Non defect-stabilized thermally stable single-atom catalyst

Surface-supported isolated atoms in single-atom catalysts (SACs) are usually stabilized by diverse defects. The fabrication of high-metal-loading and thermally stable SACs remains a formidable challenge due to the difficulty of creating high densities of underpinning stable defects. Here we report that isolated Pt atoms can be stabilized through a strong covalent metal-support interaction (CMSI) that is not associated with support defects, yielding a high-loading and thermally stable SAC by trapping either the already deposited Pt atoms or the PtO2 units vaporized from nanoparticles during high-temperature calcination. Experimental and computational modeling studies reveal that iron oxide reducibility is crucial to anchor isolated Pt atoms. The resulting high concentrations of single atoms enable specific activities far exceeding those of conventional nanoparticle catalysts. This non defect-stabilization strategy can be extended to non-reducible supports by simply doping with iron oxide, thus paving a new way for constructing high-loading SACs for diverse industrially important catalytic reactions. Developing stable single-atom catalysts (SACs) with a high metal loading remains a challenge due to the difficulty of creating high densities of defects on support materials. Here the authors prepare Pt SACs with high Pt loadings by virtue of strong covalent metal-support interaction, rather than support defects.

H eterogeneous catalysis is pivotal to the modern chemical industry 1 , with many heterogeneous catalysts comprising transition metals deposited over a solid support phase 2 . A recent evolution in this field is the dispersion of isolated, individual metal atoms over the support to form single-atom catalysts (SACs) 3 . Such SACs have attracted considerable attention as a new frontier in heterogeneous catalysis 4,5 , affording enhanced precious metal thrifting 6 and atom economy [7][8][9] , improved active site homogeneity [10][11][12] , and the ability to tune the metal-support interface with unprecedented control [13][14][15][16] , resulting in superior catalytic activity [17][18][19][20] and/or high selectivity [21][22][23] . A diverse range of supports are known to stabilize isolated metal atoms, notably through pinning at electronic and/or structural defects associated with coordinatively unsaturated sites (CUS) by experimental and density functional theory (DFT) studies (Supplementary Appendices I and II) [24][25][26][27][28][29][30][31][32][33] . The concentration of atomically-dispersed metal atoms, and their stability, are therefore intrinsically linked to the density and stability of defects as described in recent reviews (Supplementary Appendix III) 34,35 . However, it is extremely difficult to create a high density of thermally stable support defects 36 , and hence the fabrication of high metal loading SACs remains currently a formidable challenge.
Here we report the discovery that single metal atoms can be stabilized through a strong covalent metal-support interaction (CMSI) that is not associated with surface defects, enabling the genesis of high concentrations of thermally stable single atoms, even on low-surface-area materials. In situ genesis of ferric oxide supported Pt SAC from Pt nanoparticles (NPs) is verified in methane combustion reaction, one of the primary means of human energy production and important for mitigating environmental challenge associated with CH 4 emission. Experimental and DFT studies demonstrate that metal oxide reducibility dictates the ability of a support to anchor isolated Pt atoms: α-Fe 2 O 3 favors atomically dispersed Pt, whereas Al 2 O 3 favors NP sintering. However, the non defect-stabilization strategy can be extended to non-reducible oxide by simply doping with iron oxide. This finding provides a promising method to fabricate high-metal-loading SACs with excellent thermal stability.

