Strong metal-support interaction promoted scalable production of thermally stable single-atom catalysts

Single-atom catalysts (SACs) have demonstrated superior catalytic performance in numerous heterogeneous reactions. However, producing thermally stable SACs, especially in a simple and scalable way, remains a formidable challenge. Here, we report the synthesis of Ru SACs from commercial RuO2 powders by physical mixing of sub-micron RuO2 aggregates with a MgAl1.2Fe0.8O4 spinel. Atomically dispersed Ru is confirmed by aberration-corrected scanning transmission electron microscopy and X-ray absorption spectroscopy. Detailed studies reveal that the dispersion process does not arise from a gas atom trapping mechanism, but rather from anti-Ostwald ripening promoted by a strong covalent metal-support interaction. This synthetic strategy is simple and amenable to the large-scale manufacture of thermally stable SACs for industrial applications.

Various strategies have been developed for the fabrication of SACs. Atomic layer deposition and mass-selected soft-landing methods offer precise and controllable synthesis of well-designed SACs [30][31][32] ; however, their scale-up is hindered by high production costs and low catalyst yields 33,34 . Wet chemical routes, such as incipient wetness impregnation (IWI) and strong electrostatic adsorption methods, are common in laboratory-scale catalyst synthesis. However, they are best suited to low metal loadings 1,35,36 and are often time-consuming and process intensive, which is unfavorable for scale-up 18,37 . In addition, the thermal stability of the resulting SACs is typically poor 18,35 . Large-scale synthesis of thermally stable SACs therefore remains problematic.
Atom trapping is an effective method to produce thermally stable SACs [38][39][40] but still relies on wet chemistry to prepare the nanocatalysts as precursors. Based on atom trapping, Wu and Li have developed several approaches, including thermal emitting and solid diffusion, to transform bulk metals into single atoms 18,41,42 and hence open a pathway to scalable SAC production. Unfortunately, these approaches are mainly limited to carbon or N-doped carbon supports and require ammonia or HCl, which present environmental challenges.
Herein, we report a simple route to prepare thermally stable Ru SACs directly from commercial RuO 2 powders by heating of physical mixtures of RuO 2 and strongly interacting supports. Transformation of RuO 2 powders into isolated Ru atoms is promoted by a strong covalent metal-support interaction (CMSI) with MgAl 1.2 Fe 0.8 O 4 . The resulting Ru SAC has excellent thermal stability and improved activity for N 2 O decomposition at low and high concentrations. This simple and low-cost synthesis paves a way for the large-scale production of thermally stable SACs with high metal loadings for industrial applications.

Results
Synthesis and structure of Ru SACs. We recently observed that Pt NPs supported on iron oxides can be dispersed into single atoms upon high-temperature calcination 43 . It transpires that a strong CMSI between Fe and Pt is critical to the dispersion process, which also occurs for Fe-doped (but not undoped) Al 2 O 3 . The chemical similarity of Pt group metals suggests that such interaction may provide a general approach to fabricate thermally stable SACs 29 . Spinels, mixed metal oxides with welldefined structures and excellent thermal stability, are ideal supports for the fabrication of thermally stable catalysts 44,45 . The synthesis of a Ru SAC from a Fe-substituted MgAl 2 O 4 spinel was therefore explored to verify the generality of this strategy.
X-ray diffraction (XRD) patterns showed that MAFO comprised a pure crystalline spinel phase ( Supplementary Fig. 1), indicating that Fe was uniformly incorporated throughout support. The MAFO surface area was far higher than that of commercial Fe 2 O 3 43 (~100 vs. <10 m 2 g −1 , respectively, Supplementary Table 1) offering the prospect of a higher density of anchor sites to immobilize metal atoms. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) revealed small Ru NPs in the uncalcined IWI sample ( Supplementary Fig. 2), which disappeared after 900°C calcination ( Supplementary Fig. 3a-c) implying their dispersion into single atoms 43 . Aberration-corrected (AC) HAADF-STEM images confirmed the formation of uniformly dispersed Ru single atoms ( Supplementary Fig. 3d-f). In contrast, lower-temperature (500°C) calcination resulted in severe sintering of impregnated Ru species into sub-micron RuO 2 aggregates ( Supplementary  Fig. 4), consistent with our observations for Pt sintering over Fe 2 O 3 following low-temperature calcination 43 . Since the Ru/ MAFO-IWI-900 sample transitions through lower temperatures during the heating process, we reasoned that these large RuO 2 aggregates must be thermodynamically unstable and hence should be susceptible to re-dispersion when subject to a further high-temperature calcination. HAADF-STEM confirmed that 900°C calcination of the Ru/MAFO-IWI-500 sample resulted in complete loss of the RuO 2 aggregates ( Supplementary Fig. 5).
