Functional CeOx nanoglues for robust atomically dispersed catalysts

Single-atom catalysts1 make exceptionally efficient use of expensive noble metals and can bring out unique properties1–3. However, applications are usually compromised by limited catalyst stability, which is due to sintering3,4. Although sintering can be suppressed by anchoring the metal atoms to oxide supports1,5,6, strong metal–oxygen interactions often leave too few metal sites available for reactant binding and catalysis6,7, and when exposed to reducing conditions at sufficiently high temperatures, even oxide-anchored single-atom catalysts eventually sinter4,8,9. Here we show that the beneficial effects of anchoring can be enhanced by confining the atomically dispersed metal atoms on oxide nanoclusters or ‘nanoglues’, which themselves are dispersed and immobilized on a robust, high-surface-area support. We demonstrate the strategy by grafting isolated and defective CeOx nanoglue islands onto high-surface-area SiO2; the nanoglue islands then each host on average one Pt atom. We find that the Pt atoms remain dispersed under both oxidizing and reducing environments at high temperatures, and that the activated catalyst exhibits markedly increased activity for CO oxidation. We attribute the improved stability under reducing conditions to the support structure and the much stronger affinity of Pt atoms for CeOx than for SiO2, which ensures the Pt atoms can move but remain confined to their respective nanoglue islands. The strategy of using functional nanoglues to confine atomically dispersed metals and simultaneously enhance their reactivity is general, and we anticipate that it will take single-atom catalysts a step closer to practical applications. Nanometre-sized ‘nanoglue’ islands of CeOx on high-surface-area SiO2 are shown to suppress sintering and confine on average one Pt atom per island, leading to stable single-atom catalysts under oxidizing and reducing environments.

when in contact with acidic Pt precursor solutions of pH 3-5, so that PtCl 6 2− is attracted to the CeO x nanoclusters (Extended Data Fig. 3a). To synthesize CeO x -supported Pt 1 single-atom catalysts, the Pt loading was controlled at less than or equal to 0.4 wt% (with respect to the CeO x ) such that, on average, each CeO x nanocluster contained less than one Pt atom (Extended Data Fig. 3b,c). High Pt loadings, which are industrially desirable (for example, up to 1 wt% Pt), can be realized by increasing the SiO 2 surface area and/or the number density of isolated CeO x nanoclusters (Extended Data Fig. 3d-f). Residual chloride was removed by washing and high-temperature calcination (Extended Data Fig. 3g-i).
The locations of isolated metal atoms on supports determine their catalytic properties 6 . Individual Pt atoms on well crystallized CeO 2 NPs have been observed by HAADF-STEM 25 , but unambiguous identification of Pt atoms on ultrasmall CeO x nanoclusters is beyond the capability of this technique (Extended Data Fig. 4a-c and Methods). However, an extensive HAADF-STEM investigation of Pt atoms and clusters on the CeO x /SiO 2 can firmly identify Pt clusters with sizes greater than 0.4 nm if present on the samples (Extended Data Fig. 4d). Analysis of numerous atomic-resolution images of the 0.4 wt% Pt/CeO x /SiO 2 catalyst gave no evidence of Pt atoms or clusters on the SiO 2 surfaces, confirming that the deposited Pt species were restricted to CeO x nanoislands-either as single atoms or clusters with sizes less than 0.4 nm.