Results
Co-precipitation prepared thermally stable Pt 1 /FeO x . A 1.8 wt% Pt/FeO x catalyst was first prepared by co-precipitation (the synthesis details are in the Methods section), denoted as Pt 1 / FeO x , and subsequently calcined at 800°C in air for 5 h, denoted as Pt 1 /FeO x -C800. Aberration-corrected high-angle annular darkfield scanning transmission electron microscopy (AC-HAADF-STEM) with sub-angstrom resolution was used to compare the nature of platinum species between these two materials. In neither case were Pt nanoclusters or NPs observed ( Supplementary  Figures 1 and 2). Pt 1 /FeO x comprised isolated Pt atoms dispersed on the FeO x support with some Pt atoms exactly aligned with the Fe atomic columns (Fig. 1a-b). Corresponding CO diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) of Pt 1 /FeO x (Supplementary Figure 1c) shows a coverageindependent band at~2089 cm −1 and the absence of bridged adsorption of CO band (~1860 cm −1 ), which is consistent with the CO linearly bound to isolated Pt atoms 3 . Note that our CO infrared bands are broader than those recently reported by Christopher and co-workers 37 . However, this is unsurprising because in their work ultra-dilute (0.05 wt%) Pt loadings were used to immobilize only one Pt atom per 5 nm monodispersed anatase nanocrystal. Such low loadings favor the population of only the most reactive titania surface sites, and hence a homogeneous local environment. Platinum sintering was not observed following 800°C calcination by either AC-HAADF-STEM  Figure 2c) analysis of the Pt 1 /FeO x -C800 sample. X-ray absorption spectroscopy (XAS) studies provide further evidence for the complete dispersion of Pt atoms. Figure 1e presents the Fourier transform radial distribution functions of k 3 -weight extended X-ray absorption fine structure (EXAFS) spectra for Pt 1 /FeO x before and after calcination. Similar to our previous reports 3 , no Pt-Pt bond contribution was found. This observation is in marked contrast to the significant sintering typically noted for small metal species/atoms sintered to calcination at elevated temperature 38,39 , and indicates a very strong interaction between Pt atoms and Fe 2 O 3 that inhibits migration and sintering of the former. X-ray absorption near edge structure (XANES) measurements confirmed that the chemical state of Pt atoms before and after calcination resembled that PtO 2 with tetravalent Pt(IV) (Supplementary Figure 2d) rather than metallic platinum with zerovalent Pt(0) 3 .
Pt NP-generated high loading Pt SACs on Fe 2 O 3 support. Pt NPs are known to liberate mobile PtO 2 species at elevated temperature under an oxidizing atmosphere 40 , which may be subsequently trapped by either larger Pt NPs (Ostwald ripening) or unique oxide sites or crystalline facets resulting in metal dispersion [41][42][43][44][45] . In the present case, wherein a strong interaction between Pt atoms and the Fe 2 O 3 support is apparent, we anticipate that Pt atoms vaporized from a NP could be trapped and stabilized by the Fe 2 O 3 surface. To test this hypothesis, preformed colloidal Pt NPs were supported on Fe 2 O 3 at a 0.3 wt% loading and the ethylene glycol stabilizer was removed by a 500°C calcination (denoted as 0.3Pt/Fe 2 O 3 -NP), prior to various high-temperature calcination treatments. HAADF-STEM imaging revealed the presence of 2-3 nm diameter Pt NPs (Supplementary Figure 3a  Note that some Pt NPs remained visible by electron microscopy after a lower temperature (600°C) calcination, albeit more loosely coordinated to the underlying support (Supplementary Figure 4c). An oxidizing atmosphere was critical to observe this high-temperature dispersion; annealing 0.3Pt/Fe 2 O 3 -NP under Ar at 800°C promoted Pt aggregation into larger 3-5 nm particles (Supplementary Figure 4d). This observation strongly implicates oxidized Pt species in Pt NP dispersion (Fig. 3), and indeed X-ray photoelectron spectra (XPS) (Supplementary Figure 6c) provide direct evidence for high-valent (near + IV) Pt in the 0.3Pt/Fe 2 O 3 -C800 catalyst. This is quite different from those divalent (+II) Pt ions dominating on CeO 2 support after high-temperature calciantion 46,47 , and hence further evidences a stronger Pt-Fe 2 O 3 interaction that partially arises from covalent bonding interaction 48,49 . Three requirements emerge for the successful conversion of Pt NPs into atomically dispersed Pt: a high temperature to promote Pt mobility; the presence of molecular O 2 to partially oxidize the surface of Pt NPs, thereby enabling vaporization of (mobile) PtO 2 ; and a strong interaction between support and surface Pt atoms. Consequently, highly stable static single atoms on heterogeneous surface can be fabricated 50 .
