2 powder exhibits activity for ammonia synthesis, even at temperatures up to 540 °C. By contrast, the Ba3SiO5−xNyHz powder (Ar/650 °C) can effectively activate N2 to produce ammonia at temperatures down to 300 °C (0.20 mmol g−1 h−1, Fig. 2d) with a low apparent activation energy (Ea) of approximately 68.5 kJ mol−1, which outperforms that of the conventional Ru-loaded MgO catalyst (300 °C, 0.01 mmol g−1 h−1, Ea = 114.5 kJ mol−1). TPD and hydrogen temperature-programmed reduction (H2-TPR) analysis results show that lattice N3− ions in conventional Ba–Si oxynitrides with Si–N–Si bonding are very stable and could not be reduced to ammonia up to 900 °C (Supplementary Fig. 16). This is totally different from Ba3SiO5−xNyHz. We suggest that easy thermal desorption of N3− and H− mainly comes from the orthosilicate structure of Ba3SiO5−xNyHz, where SiX4 tetrahedra (X = O, N, H) do not connect with each other and lattice N and H are coordinated by not only Si but also Ba. Therefore, when the Va sites are formed, electrons at Va sites are stabilized by the Coulombic interaction with Ba (Fig. 2c). The lattice H− ions do not directly contribute to hydrogenation of N2 but provide a number of Va sites to capture N2 molecules in the gas phase (Supplementary Fig. 17). It can be expected that lattice H− and N3− ions are continuously exchanged with molecular H2 and N2 via the Va formation on Ba3SiO5−xNyHz during ammonia synthesis. Thus, the Ba–Si orthosilicate oxynitride-hydride was shown to function as a TM-free ammonia synthesis catalyst.

a, Time courses for ammonia synthesis over as-prepared Ba3SiO5−xNyHz, Ba3SiO5−xNyHz pretreated at 650 °C in Ar for 2 h and Ba3SiO5 at 400 °C and 0.9 MPa. b, XRD patterns of as-prepared Ba3SiO5−xNyHz (fresh) and Ba3SiO5−xNyHz pretreated at 650 °C in Ar for 2 h before and after the catalytic test in a. c, Schematic illustration of thermal induced Va and electron formation and the N, H regeneration under NH3 synthesis conditions in Ba3SiO5−xNyHz. d,e, Temperature dependence of the NH3 synthesis rate over TM-free materials (d) and the Ru (1.5 wt%)-loaded samples under a pressure of 0.9 MPa (e). f, NH3 synthesis rates for various Ru-based catalysts at 300 °C. g, HAADF-STEM image (top) and EDX-mapping (bottom) of used Ru/Ba3SiO5−xNyHz_HS.

Source data

The Ru (1.5 wt%)/Ba3SiO5−xNyHz catalyst functioned as an efficient catalyst for ammonia synthesis at above 200 °C (0.32 mmol gcat−1 h−1) and reached 27.1 mmol gcat−1 h−1 at 400 °C (Fig. 2e and Supplementary Fig. 18), which is far beyond that of the Ru/Ba3SiO5, Ru/Ba3Si6O9N4 and Ru/BaSi2O2N2 catalysts. The high ammonia synthesis activity was maintained for more than 170 h without crystal structure decomposition (Supplementary Fig. 19 and Supplementary Table 5), which demonstrates its excellent catalytic stability. This result is in contrast to that of the reported TM-free catalyst, potassium hydride-intercalated graphite composite catalyst (KH0.19C24). The activity of KH0.19C24 is not promoted by the supported TM catalysts19. The Ea of Ru/Ba3SiO5−xNyHz was as low as 64.4 kJ mol−1, which is much lower than that of Ru/Ba3SiO5 (110.3 kJ mol−1) and the reported conventional Ru-based catalysts (85–121 kJ mol−1). When Ru/Ba3SiO5−xHyNz particles (Ru: 1.5 wt%) were dispersed on the SiO2 surface, the resultant sample (Ba3SiO5−xNyHz_HS) with high surface area (105 m2 g−1) (Supplementary Fig. 20) exhibited much higher catalytic performance than that of Ru/Ba3SiO5−xHyNz. After the optimization of the reaction conditions (Supplementary Fig. 21), Ru (5.0 wt%)/Ba3SiO5−xNyHz_HS recorded the highest ammonia synthesis rate among the reported catalysts (Fig. 2f and Supplementary Table 7)12,15,16,30,31,32,33. In addition, the activity of Ru/Ba3SiO5−xNyHz_HS was maintained even after exposure to air for 2 h (Supplementary Fig. 22). The excess SiO2 may improve the stability of the catalyst in air.