The remarkable efficacy of MAFO for dispersing sub-micron Ru aggregates into single atoms at high temperatures inspired us to explore whether commercial RuO 2 powders (rather than costly organometallic complexes) could be used as the metal precursor to synthesize Ru SACs. To maximize the interface between commercial RuO 2 powders (containing sub-micron particles) and MAFO, a physical mixture of the two components was simply ground and calcined at either 900°C for 5 h in air (denoted as Ru 1 /MAFO-900) or 500°C (denoted as Ru/MAFO-500). This synthesis is illustrated in Supplementary Fig. 6; the nominal Ru loadings in both cases were 2 wt%.
The resulting Ru/MAFO-500 sample contained sub-micron RuO 2 aggregates (Fig. 1a-c Table 2). The absence of Ru aggregates in Ru 1 /MAFO-900 must therefore reflect dispersion, not loss, of Ru species; indeed AC-HAADF-STEM evidenced a high density of uniformly dispersed Ru single atoms on the MAFO spinel support (Fig. 1e, f and Supplementary  Fig. 10e-g).
XRD corroborated the preceding observations (Fig. 2a). The untreated physical mixture exhibits reflections characteristic of the rutile structure of RuO 2 and the MAFO support; the former remain visible following 500°C calcination but are completely lost after 900°C consistent with Ru dispersion. MAFO reflections are slightly sharpened by the 900°C calcination, indicating partial support sintering in accordance with the concomitant decrease in Brunauer-Emmett-Teller (BET) surface area (Supplementary Table 1). The chemical state of Ru was investigated by X-ray photoelectron spectroscopy (XPS). Note that the C 1s and Ru 3d photoemissions overlap, and hence Ru 3p XP spectra were measured, revealing identical Ru 3p 3/2 binding energies of 463.2 eV for the Ru 1 /MAFO-900 and Ru/MAFO-500 samples (Fig. 2b), characteristic of Ru 4+ species 46,47 . However, the spectrum intensity for Ru 1 /MAFO-900 is significantly higher than that for Ru/MAFO-500, in good agreement with its much higher dispersion. X-ray absorption spectroscopy was also measured to elucidate the local chemical environment of Ru within both samples (Fig. 2c). The absorption edge energies of Ru 1 /MAFO-900 and Ru/MAFO-500 were identical and matched that for RuO 2 , consistent with the presence of Ru 4+ species observed by XPS; however, the X-ray absorption near-edge structure (XANES) of Ru 1 /MAFO-900 differed from that of Ru/MAFO-500 and RuO 2 , i.e., Ru atoms in Ru 1 /MAFO-900 are in a different coordination environment to those in RuO 2 NPs/aggregates 48,49 . Fourier transforms of the corresponding extended X-ray absorption fine structure (EXAFS) reveal two well-defined coordination shells at~1.97 and 3.54 Å for RuO 2 and Ru/MAFO-500 associated with Ru-O and Ru-O-Ru scattering contributions, respectively (Fig. 2d, Supplementary Fig. 11 43 , unambiguously evidencing Ru single atoms. In addition to a nearest neighbor Ru-O shell, significant Ru-Fe scattering was observed for Ru 1 /MAFO-900 consistent with a strong chemical bonding to FeO x surface sites 43 . We can therefore conclude that high-temperature calcination of a physical mixture of commercial RuO 2 powders and MAFO results in a 2 wt% Ru SAC. Catalysts with higher Ru loadings such as 2.5 and 3 wt% were further prepared with the same procedure. Obvious RuO 2 diffraction peaks were observed for both samples ( Supplementary  Fig. 12), indicating that the maximum Ru loading is in fact around 2 wt%. We estimated the theoretical maximum loading of dispersed Ru atoms over MAFO support by assuming only surface Fe as the stabilization sites to be around 1.6 wt% (for details, see "Methods") 43 , which agree well with the experimental data. The good consistency suggested that the Ru atoms mainly located on the surface/subsurface rather than diffused into the bulk of the support because the latter case will