The first condition is met for H 2 PtCl 6 functionalized α-Al 2 O 3 (Supplementary Figure 7a between alumina and Pt species. We note that dispersion of < 1 nm Pt NPs was recently reported over MCM-22 51 , which may reflect restricted migration of Pt species through the microporous zeolite network. Having identified conditions under which Pt NPs can be dispersed over an Fe 2 O 3 surface, we explored the effect of Pt loading with a view to maximizing the density of isolated atoms achievable, and thereby compensate for the extremely low surface area (only 5-10 m 2 g −1 ) of the Fe 2 O 3 support after the requisite 800°C calcination. A higher 1 wt% loading of the same 2-3 nm Pt NPs (denoted as 1Pt/Fe 2 O 3 -NP) were also fully transformed into isolated atoms by calcination at 800°C (Supplementary Figure 8a-b). Low energy ion scattering (LEIS) evidenced a clear decrease in the Pt:Fe atomic ratio in the outermost surface layer upon calcining the stabilizer free 1Pt/Fe 2 O 3 -NP to 800°C (Supplementary Figure 9). Since the total Pt loading determined by inductively coupled plasma spectrometry-atomic emission spectrometry (ICP-AES) remained unchanged, it appears that some Pt atoms diffuses into the near sub-surface region of the Fe 2 O 3 support, possibly by a cation-exchange process 52,53 . However, a further increase in Pt loading to 2 wt% resulted in the formation of Pt clusters alongside isolated atoms (Supplementary Figure 8c-d), in line with the theoretical maximum loading (~1.5 wt% for Fe 2 O 3 possessing a surface area of 10 m 2 g −1 , see Methods). A high density of clusters and few single isolated atoms were observed for a 4.5 wt% Pt loading (Supplementary Figure 8e-f). The maximum concentration of atomically-dispersed Pt loading is clearly dictated by the very low support surface area. The emergence of Pt clusters for loadings ≥2 wt% confirms that any Pt atoms diffusing into the support must be confined to the immediate subsurface region and not migrated into the bulk Fe 2 O 3 lattice, since the latter pathway would enable isolated atoms to remain the dominant surface species at far higher Pt loadings.
Disintegration of Pt NPs during high-temperature calcination was directly visualized by in situ HAADF-STEM, and one Fe 2 O 3 crystal of 1Pt/Fe 2 O 3 -NP sample was randomly picked to observe the process. Prior to calcination, about 300 Pt particles were visible across the Fe 2 O 3 support, and the average particle size was about 3 nm ( Fig. 4a-c). However, after heating to 800°C under a flow of 1 bar O 2 for 20 min, the total number of Pt NPs decreased to~200, since many of the 2-3 nm Pt NPs shrank and/or disappear entirely, while a few larger particles (size above 10 nm) remained ( Fig. 4b-c versus Fig. 4e-f). The loss of small clusters was confirmed by higher resolution, time-resolved imaging ( Fig. 4g-i taken from the Supplementary Movie 1), which reveal smaller Pt clusters (e.g., particle 1) disintegrated in less than 35 s, while larger Pt clusters (e.g., particle 2) vanished in 45 s. Previous reports suggest that smaller metal crystallites disperse as adatoms, which migrate towards neighboring (larger) particles and eventually coalesce at elevated temperature via Ostwald Ripening, driven by lowering of the total surface energy 54 . In this case, however, particle disappearance occurs in the absence of such coalescence, consistent with the accompanying genesis of atomically-dispersed Pt entities.