High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) analysis showed that small-sized Ru nanoparticles (approximately 3 nm) are highly dispersed on the surface of Ba3SiO5−xNyHz (Supplementary Fig. 23). The Ru particle size of Ru/Ba3SiO5−xNyHz is smaller than that of Ru/Ba3SiO5, while the large activity difference between them cannot be explained by the Ru particle size. For Ru/Ba3SiO5−xNyHz_HS, the Ba3SiO5−xNyHz particles with sizes of 100–200 nm are dispersed on the surface of SiO2 (Fig. 2g and Supplementary Fig. 24). As a result, the active Ba3SiO5−xNyHz loaded with tiny Ru nanoparticles (approximately 2.5 nm) is effectively exposed on the catalyst surface for Ru/Ba3SiO5−xNyHz_HS. The valance state of Ru on Ba3SiO5−xNyHz is almost neutral (metallic) but slightly positive according to the Ru K-edge X-ray absorption near edge structure (XANES) (Supplementary Fig. 25) and X-ray photoelectron spectroscopy (XPS) results (Supplementary Fig. 26). This suggests that the electrons are donated from Ru to Ba3SiO5−xNyHz due to the strong Ru–N interaction. This result is rather different from those of conventional Ru-based ammonia synthesis catalysts that require negatively charged Ru to activate N2. The reaction orders of N2 (α), H2 (β) and NH3 (γ) for Ru/Ba3SiO5−xNyHz were determined to be 0.55, 0.48 and −0.63, respectively (Supplementary Fig. 27). In particular, the N2 reaction order (α = +0.55) is only half that of conventional catalysts (α = ~1.0), where the overall reaction rate is limited by the N2 cleavage step. This result indicates the change of RDS from N2 dissociation to other elementary steps over the Ru/Ba3SiO5−xNyHz catalyst. The H2 reaction order for Ru/Ba3SiO5−xNyHz is positive (β = +0.48), which suggests that the Ru/Ba3SiO5−xNyHz is not subject to H2 poisoning.

Ammonia synthesis employing a mixed gas flow of 15N2 and H2 at 400 °C was conducted to elucidate the catalytic reaction mechanism over Ba3SiO5−xHyNz and Ru/Ba3SiO5−xHyNz. The m/z = 16, 17 and 18 signals gradually increased with decreasing m/z = 30 (15N2) and m/z = 2 (H2) (Supplementary Figs. 28 and 29), which indicates the effective formation of ammonia over Ba3SiO5−xHyNz in the absence of any TM sites (Fig. 3a). No 29N2 formation indicates that the N2 isotope exchange reaction (N2-IER) does not proceed since the Ba3SiO5−xHyNz does not have direct N2 dissociation ability. In contrast to Ru/MgO (Supplementary Fig. 30), the mass signal intensity ratios of (m/z = 16)/(m/z = 18) and (m/z = 17)/(m/z = 18) over Ba3SiO5−xNyHz were much larger than the theoretical values of 0.8 and 0.075 (Fig. 3b), which indicates the formation of 15NH3 as well as 14NH3. N2 gas with 15N was desorbed from the Ba3SiO5−xNyHz used after the isotope-labelled ammonia synthesis test (Fig. 3c), indicating that lattice 14N3− of Ba3SiO5−xNyHz was partially replaced with 15N3− from gaseous 15N2 during the isotope test. Similar results were observed for Ru/Ba3SiO5−xHyNz (Fig. 3d–f and Supplementary Fig. 31), which suggests the Ru-loading does not change the Va-mediated N2 activation processes. The concentration of lattice 15N3− in Ru/Ba3SiO5−xNyHz reaches approximately 37%, which is more than five times higher than that of bare Ba3SiO5−xNyHz (approximately 7%) and illustrates that loading of Ru nanoparticles would facilitate the formation of Va sites at the Ru–support interface as N2 activation centres. The catalytic cycle for Va-mediated N2 activation to ammonia was further confirmed by switching the reaction gas atmosphere between H2 and N2 (Supplementary Fig. 32). Accordingly, the lattice N3− ions in Ba3SiO5−xNyHz were consumed by hydrogenation and were immediately regenerated by the reaction of Va sites with molecular N2. This shows that the transiently formed Va sites function for N2 activation. To confirm the validity of this model, we examined the effect of Ni (which is known to have weak interaction with N) loading, and found no significant enhancement of activity (2.1 mmol g−1 h−1 at 400 °C and 0.9 MPa) and Va formation compared with the bare Ba3SiO5−xNyHz (Supplementary Figs. 33 and 34).