give rise to a much higher maximum Ru loading. The Fe content in MAFO is tunable. We further investigated the effect of Fe content by preparing three MgAl 2−x Fe x O 4 supports with different Fe contents (x = 0.5, 1, 1.5). As shown in Supplementary Table 1, the substitution of Fe weakens the sintering resistance of the MgAl 2 O 4 spinel, thus inducing a surface area decrease after being calcined at high temperatures. Meanwhile, excess Fe substitution would result in the appearance of an impure phase of iron oxide ( Supplementary  Fig. 13a). We then tried to synthesize Ru SACs by using the newly However, weak diffraction peaks of RuO 2 were observed in the 2Ru/MgAl 1.5 Fe 0.5 O 4 -900 sample, suggesting that RuO 2 cannot be completely dispersed on this sample. This likely reflects the low Fe content in the MgAl 1.5 Fe 0.5 O 4 spinel that cannot provide sufficient sites to stabilize all Ru single atoms, consistent with the calculated theoretical maximum Ru loading for MgAl 1.5 Fe 0.5 O 4 support (up to 1.0 wt%; for details, see "Methods"). Based on the above analysis, we propose that for the catalyst with 2 wt% Ru loading the optimized Fe ratio should be around x = 1. For lower Ru loading, the optimized Fe content needs further study; we believe that provided sufficient stabilizing sites are present, the smaller the Fe content the better.
Catalytic performance of Ru/MAFO samples. The catalytic performance of the preceding Ru/MAFO catalysts was subsequently studied for nitrous oxide (N 2 O) decomposition, an important reaction in an environmental context and satellite propulsion systems. N 2 O is a potent greenhouse gas facilitating ozone depletion even at very low concentrations [53][54][55] . However, at high concentrations, N 2 O is a potential "green" propellant in the aerospace sector [56][57][58] . Catalytic decomposition of N 2 O into N 2 and O 2 is therefore a promising route to eliminate (undesirable) low concentrations in the atmosphere and exploit high concentrations as a fuel, and hence both limits (1000 ppm and 20 vol% N 2 O in Ar) were explored in this work (Fig. 3a). The Ru 1 /MAFO-900 SAC exhibited much greater activity than Ru/MAFO-500 at both N 2 O concentrations, reflected in lower light-off temperatures. Ru 1 /MAFO-900 also displayed excellent stability at 550°C for decomposition of low N 2 O concentration, with conversion remaining~76% for 100 h on-stream (Fig. 3b); although Ru/MAFO-500 was also very stable under these conditions, N 2 O conversion was only~25% (a small activity increase at long reaction times may reflect dispersion of small amount of the sub-micron RuO 2 aggregates). XRD ( Supplementary Fig. 15) and HAADF-STEM ( Supplementary Fig. 16) evidenced no Ru NCs or NPs for Ru 1 /MAFO-900 post-reaction, demonstrating  Table 2). Interestingly, decomposition of high N 2 O concentration at elevated temperatures (800°C) over Ru/MAFO-500 resulted in a step change in conversion after only a few minutes on-stream ( Supplementary Fig. 17), which we attribute to dispersion of the initial RuO 2 aggregates; a similar phenomenon was observed for CH 4    Mechanism of RuO 2 dispersion. RuO 2 powders/aggregates can be dispersed into single atoms on MAFO by high-temperature calcination. We believe that substituted Fe plays a critical role in trapping and stabilizing Ru atoms or RuO 2 single clusters through a CMSI effect 43 , a conjecture easily verified by control experiments with an Fe-free spinel (MA). As anticipated, XRD indicated that RuO 2 aggregates are not dispersed into isolated atoms over the MA support by high-temperature calcination (Supplementary Fig. 20) but rather undergo sintering resulting in sharper RuO 2 reflections. AC-HAADF-STEM confirmed that large RuO 2 aggregates were retained in the Ru/MA-900 sample, although a small number of RuO 2 NPs or NCs were also