DFT Studies. The preceding electron microscopy, X-ray spectroscopy and diffraction, and CO chemisorption measurements provide both clear evidence of, and insight into the driving force for, the formation of atomically-dispersed Pt following high-  Figure 4d) can reach as high as 4.00 eV, and hence is energetically strongly disfavoured due to CMSI of Pt 1 to the surface atoms. By comparison, PtO 2 evaporation at 600°C is virtually thermo-neutral or slightly endothermic (0.07 eV), indicating the critical role of temperature in dispersing Pt NPs. This result is consistent well with the experimental results that the dispersion of NPs only occurred with calcination >600°C (Fig. 3, Supplementary Figures 4a-c). This temperature-dependent evolution of PtO 2 dissociation energetics from a Pt(221) step (representative of the partially oxidized surface of Pt NPs following high-temperature calcination) mirrors that of the catalytic activity for methane oxidation, as discussed below. The dissociative adsorption of PtO 2 on an oxygen terminated Fe 2 O 3 (0001) surface (and concomitant desorption of one O 2 molecule) is highly exothermic at 800°C (calculated as −2.46 eV, Fig. 5c-d), with PtO 2 dissociative capture even more favorable at lower temperature. Note that the evaporation, migration, and trapping do not require the presence of anion or cation vacancies on the support: no Lewis acid sites were detected on the Fe 2 O 3 support by NH 3 temperatureprogrammed desorption, suggesting that Fe 2 O 3 possessed negligible oxygen anion defects following a high-temperature calcination. Therefore, Pt atoms disperse across the ferric oxide support where they are anchored as isolated atoms through a strong local surface interaction. Calculations indicate that these reactivelyformed Pt atoms coordinate with four surface oxygen atoms in a distorted square geometry with an average Pt-O length of 1.94 Å, intermediate between that in gas PtO 2 (CN = 2, 1.70 Å) and bulk PtO 2 (CN = 6, 2.04 Å). Bader charge analysis of an isolated Pt   Table 2); both findings are attributed to the irreducibility of Al 3+ , i.e. strong non-redox Al-O bonding network, which prevents any significant metal-support interaction. DFT calculations are hence in excellent agreement with experimental observations. The finding of the stability of SAC closely related to the reducibility of the support is consistent with previous work as well 55 .
Catalytic performance in methane combustion. The impact of oxygen-induced restructuring at elevated temperature on catalytic performance is striking (Fig. 6). As the reaction temperature of 1Pt/Fe 2 O 3 -NP under an oxygen-rich methane gas feed was raised from 300 to 700°C, the combustion activity increased such that a modest~18 % conversion was initially attained.   accompanied by the disappearance of Pt NPs present in the asprepared material (Fig. 6 left inset), and formation of atomicallydispersed Pt (Fig. 6 right inset, Supplementary Figure 14a-b), demonstrating a causal relationship between in situ genesis of a Pt SAC and dramatically enhanced performance. Note that preactivation of 1Pt/Fe 2 O 3 -NP by calcination at 700°C for 5 h prior to methane addition resulted in an extremely stable Pt SAC that maintained similar activity for > 16 h on-stream (Supplementary Figure 14c). Since the mass of 1Pt/Fe 2 O 3 -NP catalyst (and Pt loading) in this methane light-off experiment is constant, the time-dependent increase in CH 4 conversion at 700°C due to catalyst restructuring is directly proportional to the catalytic activity; single-atom formation induces a four-fold enhancement (18→65 % conversion) in the specific activity. Additional experiments with the same catalyst at 20 % iso-conversion in either its SAC (1Pt/Fe 2 O 3 -C700) or nanoparticulate (1Pt/Fe 2 O 3 -NP) form confirmed that the specific activity of the former was 4 times greater than that of the latter (2.01 mol CH4 h −1 g Pt −1 vs. 0.47 mol CH4 h −1 g Pt −1 ) and 20 times than that reported for Pt/ Al 2 O 3 56 (Supplementary Table 4). Turnover frequencies based on the number of surface Pt atoms (dispersion) were similar for SAC and nanoparticulate counterparts (0.1 s −1 vs. 0.08 s −1 ), indicating a common active site, and hence the superior specific activity of the single-atom catalyst reflects its improved atom efficiency (every Pt atom directly activates methane).

Interaction between Pt atoms and iron-modified support.