X-ray absorption spectra provided further insights into the nature of the Pt species. Pt L III -edge X-ray absorption near-edge structure (XANES) data (Fig. 3a) characterizing the calcined 0.4 wt% Pt/CeO x /SiO 2 show that the Pt was cationic, with an oxidation state close to that in bulk PtO 2 (ref. 26 ). Extended X-ray absorption fine structure (EXAFS) spectra ( Fig. 3b) indicate atomically dispersed Pt, with no evidence of a Pt-Pt scattering path in the spectra of the 0.4 wt% Pt/CeO x /SiO 2 ( Fig. 3b and Extended Data Fig. 4e-j). A Pt-O shell indicating Pt-CeO x bonding was found with a coordination number of 4.5 ± 0.5 and a bonding distance of 1.97 ± 0.02 Å (error bounds defined in Extended Data Fig. 4), consistent with EXAFS data for site-isolated platinum on cerium dioxide or iron oxide 1,26 . Diffuse-reflectance infrared Fourier-transform spectra (DRIFTS) of the 0.4 wt% Pt/CeO x /SiO 2 , after CO adsorption, show a sharp single peak near 2,103 cm −1 with a full-width at half-maximum of 12.8 cm −1 (Fig. 3c), assigned to CO adsorbed on isolated cationic Pt 5 ultrasmall CeO x nanocrystallites. These ultrasmall, isolated CeO x nanoclusters act as functional nanoglues to localize metal atoms and to provide active oxygen species. By judiciously adjusting the aqueous solution pH so that CeO x nanoclusters are positively charged, negatively charged Pt-containing species selectively adsorb onto the CeO x nanoclusters. Subsequent vigorous washing and high-temperature calcination eliminate solution residues and facilitate confinement of Pt atoms to isolated CeO x nanoclusters.

Article
To further understand the stability of the atomically dispersed Pt, the behaviour of Pt atoms on other supports under various conditions was compared. The data include those obtained for Pt atoms on SiO 2 and on CeO 2 . Under reducing or oxidizing environments at temperatures greater than 300 °C, Pt atoms on SiO 2 sintered considerably (Extended Data Fig. 5a-f), demonstrating weak Pt-SiO 2 interactions 10 . Single Pt atoms on CeO 2 NPs did not sinter during calcination, even at high temperatures (Extended Data Fig. 5g-h) 5,8 . Exposure to H 2 at 300 °C for 1 h, however, caused breaking of Pt-O bonds, migration of Pt atoms on CeO 2 NPs and sintering of Pt atoms to form clusters and NPs (Extended Data Fig. 5i-k). After similar treatment, the Pt NPs formed on CeO 2 were smaller than those formed on SiO 2 (Extended Data Fig. 5), suggesting that the Pt-CeO 2 interaction is stronger than the Pt-SiO 2 interaction during the H 2 -reduction treatment. Investigations of Pt sandwiched between hetero-structured CeO 2 -SiO 2 core-shell model systems confirmed the stronger interaction of Pt atoms with CeO 2 than with SiO 2 , resulting in the suppression of Pt diffusion from CeO 2 onto SiO 2 (Extended Data Fig. 6 and Methods).
The stability of the CeO x -supported Pt single-atom catalyst (Pt 1 /CeO x / SiO 2 ) was investigated under reducing conditions. CeO 2 NP-supported Pt atoms became mobile at temperatures greater than 300 °C for 1 h in H 2 , but Pt atoms, confined to the isolated CeO x islands, did not sinter, even after H 2 reduction at 300 °C for 10 h, as shown by DRIFTS and HAADF-STEM data ( Fig. 3d and Extended Data Fig. 7a). The minimal difference in the CO absorption bands characterizing Pt atoms in the reduced versus the as-synthesized catalysts is evidence that the CO probe molecules used in the DRIFTS experiments reduced the Pt atoms in the as-synthesized Pt 1 /CeO x /SiO 2 , as expected 27 . To further probe the stability of Pt atoms localized on the CeO x islands, samples were exposed to H 2 at temperatures of 400-600 °C. Even under these harsh reducing conditions, the Pt remained as isolated single atoms (Extended Data Fig. 7b-d). The infrared absorption bands characterizing CO on Pt atoms were shifted after reduction in H 2 at temperatures greater than 500 °C, suggesting modifications of the Pt-CeO x interactions 28 , which were corroborated by density functional theory calculations (Extended Data Fig. 7e). Because the DRIFTS experiments were conducted under oxygen-free environments, formation of Pt x O y clusters was excluded (Extended Data Fig. 7f-h,k-p and Methods section 'Extended results and discussions') 29 . Both Ce XPS and Pt XAS data show that formation of Pt clusters or Pt-Ce alloy species under the reduction conditions did not occur (Extended Data Figs. 7i, 9i-p). Thus, notwithstanding changes in the catalysts, the Pt remained site-isolated, with the changes restricted to the host CeO x islands. By contrast, decreasing Pt atom number density on CeO 2 supports (with loadings lower than that of the 0.4 wt% Pt 1 /CeO x /SiO 2 ) did not prevent Pt atoms from forming clusters and NPs after reduction in H 2 at 400 °C (Extended Data Fig. 7j).