observed (Supplementary Fig. 21). The question arises as to the mechanism of Ru dispersion. Gasphase atom trapping is a common process by which hightemperature dispersion may occur but is usually accompanied by metal losses 18,40 . In the present case, no detectable Ru loss was observed, suggesting the operation of a different mechanism, and confirmed by the following control experiments. Hightemperature calcination of RuO 2 and MAFO was repeated using different locations for the two components ( Supplementary  Fig. 22): RuO 2 powders were placed (a) on the surface of or (b) beneath the MAFO spinel or (c) randomly mixed with the spinel by applying a mechanical vibration. Considering that RuO 2 can oxidize to form volatile RuO 3 and/or RuO 4 at very high temperatures [59][60][61] , if gas-phase atom trapping dominated the dispersion process, then all three geometries should result in efficient Ru dispersion over MAFO since volatilized gas-phase atoms can diffuse to large (in cm level) distances 18,61 . In practice, the RuO 2 powders were unchanged and clearly visible as a separate phase following calcination in scenarios (a) and (b) ( Supplementary Fig. 22), and we can therefore discount a gasphase atom trapping mechanism. This is in accordance with additional control experiments in which RuO 2 powders were calcined without the MAFO support, which resulted in minimal weight loss (<10%) under static or flowing conditions (Supplementary Table 4, entry 1, 2). Note that in scenario (c), although the ochre color of the calcined sample darkened somewhat (a characteristic of Ru 1 /MAFO-900, Supplementary Fig. 23), XRD reflections of RuO 2 remained visible ( Supplementary Fig. 24), and black insoluble substances were observed following dissolution of the MAFO support in aqua regia (Supplementary Fig. 25) consistent with large RuO 2 aggregates. The Ru loading in the vibration mixed Ru/MAFO-VM-900 sample was only 0.72 wt% (Supplementary Table 2), far less than the nominal loading, indicating that only a small amount of RuO 2 aggregates were dispersed over the spinel. We can therefore conclude that intimate physical mixing (PM) of RuO 2 and MAFO prior to their calcination is essential to maximize the resulting dispersion of Ru single atoms.
The control experiment highlighted that RuO 2 volatilization was minimized under an inert environment (Supplementary Table 4, entry 3), and hence RuO 2 dispersion over MAFO was also attempted by annealing at 900°C under Ar and He atmospheres (conditions strongly disfavoring gas-phase atom trapping). In both cases, XRD confirmed the loss of RuO 2 reflections following 5 h anneals (Supplementary Fig. 26) consistent with at least partial Ru dispersion. The resulting Ru/ MAFO loadings of~1.6 wt% (Supplementary Table 2) were slightly lower than the nominal 2 wt% value, suggesting that a small proportion of the parent RuO 2 remained intact, and indeed a subsequent aqua regia treatment of both Ru/MAFO materials revealed trace insoluble component ( Supplementary Fig. 27). Extended annealing under He increased the final Ru loading to 2 wt% (Supplementary Table 2), indicating complete dispersion of this residual RuO 2 into single atoms over the MAFO support. In summary, there is no evidence that Ru volatilization and subsequent gas-phase atom trapping is mainly responsible for RuO 2 dispersion.
The kinetics of RuO 2 dispersion by air calcination was also explored. XRD showed the immediate disappearance of RuO 2 reflections on heating to 900°C (0-h sample, Supplementary  Fig. 28), although HAADF-STEM highlighted trace residual RuO 2 aggregates that required ≥1 h at 900°C to fully disperse into Ru single atoms ( Supplementary Figs. 29 and 30). The dispersion process was directly visualized by in situ AC-HAADF-STEM and simultaneous secondary electron (SE) detection (Fig. 4 Fig. 32).