Fundamental insight into the reaction-induced restructuring of metal NPs over a reducible metal oxide support offers a facile route to synthesize high-loading and thermally stable SACs. Platinum NPs introduced either as pre-formed colloids or e.g. by simple wet impregnation with 1 wt% H 2 PtCl 6 (Supplementary Figure 15a-b) over a low area Fe 2 O 3 support, are readily transformed into isolated Pt atoms by high-temperature calcination ( Supplementary Figure 15c-d). Flytzani-Stephanopoulos and coworkers have shown that Na or K ions can stabilize surface hydroxyls, which can in turn stabilize metal (Au and Pt) atoms and/or subnanometer clusters over diverse supports [57][58][59] . However, our recent work 60 has not identified such a role for alkalis, suggesting that Fe 2 O 3 may have unique properties to some degree. To test our hypothesis, we synthesized a sodium-free material using a (NH 4 ) 2 CO 3 precipitant, denoted as Fe 2 O 3 (N). Pt NPs exhibited the same dispersion and single-atom formation after 800°C calcination over this Fe 2 O 3 (N) support (Supplementary Figure 16), confirming that the reducible Fe 2 O 3 support, and not presence of Na + , was responsible for stabilizing Pt singleatoms.
The propensity for stabilizing single Pt atoms is dictated by the magnitude of the metal-support interaction, which itself can be tuned by e.g. doping a reducible (Fe 2 O 3 ) into a non-reducible (Al 2 O 3 ) oxide (Supplementary Figure 17a) through coprecipitation. As shown in Fig. 7, functionalization of such an Fe 2 O 3 -Al 2 O 3 mixed metal oxide by Pt NPs, and subsequent 800°C calcination, delivers atomically-dispersed Pt precisely as observed over the pure ferric oxide (Supplementary Figure 17b-c), but in a higher area form (~30 m 2 g −1 ).

Discussion
We reported a facile synthesis of thermally-stable Pt SAC that achieved remarkable catalytic performance towards methane combustion. Isolated Pt atoms with high oxidation state are not stabilized by surface defects but through a strong covalent interaction with iron and oxygen atoms on the surface. Since Fe can be simply incorporated into a variety of oxides with low area and/or defect density, e.g. perovskites and spinels, our approach may afford a generic route to fabricate high loading Pt SACs over diverse supports being able to operate under harsh reaction conditions.

Methods
Synthesis of FeO x and Fe 2 O 3 . Ferric nitrate (Fe(NO 3 ) 3 ·9H 2 O), 98%, Sigma-Aldrich) and sodium carbonate (Na 2 CO 3 , 99.5% Sigma-Aldrich) were used as purchased. An aqueous solution of Fe(NO 3 ) 3 (1 mol L −1 ) was obtained by dissolving Fe(NO 3 ) 3 ·9H 2 O (122 g, 0.3 mol) in deionized water (300 mL). FeO x was subsequently prepared from this solution by precipitation. First, Na 2 CO 3 (11 g) was added in water (100 mL) and stabilized at 50°C in a water bath. To this, 40 mL Fe (NO 3 ) 3 solution was slowly added (1 mL min −1 ) under vigorous stirring at 50°C for 3 h, and then aged static for a further 2 h. The recovered FeO x solid was washed with deionized water and dried at 60°C overnight. Fe 2 O 3 was obtained by calcining FeO x at 800°C (ramp rate 3°C min −1 ) for 5 h under flowing air (100 mL min −1 ).