To confirm that during H 2 reduction the Pt confinement on SiO 2 -supported CeO x islands also applies to SiO 2 -supported CeO 2 NPs, a sample containing Pt atoms dispersed on SiO 2 -supported 8-nm CeO 2 NPs (0.4 wt% Pt/CeO 2 NPs/SiO 2 ) was exposed to H 2 at 300 °C for 1 h. The data (Extended Data Fig. 8a-c) show that the Pt atoms did not migrate onto SiO 2 , but they did sinter to form small Pt clusters, because larger CeO 2 NPs, on average, contained more than one Pt atom, further confirming the conclusion that under reduction conditions Pt atoms sinter on CeO 2 surfaces. The CO DRIFTS data (Extended Data Fig. 8d- Fig. 8g). HAADF-STEM images show uniform Pt clusters (an average size of approximately 0.9 nm) attached to the isolated CeO x islands when the reduction temperature increased to 400 °C (Extended Data Fig. 8h). After reduction at 500 °C for 12 h, the CeO x islands became amorphous, but the Pt clusters nonetheless retained their sizes and were still attached to the islands (Extended Data Fig. 8i-j). The absence of Pt NPs with sizes greater than 1 nm in these severely reduced catalysts with high Pt loading highlights the effectiveness of the CeO x nanoglue strategy for stable localization of Pt species (from isolated metal atoms to subnanometre clusters) and for preventing the formation of larger Pt NPs, thus expanding their potential for practical applications.
We note that the active phase in many catalysts is pre-formed by reduction, and that H 2 treatment at temperatures greater than 200 °C usually causes sintering of atomically dispersed noble metals 4,8,9 . In addition to limiting applications, this restructuring also hinders fundamental investigations of single-atom catalyst performance and structure-property relationships. The robustness of the 0.4 wt% Pt 1 / CeO x /SiO 2 under both oxidizing and reducing environments offers new opportunities in this regard.
Ceria-supported single Pt atoms were reported to be less active for CO oxidation than CeO 2 -supported Pt clusters 8,9,27 , and our data verify this pattern (Extended Data Fig. 9a-c). Because our CeO x nanoisland-confined Pt single-atom catalysts remain stable after reductive activation, we could explore CO oxidation performances and show (Fig. 4) that the temperature for 50% CO conversion is 133 °C for the H 2 -activated Pt 1 /CeO x /SiO 2 and 226 °C for the non-activated Pt 1 / CeO x /SiO 2 . This result demonstrates that H 2 activation increases the CO oxidation rate by two orders of magnitude and decreases the apparent activation energy (Extended Data Fig. 9q). The XANES and EXAFS data indicate that H 2 reduction removes oxygen ligands, reduces the coordination number of the first Pt-O shell from 4.5 ± 0.5 to 3.2 ± 0.3 and reduces Pt 4+ to Pt δ+ (refs. 6,27,30 ) without sintering the isolated Pt atoms (Extended Data Fig. 9i-p). Comparison against catalysts prepared on other CeO 2 supports or by impregnation of CeO x /SiO 2 and SiO 2 with Pt salt solutions shows that the H 2 -activated 0.4 wt% Pt 1 /CeO x /SiO 2 is much more stable and active for CO oxidation (Extended Data Fig. 9d-h,r-t).