In situ electron microscopy cannot (yet) directly record the movement of individual atoms over practical catalysts under such conditions; however, the preceding images enable us to exclude certain dispersion processes such as Brownian motion of RuO 2 aggregates throughout the MAFO matrix, distributing Ru atoms/ RuO 2 sub-units as it passes. The only plausible dispersion model is therefore an anti-Ostwald ripening process wherein Ru atoms/ RuO 2 sub-units break away from static RuO 2 aggregates and diffuse across the MAFO surface until being trapped by a CMSI. The rapidity of RuO 2 dispersion over MAFO vs. MA supports at 900°C suggests that CMSI involving FeO x sites may promote such ripening.
To further verify the CMSI between RuO 2 and FeO x , we performed a H 2 temperature-programmed reduction (H 2 -TPR) characterization. As shown in Supplementary Fig. 33, on Ru/MA-500, Ru/MA-900, and Ru/MAFO-500 samples two reduction peaks were observed between 100 and 200°C. The former corresponds to the reduction of RuO 2 to RuO while the latter is ascribed to the reduction of RuO to Ru metal 63,64 . The slightly higher temperature for the reduction of RuO on Ru/MAFO-500 than that on Ru/MA-500 may suggest that Ru species interact stronger with MAFO than with MA. Of more importance, the low-temperature reduction of Ru nearly vanished on the Ru 1 / MAFO-900 sample with only a very tiny reduction peak (marked by arrow). The majority of the Ru species must have been reduced together with Fe at higher temperatures, suggesting a strengthened interaction between RuO 2 and FeO x after being calcined at 900°C. A quantitative analysis (Supplementary Table 5) revealed that the H 2 consumptions on Ru/MA-500, Ru/MA-900, and Ru/ MAFO-500 samples are similar to the theoretical one for complete reduction of RuO 2 to Ru. However, for Ru 1 /MAFO-900 sample, the H 2 consumption of the tiny reduction peak is only about 1/27 of the theoretical one, corresponding to a reduction of Ru loading of~0.07 wt%. We propose that these Ru species may be stabilized by Mg or Al sites since the MA support itself can stabilize very low loading of Ru single atoms 65 .
A recent theoretical study proposed that strong metal atom-support interactions can decrease the activation energy (and hence promote the occurrence) of Ostwald ripening 66 , in good agreement with our experimental observations. The possibility that a CMSI promotes RuO 2 dispersion in our system was investigated by density functional theory (DFT) calculations (for details, see "Methods"). The small RuO 2 clusters (Ru 5 O 10 or Ru 10 O 20 ) supported on MgAl 2 O 4 (100) and two-layer Fe-substituted MgAl 2 O 4 (100), respectively, were studied for comparison. Geometry optimization revealed that either the longest or the average Ru-Ru distance in RuO 2 clusters supported on Fesubstituted MgAl 2 O 4 (100) were significantly elongated, compared to those on MgAl 2 O 4 (100) surface ( Supplementary Fig. 34 and Supplementary Table 6). In particular, one RuO 2 moiety evidently moves away from the RuO 2 cluster on Fe-substituted MgAl 2 O 4 (100). Further calculations on the binding energy and reaction Gibbs free energy showed that the farthest RuO 2 moiety dissociation from the RuO 2 cluster supported on Fe-substituted MgAl 2 O 4 (100) surface is preferred over on MgAl 2 O 4 (100) surface (Supplementary Table 7), which comes from the Fe effect on the metal-support interaction, as confirmed by electron density difference maps in Supplementary Fig. 35. These results revealed that the presence of Fe atom weakens Ru-Ru interaction in cluster and promotes RuO 2 dispersion. We therefore propose that a strong CMSI between FeO x sites in the MAFO support and RuO 2 aggregates promotes anti-Ostwald ripening of Ru atoms/ RuO 2 sub-units.