a,b, Reaction time profiles for NH3 synthesis from 15N2 and H2 over Ba3SiO5−xNyHz (a) and the change in mass signal ratio over reaction time (b). c, TPD for Ba3SiO5−xNyHz collected after the isotope-labelled ammonia synthesis. d,e, Reaction time profiles for NH3 synthesis from 15N2 and H2 over Ru/Ba3SiO5−xNyHz (d) and the change in mass signal ratio over reaction time (e). The dotted lines in b and e represent the theoretical fragment ratio (NH2/NH3 = 0.8 and NH/NH3 = 0.075). f, TPD for Ru/Ba3SiO5−xNyHz collected after the isotope-labelled ammonia synthesis. The content of 15N in the desorbed nitrogen gas was calculated from the shaded area in panels c and f.

Source data

N2 activation on the Ba3SiO5−xNyHz catalyst was further examined using diffuse-reflectance infrared Fourier transform spectroscopy (DRIFTS). It should be noted that Ba3SiO5−xNyHz without Ru exhibits a peak centred at around 2,016 cm−1, which is much lower than that of conventional Ru catalysts (Ru/MgO, 2,231 cm−1)34,35 and electride-based catalysts (Ru/C12A7:e−, 2,176 cm−1)12 (Fig. 4a and Supplementary Fig. 35). This is in contrast to N2 adsorption on conventional non-TM sites through electrostatic interaction, where N2 serves as a weak base molecule interacting with acid sites (electropositive sites) resulting in a blue-shift of the N≡N bond frequency36,37,38. In the case of Ba3SiO5−xNyHz, electrons trapped at the Va sites directly interact with N2 and facilitate π backdonation into the antibonding orbital of the N2 molecule, which leads to a red-shift of the N2 adsorption band. The N2 peak position slightly blue-shifts after Ru loading on the Ba3SiO5−xNyHz surface, which indicates that N2 molecules are mainly captured at the Va sites of Ba3SiO5−xNyHz accompanied by the influence of the positively charged Ru site. The N2-IER rate of Ru/Ba3SiO5−xNyHz is negligibly smaller than that of Ru/MgO although Ru/Ba3SiO5−xNyHz showed an over ten times higher ammonia synthesis rate than Ru/MgO (Fig. 4b and Supplementary Fig. 36). This result indicates that the Ru nanoparticles on Ba3SiO5−xNyHz do not function to dissociate N2, which is a striking difference from the well-known role of Ru in NH3 synthesis39,40,41.

a, DRIFTS N2 adsorption on Ru/MgO, Ru/Ba3SiO5−xNyHz and Ba3SiO5−xNyHz at −170 °C. b, Comparison of N2-IER rate and ammonia synthesis rate (0.9 MPa) at 400 °C for Ru/MgO and Ru/Ba3SiO5−xNyHz. The error bars represent the standard deviation of the mean based on n = 3 independent measurements. N.D., not detected. c, In situ DRIFTS observation of formation of intermediates on Ba3SiO5−xNyHz (Ru free) and Ru/Ba3SiO5−xNyHz (Ru loaded). The measurement temperature was increased from 30°C (blue) to 300°C (brown).