These observations illustrate the value of the CeO x nanoglue design strategy, implemented with our scalable strong electrostatic adsorption process for dispersing CeO x (x ≈ 1.86) nanoglue islands with an average dimension of 2 nm or smaller onto a robust, high-surface-area SiO 2 support and then selectively localizing Pt atoms on these islands. The CeO x nanoglue islands incorporate abundant Ce 3+ that strongly anchors Pt atoms and small clusters under either O 2 or H 2 environments, even at elevated temperatures. Challenges for practical applications remain, however, such as the partial and reversible oxidation of the Pt 1 atoms and associated lowering of catalyst activity seen at higher temperatures (for example, 300 °C) under oxidizing conditions (Methods section 'Extended results and discussions' and Extended Data Fig. 9i-p). But with our strategy for confining metal atoms by functional nanoglues applicable to metals other than Pt (including Pd and Rh; see Extended Data Fig. 10) and in principle able to produce a wide range of robust 11 Article single-atom and cluster catalysts, we anticipate that it will prove useful for numerous catalytic transformations.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05251-6.
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.   4 . Such a time difference in the dynamic adsorption and precipitation processes allows efficient, facile production of uniformly decorated high-surface-area SiO 2 with Ce species via the SEA process.

Preparation of nanoglues and support materials
In a typical procedure for synthesizing 12 wt% CeO x nanoglue islands on SiO 2 , 360 mg of fumed SiO 2 powder (Alfa Aesar, total Brunauer-Emmett-Teller (BET) surface area 278 m 2 g −1 ) was mixed with 100 ml of deionized (DI) water, and the resultant aqueous solution was sonicated to produce a uniform mixture. Under vigorous stirring, 0.4 mmol of Ce(NO 3 ) 3 ·6H 2 O was dissolved in the above solution, and 0.8 ml of NH 3 ·H 2 O (2 M) was quickly injected into the solution with further stirring for 3 min prior to vacuum filtration. The final solution pH was about 8.7. The resultant precipitates were removed for air drying overnight at room temperature. The dried powders were ground to smaller particles and were then calcined in a muffle furnace at 500 or 600 °C (with a temperature ramp of 5 °C min −1 ) for 12 h. This facile synthesis protocol was easily scaled up to routinely produce 10 times more of the final product.
SiO 2 -supported CeO 2 NPs (IMP) by impregnation. The SiO 2 -supported 12 wt% CeO 2 NPs were prepared using the same precursors as described above. The pore volume of the SiO 2 was determined to be ~2.1 ml g −1 . Therefore, 0.8 mmol of Ce(NO 3 ) 3 ·6H 2 O was dissolved in 2.1 ml of DI water. As the sample was stirred, 2.1 ml of precursor solution was slowly added to 1 g of SiO 2 powder to ensure complete wetting of the support by the Ce-containing solution. The final mixture was air-dried overnight prior to oven drying at 60 °C for 12 h. The dried powders were ground and then calcined at 600 °C (the temperature was ramped at a rate of 5 °C min −1 ) in a muffle furnace.
Pure CeO 2 powders. To synthesize pure CeO 2 , 2 g of Ce(NO 3 ) 3 ·6H 2 O was dissolved in 100 ml of DI water. With the sample vigorously stirred, 2 ml of 15% ammonia solution was slowly injected into the above solution to form a suspension. The product was continuously stirred overnight in open air. Vacuum filtration was used to repeatedly wash the precipitates with DI water. The final light-yellow powders were ground and then calcined at 500 °C for 4 h.

Selective deposition of Pt atoms and preparation of Pt catalysts
Pt/CeO x /SiO 2 , Pt/CeO 2 NPs/SiO 2 and Pt/CeO 2 . In a typical synthesis protocol, 300 mg of support (500 °C-calcined CeO x /SiO 2 , CeO 2 NPs/SiO 2 ) was immersed in 70 ml of DI water and sonicated. The pH of the solution was adjusted to ~3 by addition of HCl. Simultaneously, 3.0 μmol of H 2 PtCl 6 ·6H 2 O was diluted in 50 ml of aqueous solution (pH of ~3 by addition of HCl). With the sample vigorously stirred and within 5 h, the Pt precursor solution was slowly pumped to the aqueous solution containing the corresponding support powders. After being stirred for another 5 h, the final mixture was filtered via vacuum filtration to obtain the precipitates, which were washed several times with DI water to remove residual Pt ions and other species. After the final catalyst precursor powders were dried in air, they were calcined at 600 °C for 12 h to obtain the as-synthesized catalysts. 0.3 wt% Pt/CeO 2 catalyst was synthesized by using 30 μmol of H 2 PtCl 6 ·6H 2 O and 300 mg of CeO 2 powder. The high Pt loading (4 wt% Pt with respect to CeO x ) Pt/CeO x / SiO 2 catalyst was synthesized by using 61 μmol of H 2 PtCl 6 in the above process. After calcination, 4 wt% Pt/CeO x /SiO 2 catalyst was reduced in H 2 at 400 °C to produce Pt clusters. The Pt x O y /CeO x /SiO 2 samples were produced by mildly calcining the CeO x /SiO 2 supported Pt cluster catalysts at 100 °C. For Pd/CeO x /SiO 2 and Rh/CeO x /SiO 2 , 14 μmol of PdCl 2 and 5.8 μmol of RhCl 3 were used in the synthesis process, respectively.