Scalable production of SACs. Scale-up represents a key barrier to progressing SACs from intellectual curiosity to practical solution for industrial chemical processes. The utility of our simple mixing/calcination protocol was therefore exploited to prepare 10 g Ru 1 /MAFO (Ru 1 /MAFO-10g-900, Supplementary Fig. 36 Fig. 42) due probably to the lower redox activity and/ or the significantly lower surface area of Fe 2 O 3 , the ability to prepare 1 kg of SAC by mixing and heating two commercial bulk oxides may have a profound influence on the future direction of catalyst manufacturing.

Discussion
We have developed a simple strategy to prepare Ru SACs by PM of commercially available RuO 2 powders with Fe-containing supports. RuO 2 powders undergo complete dispersion into isolated single atoms following high-temperature treatment under oxidizing and inert atmospheres. A strong metal-support interaction between Ru and Fe plays a critical role not only in trapping and stabilizing Ru atoms but also in promoting the ripening of RuO 2 aggregates. The approach is simple, general, environmentally friendly, and highly scalable, unlocking the large-scale manufacture of thermally stable SACs for industrial applications.
Preparation of MA spinel. MgAl 2 O 4 spinel (designated as MA) was prepared by hydrolysis of aluminum isopropoxide and magnesium acetate tetrahydrate in ethanol. In all, 0.15 molar of magnesium acetate tetrahydrate and 0.30 molar of aluminum isopropoxide were mixed in 900 mL of ethanol and sealed in a 2-L autoclave. The mixture was heated to 120°C and held there for 10 h, then increased to 160°C and held there for another 10 h under vigorous stirring. After cooling to room temperature, the obtained product was filtrated and then dried at 120°C for 1 h, and finally calcined in ambient air at 700°C for 5 h with a heating rate of 2°C min −1 , resulting in the formation of MA spinel with pure spinel crystal phase.
Preparation of MAFO spinels. MgAl 1.2 Fe 0.8 O 4 spinel (designated as MAFO) was prepared by hydrolysis of aluminum isopropoxide and iron(III) acetylacetonate with magnesium nitrate hexahydrate in ethanol. In all, 0.15 molar of magnesium nitrate hexahydrate, 0.18 molar of aluminum isopropoxide, and 0.12 molar of iron (III) acetylacetonate were mixed in 900 mL of ethanol and sealed in a 2-L autoclave. The mixture was heated to 120°C and held there for 10 h, then increased to 160°C and held there for another 10 h under vigorous stirring. After cooling to room temperature, the obtained product was filtrated and then dried at 120°C for 1 h, and finally calcined in ambient air at 700°C for 5 h with a heating rate of 2°C min −1 , resulting in the formation of MAFO spinel with pure spinel crystal phase. MgAl 1.5 Fe 0.5 O 4 , MgAl 1 Fe 1 O 4 , and MgAl 0.5 Fe 1.5 O 4 spinels were prepared via adjusting the ratio of aluminum isopropoxide and iron(III) acetylacetonate and used the same preparation procedure of MAFO spinel.
Preparation of Ru/MAFO-IWI samples. The Ru/MAFO-IWI samples (nominal weight loadings of Ru were 1 wt%) were prepared using the IWI method. The sample was synthesized using a solution of ruthenium(III) acetylacetonate in toluene. After impregnation, the sample was dried at room temperature for 24 h and 60°C for 10 h. Then the sample was calcined in ambient air at 500/900°C for 5 h with a heating rate of 2°C min Large-scale preparation of Ru 1 /Fe 2 O 3 SAC. In all, 1000 g of Fe 2 O 3 was physically mixed with 3.9620 g of RuO 2 and calcined in ambient air at 900°C for 5 h with a heating rate of 2°C min −1 . The nominal weight loading of Ru was 0.3 wt%. The resulting sample is designated as Ru 1 /Fe 2 O 3 -1000g-900.
Three control experiments with different contact manners. RuO 2 powders were located on the surface of or underneath the MAFO spinel support or randomly mixed by vibration. The comparative experiments were performed and calcined at 900°C for 5 h with a heating rate of 2°C min −1 . Vibration mixing was carried out using a vibrating plate, and the calcined sample is designated as Ru/MAFO-VM-900. The nominal weight loadings of Ru were 2 wt%.