Source data

In situ DRIFTS measurements (Fig. 4c and Supplementary Fig. 37) showed N2 adsorption peaks at 2,161 and 1,976 cm−1, which may be attributed to N≡N stretching of N2 molecules adsorbed at different Va sites close to Si and Ba sites on Ba3SiO5−xNyHz. In particular, the latter peak was red-shifted relative to that observed at −170 °C (2,016 cm−1). This means that the N≡N bond is effectively weakened at reaction temperatures42. In addition, the peak intensity steeply increased with increasing reaction temperature especially above 200 °C. This is due to the accumulation of N2 molecules at the Va sites not only on the top surface but also in the subsurface region of Ba3SiO5−xNyHz. It should be noted that another intense band was observed at approximately 1,417 cm−1, which can be assigned to the N=N bond43. Moreover, the N–H stretching vibration was observed at 3,187–3,244 cm−1 (Supplementary Fig. 37b), which is attributed to the formation of imide species. Similar intermediates were observed for Ru/Ba3SiO5−xNyHz, where N2 peaks above 2,000 cm−1 are immediately consumed to form other nitrogen species, giving negative peaks. Accordingly, NNH species could be generated as intermediates on this catalyst during ammonia synthesis. The N≡N (1,985 cm−1) and N=N (1,439 cm−1) stretching vibration of Ru/Ba3SiO5−xNyHz red-shifted to 1,903 cm−1 and 1,386 cm−1, respectively when 15N2 and H2 flow was conducted (Fig. 4c). The peak shift is reasonably explained by the isotope effect (1,985 cm−1 × (14/15)1/2 = 1,917 cm−1, 1,439 cm−1 × (14/15)1/2 = 1,390 cm−1). The 1,439 cm−1 peak is not due to the H–N–H bending vibration since the peak position is not largely shifted under 14N2 and D2 flow conditions. On the other hand, the NNH bending vibration would appear at around 1,400–1,480 cm−1, overlapping with the broad ν(N=N) band centred at around 1,420 cm−1. The small shift from 1,439 to 1,422 cm−1 is consistent with the isotope effect from the N=N bond in NNH (1,439 cm−1) to NND (1,417 cm−1). From these results, it can be concluded that the N2 molecule is activated at Va sites on Ba3SiO5−xNyHz and sequentially hydrogenated to form NNHx (x = 1–3) species through the associative reaction mechanism.

A DFT calculation study of reaction pathways was performed using Ba48Si16O40N15H32 loaded with Ru12 clusters and the reference Ru/MgO catalyst (Fig. 5 and Supplementary Figs. 38 and 39). First, H2 was dissociated on Ru with a negligible energy barrier (I → II). The dissociated H* was then migrated from the Ru surface to a nearby lattice N of the support (II → III). Further hydrogenation reactions occurred sequentially to form NH3 and a nitrogen vacancy site (VN) site was formed after NH3 desorption (III → VII). The formation energy of an N vacancy (ENV of Ru/Ba3SiO5−xNyHz (−1.21 eV) is much lower than that of bare Ba3SiO5−xNyHz (−0.87 eV), which indicates that the supported Ru nanoparticles facilitate the formation of VN sites (Supplementary Fig. 40). The adsorption energies of N2 on both Ru and VN sites of Ru/Ba3SiO5−xNyHz are calculated to be −0.4 eV (Supplementary Fig. 41), suggesting that N2 is likely to be adsorbed on both sites. The energy barrier for N2 dissociation on the Ru surface was 0.47 eV, which was lower than that of NNH formation (0.79 eV, TS IV). However, CO adsorption experiments confirmed that the Ru surface was barely exposed on Ru/Ba3SiO5−xNyHz (Supplementary Fig. 42). Therefore, the N2 activation at the VN site is more plausible than that on the Ru surface. The N2 molecule adsorbed at the VN site was coordinated to the adjacent Ba, Si and Ru sites, resulting in elongation of the N–N bond (1.31 Å) (Step IX). The hydrogenation of N2 at the VN site to form *NNH (IX → X) occurs in preference to direct dissociation of N2 (Supplementary Fig. 43). *NNH2 and *NNH3 are generated by