Pt/CeO x /SiO 2 (IMP) by impregnation. The pore volume of the as-synthesized CeO x /SiO 2 was determined to be 1.8 ml g −1 . With stirring, 0.54 ml of aqueous solution containing H 2 PtCl 6 precursor salt was slowly added to 300 mg of CeO x /SiO 2 powder to ensure complete wetting of the support surfaces by the precursor salt solution. The mixture was then dried at 60 °C for 12 h in an oven. The resultant 0.4 wt% Pt/CeO x / SiO 2 (IMP) powders were ground and then calcined at 300 °C for 3 h in a muffle furnace. The high-loading Pt/CeO x /SiO 2 (IMP) powders were directly reduced in H 2 at 300 °C for STEM examination.
Pt/SiO 2 (IMP) by impregnation. 50 μl of H 2 PtCl 6 solution (3 mg ml −1 ) was diluted with 580 μl of DI water. With stirring, 630 μl of precursor solution was slowly added to 300 mg of SiO 2 powder to ensure complete wetting of the support by the metal-containing solution. The final mixture was dried at 60 °C for 12 h in an oven. The dried powders were ground and then calcined at 300 °C for 3 h in a muffle furnace.

Pt/SiO 2 (SEA).
In a typical synthesis process, Pt(NH 3 ) 4 Cl 2 was used as precursor to adsorb Pt(NH 3 ) 4 2+ onto negatively charged SiO 2 surfaces in the presence of alkaline solution. Commercial SiO 2 powders (180 mg) were dispersed in 50 ml of DI water and the solution was kept alkaline by adding 0.2 ml of 2-M ammonia solution. Then 1 ml of aqueous solution containing 3.2 mg of Pt(NH 3 ) 4 Cl 2 was injected into the SiO 2 suspension that was vigorously stirred for 3 h. The as-prepared Pt/SiO 2 catalysts, without calcination, contained both Pt single atoms and Pt clusters (Extended Data Fig. 5a,b). The Pt loading on commercial SiO 2 powders was determined to be 0.5 wt% by inductively coupled plasma mass spectrometry (ICP-MS) measurements.
Preparation of core-shell samples SiO 2 spheres. SiO 2 spheres were produced by the Stöber method 32 : 30 ml of DI water, 150 ml of ethanol, and 18 ml of ammonia (~25%-28%) were mixed in a flask. Then 12 ml of tetraethyl orthosilicate (TEOS) was added to the solution, which was heated to 58 °C and stirred for 2 h. Then the solvent was removed by evaporation at 100 °C, and the resultant powder was calcined for 2 h at 500 °C.
Pt/SiO 2 spheres. 500 mg of SiO 2 spheres were dispersed in 100 ml of DI water while 1 ml of 2-M ammonia solution was added. Then 50 ml of solution containing 20 mg of Pt(NH 3 ) 4 (NO 3 ) 2 was pumped into the suspension within 1 h. After stirring for another half an hour, the mixture was filtered. The Pt loading on SiO 2 spheres was determined to be 0.2 wt% by ICP measurement.