Catalyst characterization. HAADF-STEM images were obtained on a JEOL JEM-2100F operated at 200 kV. AC-HAADF-STEM images were obtained on a FEI Titan Cubed Themis G2 300 operated at 200 kV. TEM specimens were prepared by depositing a suspension of the powdered sample on a lacey carbon-coated copper grid.
The in situ AC-HAADF-STEM/SEM experiment was performed on a Hitachi field emission scanning transmission microscope HF5000 using the MEMS heating holder, and the gas flow was controlled by MFC system. The MEMS heating holder was manufactured by Hitachi High Technologies Canada. And the chips were manufactured by Norcada Inc. The Ru/MAFO-UC sample was supported on the 50-nm-thick Si 3 N 4 membrane. And the gas was injected to the sample area by special designed gas injection nozzle. The oxygen purity used for the in situ calcination experiment was 99.999%.
XRD patterns were recorded on a PANalytical Empyrean diffractometer equipped with a Cu Kα radiation source (λ = 0.15432 nm), operating at 40 kV and 40 mA.
The BET surface area, pore volume, and average pore size were measured with a Micromeritics ASAP 2460 instrument using adsorption of N 2 at 77 K. All of the samples were degassed under vacuum at 300°C for 5 h before the adsorption measurements.
Inductively coupled plasma optical emission spectrometry was performed on an Optima 7300DV instrument (PerkinElmer Instrument Corporation). All the samples were dissolved by using aqua regia heated on a hotplate until it was clear or continuously heated for 2 h. X-ray fluorescence spectrometry (XRF) was performed on a PANalytical Zetium instrument. The samples were pressed into tablets before XRF analyses. In order to obtain an accurate Ru content, we prepared a calibration curve: briefly, 2.5 g of MAFO spinel was physically mixed with corresponding proportion of RuO 2 by using an agate mortar (for details, see Supplementary Fig. 43 and Supplementary Table 8).
XPS was measured on a Thermo Fisher ESCALAB 250Xi spectrometer equipped with an Al anode (Al Kα = 1486.6 eV), operated at 15 kV and 10.8 mA. The background pressure in the analysis chamber was <3 × 10 −8 Pa, and the operating pressure was around 7.1 × 10 −5 Pa. The survey and spectra were acquired at a pass energy of 20 eV. Energy calibration was carried out using the C 1s peak of adventitious C at 284.8 eV.
XANES and EXAFS spectra at the Ru K-edge were recorded at the BL14W1, Shanghai Synchrotron Radiation Facility, China. A Si (311) double-crystal monochromator was used for the energy selection. The energy was calibrated by Ru foil. Ru foil and RuO 2 were used as reference samples and measured in the transmission mode. The Ru/MAFO-500, Ru 1 /MAFO-900, and Ru 1 /MAFO-10g-900 samples were measured in the transmission mode. The Athena software package was used to analyze the data. H 2 -TPR was carried out on a Micromeritics AutoChem II 2920 apparatus. The sample (~100 mg) was placed in the U-shaped quartz reactor and heated at 300°C in Ar for 30 min to remove the physically adsorbed water and other contaminants. After cooling the sample down to 50°C, the gas was switched to 10 vol% H 2 /Ar, and the sample was heated to 900°C at a ramp rate of 10°C min −1 for reduction. H 2 consumption during sample reduction was monitored via TCD. The amount of H 2 consumption was calculated with the H 2 peak area and calibration curve of the 10 vol% H 2 /Ar standard gas.