Synthesis of Pt 1 /FeO x and Pt 1 /FeO x -C800. Pt 1 /FeO x was prepared by coprecipitation of an aqueous mixture of chloroplatinic acid (H 2 PtCl 6 ·6H 2 O, 37 mg Pt mL −1 , 2.6 mL, 99.9%, Sigma-Aldrich) and Fe(NO 3 ) 3 ·9H 2 O (1 mol L −1 , 40 mL) with Na 2 CO 3 solution (11 g Na 2 CO 3 in 100 mL H 2 O) at 50°C under stirring for 3 h, and ageing static for a further 2 h. The resulting solid was recovered by filtration, washed with deionized water and dried at 60°C overnight. A portion of the Pt 1 /FeO x was then calcined as above at 800°C and denoted as Pt 1 /FeO x -C800. The Pt loading determined by ICP was 1.8 wt%.  Characterization. Pt loadings were determined by inductively coupled plasma spectrometry-atomic emission spectrometry (ICP-AES) on an IRIS Intrepid II XSP instrument (Thermo Electron Corporation). X-ray diffraction (XRD) patterns were collected on a PW3040/60 X'Pert Pro super (PANalytical) diffractometer, operating at 40 kV and 40 mA using a Cu Kα radiation source (λ = 0.15432 nm) with a scanning angle (2θ) of 10-80°. In situ diffuse reflectance infrared Fourier transform spectra (DRIFTS) were collected at 25°C with a Bruker Vertex 70 spectrometer equipped with a mercury cadmium telluride (MCT) detector at a resolution of 4 cm −1 over 32 scans. Samples were first heated in situ at 200°C under flowing He for 0.5 h, then cooled to room temperature prior to recording of background spectra. Subsequently, a 1 vol% CO/He flow was introduced to the sample for 10 min. The environmental cell was then purged with flowing helium to remove gas phase CO, prior to spectral acquisition of chemisorbed CO species. Temperatureprogrammed desorption (TPD) was conducted on a Micromeritics AutoChem II 2910 automatic catalyst characterization system equipped with a mass spectrometer. First, the Fe 2 O 3 support was loaded into a U-shape quartz reactor and pretreated at 200°C in He for 0.5 h to remove adsorbed hydrates and carbonates. After cooling to 60°C, NH 3 gas pulses were injected until a stable signal was obtained, and the sample then heated to 900°C under He at 10°C min −1 . Synchrotron radiation experiments were performed at BL14B2 of SPring-8 62 and BL14W1 of the SSRF light sources. Pt L III -edge transmission X-ray absorption spectra were acquired on samples loaded into the Al 2 O 3 tube, in addition to Pt foil and PtO 2 powder references. Spectra were background subtracted, normalized, and fitted using the IFEFFIT software suite 63 . X-ray photoelectron spectra (XPS) were performed on a VG ESCALAB MK2 apparatus using Al K α radiation (1486.6 eV, 12.5 kV, 250 W) to obtain the binding energies and oxidation states of Pt. Binding energies are referenced to adventitious carbon at 284.8 eV. High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) images were obtained on a JEOL JEM-2100F. Aberration-corrected HAADF-STEM images were obtained on a JEOL JEM-ARM200F equipped with a CEOS probe corrector. Samples were dispersed by ultrasonication in ethanol, and the resulting solution dropped on to carbon films supported on copper grids. In situ imaging was performed on a Titan Cubed Themis G2 300 microscope using a DENSsolutions Climate S3 and Gas flow and Heating System, which comprised a heated sample holder and gas delivery manifold. The sample was mounted in a 5 μm gap between two 30 nm thick SiN windows. The oxygen purity used for the in situ calcination experiment was 99.999%. Reported temperatures were based on the DENSsolutions calibration. Low energy ion scattering (LEIS) spectra were acquired on a Kratos AXIS Supra spectrometer using 1 keV He + ions generated from a Minibeam 6 Gas Cluster Ion Source, at a scattering angle of 130°, and 320 eV analyzer pass energy, averaged over a sample area of 500 μm 2 .

Synthesis of xPt/Fe
Catalytic evaluation. CH 4 oxidation was performed in a U-shaped quartz reactor using 20 mg catalyst and a feed gas comprising 0.5 vol% CH 4 , 3 vol% O 2 , and 97 vol% He balance at 30 mL min −1 . The effluent gas composition was analyzed by an on-line Agilent 6890 A gas chromatograph equipped with a TDX-01 column and a thermal conductivity detector. CH 4 conversion was calculated based on the difference between inlet and outlet concentrations.