Porous SiO 2 shell. The (Pt/CeO 2 )@SiO 2 (Extended Data Fig. 6) was prepared by a modified Stöber method 33 ; 500 mg of the as-prepared 0.3 wt% Pt/CeO 2 was dispersed in 300 ml of ethanol by ultrasonication. Then 2.5 ml of TEOS was added, and the solution was stirred for 4 h. Then 5 ml of ammonia (~25%-28%) and 20 ml of DI water were added to the slurry, which was stirred for another 4 h. Finally, the product was washed with ethanol and dried at 100 °C. For (Pt/SiO 2 )@SiO 2 , 150 mg of 0.2 wt% Pt/SiO 2 spheres and 1.5 ml of TEOS were used in the synthesis.
Cu nanoparticles. A specified amount of (Pt/CeO 2 )@SiO 2 or (Pt/SiO 2 )@ SiO 2 was added to 25 ml of DI water. The pH value of the solution was kept at 9 by adding Na 2 CO 3 solution. Then Cu(NO 3 ) 2 solution was added dropwise to the above suspension. The target loading of Cu was 2 wt% in each sample. After ageing for 1 h, the products were filtered and dried at 70 °C overnight. The final products, denoted as Cu[(Pt/CeO 2 )@ SiO 2 ] or Cu[(Pt/SiO 2 )@SiO 2 ], were in situ reduced in 20 ml min −1 of 10% H 2 /Ar at 400 °C for 3 h in preparation for DRIFTS and HAADF-STEM characterizations.

Catalytic testing of the prepared catalysts
The CO oxidation reaction was conducted in a fixed-bed, plug-flow reactor at atmospheric pressure. Typically, 30 mg of 0.4 wt% Pt/CeO x / SiO 2 catalyst was packed between two quartz wool plugs inside a quartz tube (inner diameter of 4 mm) for each test. Prior to catalytic reaction, the H 2 -activated catalysts were pretreated with 10 ml min −1 of 5% H 2 /He at 300 °C for 1 h. After cooling to room temperature, the reaction temperature was ramped up at a rate of 1 °C min −1 . The feed gas, containing 1 vol.% CO, and 4 vol.% O 2 balanced in He, passed through the catalyst bed at a flow rate of 10.0 ml min −1 (corresponding to a weight hourly space velocity (WHSV) of 20,000 ml (g cat h) −1 ). The outlet gas composition was measured with an online gas chromatograph (7890A, Agilent) equipped with a thermal conductivity detector (TCD).
The specific reaction rate and apparent activation energy were measured as the catalyst was exposed to the gas composition stated above. During the kinetics experiments, the CO conversion was controlled to be less than 15% by adjustment of either the feed flow rate or catalyst mass.

Characterization instruments
HAADF-STEM images were collected with a JEOL ARM-200F microscope equipped with a probe-forming aberration corrector operated at 200 kV to achieve a nominal image resolution of 0.08 nm. The backscattered electron images were obtained on a JEOL JXA-8530F electron microprobe. Transmission electron microscopy (TEM) images were taken on a Hitachi HT7700 transmission electron microscope operating at an acceleration voltage of 100 kV. All reported electron microscopy images were raw images. X-ray photoelectron spectroscopy (XPS) analysis was carried out with an ESCALAB 250 X-ray photoelectron spectrometer with Al Kα as the excitation source. The loadings of Pt and Ce were determined with a ThermoFinnegan iCAP Q quadrupole ICP-MS with CCT (Collision Cell Technology), and Pd and Rh were determined with inductively coupled plasma-atomic emission spectroscopy (ICP-AES) (Atomscan Advantage, Thermo Jarrell Ash). Samples were run in KED (kinetic energy discrimination) mode, with in-line aspiration of a multi-element internal standard. The BET surface areas were measured by the nitrogen adsorption-desorption method on a surface area and porosity analyser (Quantachrome NOVA 4000e apparatus). Before measurements, the samples were degassed at 180 °C for 6 h under vacuum. The X-ray diffraction (XRD) patterns were recorded on a Rigaku TTR-III theta-theta rotating anode X-ray diffractometer using Cu Kα radiation (40 kV and 200 mA) with a step size of 0.02°. Raman spectra were collected with a JYLABRAM-HR spectrometer equipped with an integral microscope. DRIFTS experiments were carried out with samples in a diffuse reflectance reaction chamber (Harrick Scientific) equipped with ZnSe windows, mounted in a Praying Mantis diffuse reflection accessory (Harrick Scientific), and coupled to a Thermo Scientific Nicolet iS50 FTIR spectrometer with a liquid-nitrogen-cooled HgCdTe (MCT-A) detector. Hydrogen temperature-programmed reduction (H 2 -TPR) measurements were conducted on a Micromeritics Autochem II 2920 with a thermal conductivity detector (TCD). X-ray absorption spectroscopy (XAS) experiments were carried at beamlines 4-1 and 9-3 at the Stanford Synchrotron Radiation Lightsource (SSRL) and at the TPS 44A beamline of the National Synchrotron Radiation Research Center (NSRRC), Taiwan. The storage ring energies and currents were 3 GeV and 500 mA, respectively. Photon energy selection was achieved with double-crystal Si(220) monochromators at SSRL. A Si (111) channel-cut crystal was used as the quick-scanning monochromator at NSRRC.