Catalytic reactions. N 2 O decomposition was carried out at atmospheric pressure in a fixed-bed microreactor. In all, 100 mg of catalyst diluted with 1 g of quartz sand (40-80 mesh) was loaded into a U-shaped quartz reactor. A k-type thermocouple in a thin quartz tube was inserted into the catalyst bed to measure the temperature. The feed gas containing 1000 ppm N 2 O and balance Ar (low concentration) or 20 vol% N 2 O and balance Ar (high concentration) was passed through the reactor at 33.3 mL min −1 . Long-term stability was tested by running the reactor at 550°C for 100 h at low N 2 O concentration. The test of the dispersion of RuO 2 was using 50 mg of the Ru/MAFO-500 catalyst diluted with 1 g of quartz sand (40-80 mesh) and performed at high N 2 O concentration with a high gas flow (166.7 mL min −1 ). The reaction temperature increased from room temperature to 800°C with a rate of 10°C min −1 and then maintained at 800°C for 10 h. The amounts of the N 2 O in the inlet and outlet gas compositions were analyzed using a gas chromatograph (Echrom A91) equipped with Parapak Q packed column and a thermal conductivity detector using He as the carrier gas. For the Ru 1 /Fe 2 O 3 -1000g-900 catalyst, 670 mg of catalyst diluted with 1 g of quartz sand (40-80 mesh) was loaded into a U-shaped quartz reactor in low-concentration N 2 O decomposition reaction under the premise of using same Ru amount.
Computational methods. All DFT calculations were performed with Vienna Abinitio Simulation Package (VASP) 67,68 , and the exchange-correlation energy was expressed by generalized gradient approximation of Perdew-Burke-Ernzerhof functional 69 . The projector-augmented wave method 70 was used to describe the interaction between electrons and ions. The plane-wave basis energy cutoff was set to 520 eV with the gamma point only for the Brillouin zone. The convergence criteria for the electronic structure and geometry optimization were 1 × 10 −4 eV and 0.02 eV Å −1 , respectively. Because of the strongly correlated d electrons, DFT+U calculations with corresponding U-J values of 2.5 eV (Fe) and 2.0 eV (Ru) were employed 71,72 .
Computational models. The 2 × 2 supercell model of MgAl 2 O 4 (100) 73 consists of four Al-O layers and three Mg layers, of which bottom two layers were fixed in the relaxation calculations. A 15 Å vacuum layer was added to avoid interaction between periodic structures. To model the MAFO, Al in top layers of MgAl 2 O 4 (100) were partly replaced by Fe. Ru 5 O 10 and Ru 10 O 20 clusters that were cut from the RuO 2 crystal were employed as RuO 2 cluster models.
Theoretical maximum loading of dispersed Ru atoms over spinel. The BET surface area of Ru 1 /MAFO-900 was 38 m 2 g −1 , hence 1 g of MAFO support provides 38 m 2 of surface (S) after 900°C calcination. The spinels mainly have primary cuboctahedral shape with dominant {100} and {111} facets 44 . Assuming that all M 3+ on the surface can stabilize Ru atoms, the theoretical model indicates that the maximum density of atomically dispersed Ru (D) are 5.88 and 6.79 atom nm −2 for {100} and {111} facets, respectively. The total number of isolated Ru atoms (N) that could be achieved for 1 g of Ru/MAFO is therefore predicted to be N = D × S. Since NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-14984-9 ARTICLE NATURE COMMUNICATIONS | (2020) 11:1263 | https://doi.org/10.1038/s41467-020-14984-9 | www.nature.com/naturecommunications the mass of Ru equals (N/N A ) × M, where N A is Avogadro's constant (6.02 × 10 23 mol −1 ), and M is the molar mass of Ru (101 g mol −1 ), the theoretical maximum loadings of isolated Ru atoms that could be dispersed over 1 g of MAFO are 3.7 and 4.3 wt% for {100} and {111} facets, respectively. Thus the calculated maximum Ru loading is about 4 wt% assuming that all M 3+ sites can stabilize Ru atoms. However, if only Fe 3+ can stabilize Ru, the maximum Ru loading should be 4 wt% × 0.8/ 2 = 1.6 wt% for MgAl 1.2 Fe 0.8 O 4 support. Similarly, the maximum Ru loading should be 4 wt% × 0.5/2 = 1.0 wt% for MgAl 1.5 Fe 0.5 O 4 support.

Data availability
The data that support the findings of this study are available within the paper and its Supplementary Information, and all data are available from the authors on reasonable request.