Turnover frequency (TOF) calculation. TOFs were calculated from specific activities measured at 700°C and approximately 20 % CH 4 iso-conversion for 17 mg 1Pt/Fe 2 O 3 -NP and 6 mg 1Pt/Fe 2 O 3 -C700 catalysts respectively. A common gas flow rate of 50 mL min −1 was used, resulting in a space velocity of 176,000 mL h −1 g cat −1 for 1Pt/Fe 2 O 3 -NP and 500,000 mL h −1 g cat −1 for 1Pt/Fe 2 O 3 -C700. The Pt dispersion of 1Pt/Fe 2 O 3 -NP was calculated according to the following relationship between the dispersion (D) and particle radius (r): D / % = 100*5.6/r Pt . The average particle radius of 1Pt/Fe 2 O 3 -NP was 15 Å (Fig. 4c), corresponding to a dispersion of 30 %, while that for 1Pt/Fe 2 O 3 -C700 was 100 %. TOFs were calculated by normalizing specific activities to the product of the dispersion and total Pt mols (identical for both catalysts).
DFT parameters. All DFT + U calculations were performed using spin-polarized density functional theory (DFT) with the generalized gradient approximation (GGA) and Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional as implemented in VASP5.3.5 [64][65][66] . The U-J value of 3.0 eV was used to describe the strong correlation of the localized Fe 3d states 3,67,68 . Valence states of all atoms were expanded in a plane wave basis set with a cutoff energy of 400 eV, and a Monkhorst-Pack mesh of 3 × 3 × 1 k points was used for Brillouin Zone integration. Atomic positions were optimized by the conjugate gradient algorithm until forces were < 0.02 eV/Å. Free energy changes were calculated from zero-point energies, enthalpies, and entropies of gas Pt single atom and PtO 2 species corrected at 600, 700 and 800°C, respectively as summarized in Supplementary  Table 3.
Computational models. A Pt(221)-p(3 × 1) surface slab was used to model the step and edge of Pt NPs (coordination number = 7 for atoms at edges). We applied periodic models of surface slabs with five atomic layers, with the bottom two layers frozen and the remaining layers allowed to relax. The (0001) surfaces of α-Fe 2 O 3 -p (2 × 2) were also represented by a periodic slab model. Antiferromagnetic properties of α-Fe 2 O 3 were represented by a ( + --+ ) magnetic configuration, which was previously proven to be most energetically favorable for α-Fe 2 O 3 3 . We chose slabs containing 9 layers of Fe atoms and 5 atomic layers of O to model the O-terminated surface. The bottommost Fe-O layers were frozen during geometry optimization. A θ-Al 2 O 3 (010)-p(4 × 2) surface slab was used to model the alumina substrate, consisting of six O-Al layers, wherein the bottom two O−Al layers were frozen while the remaining layers were allowed to relax 69,70 . The IDIPOL tag was set to 3 to switch on dipole corrections to the total energy along the z-direction. All supercell slabs were periodically repeated with a 15 Å vacuum layer between surfaces in the direction of the surface normal.
Theoretical maximum loading of dispersed Pt atoms over Fe 2 O 3 support. The BET surface area of Fe 2 O 3 was 5-10 m 2 g −1 , hence 1 g of Fe 2 O 3 provides~5-10 m 2 of surface (S). Our DFT calculation model indicates that the maximum density of atomically dispersed Pt (D) is 4.5 atom nm −2 . The total number of isolated Pt atoms (N) that could be achieved for 1 g of Pt/Fe 2 O 3 is therefore predicted to be N = D × S. Since the mass of Pt (m) 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 Pt (195 g mol −1 ), the theoretical maximum loading of isolated Pt atoms that could be dispersed over 1 g of Fe 2 O 3 is m(g)/1(g) × 100 % = (D × S / N A ) × M × 100% = 1.5 wt%.

Data availability
The data that support the plots within this paper and other findings of this study are available from the corresponding authors upon reasonable request.