Characterization methods
ICP. ICP-MS measurements were performed to quantify the loadings (loading levels) of metal species on SiO 2 . For a particular weight of Pt/CeO x /SiO 2 catalyst, Pt and Ce atoms were dissolved in freshly prepared aqua regia solution. Then the insoluble SiO 2 was washed and centrifuged three times with all the solution in a volumetric flask. The final concentrations of Pt and Ce were diluted to meet the calibration standards. The loading of CeO x was calculated in terms of CeO 2 units. To verify that Pt atoms were selectively adsorbed by CeO x nanoglues, the CeO x /SiO 2 support was replaced by pure SiO 2 in a typical synthesis procedure. Because the loading of Pt on the pure SiO 2 was expected to be much lower than that on the CeO x /SiO 2 , larger amount of samples of Pt/SiO 2 were used for the ICP-MS measurement to assure that the concentration of Pt was in the range of the reference standards. The results are summarized in Extended Data Fig. 3a.

XPS.
For XPS investigations of the high-temperature reduced 0.4 wt% Pt/CeO x /SiO 2 , as-synthesized powder samples were reduced in 10% H 2 /Ar at 500 °C for 1 h. The reactor vessel was then tightly sealed and transferred to a glove box. The catalyst powders were then transferred into the XPS sample holder within the glove box. The XPS sample holder was tightly sealed in the glove box to protect the catalysts from oxidation by air during transfer. After the sample preparation procedure, the XPS sample holder was placed in a vacuum transfer chamber and transferred to the XPS apparatus without sample exposure to air.

HAADF imaging of CeO x /SiO 2 and Pt/CeO x /SiO 2 . The mesoporous
SiO 2 support has interconnected 3D pores and underwent considerable charging under electron beam illumination, making imaging of such materials challenging. In the HAADF images shown in Extended Data Fig. 1d,e, the adjacent regions could not be simultaneously imaged because atomic-resolution HAADF imaging is sensitive to height modulations of the mesoporous SiO 2 support: CeO x clusters located at different sample heights would not be clearly revealed in the same HAADF image. Through-focus images were needed to observe clear features of the adjacent regions. Prior to calcination, the conforming coating of Ce species on the mesoporous SiO 2 support made the SiO 2 powders less susceptible to electron beam charging effects. After high-temperature calcination, however, the mesoporous SiO 2 became charged under the influence of the electron beam, owing to the fact that the crystalline CeO x nanoclusters were isolated from each other. Because of the sample charging and electron-beam-induced effects, crystalline CeO x nanoclusters would usually be transformed quickly into amorphous species. For this reason, a low-dose STEM method was used to acquire STEM images of the sensitive CeO x nanoclusters. In some cases, a thin layer (<2 nm) of continuous carbon coating of the CeO x /SiO 2 or Pt/CeO x /SiO 2 powders was used to reduce the sample charging and other electron-beam-induced effects.
The HAADF imaging is sensitive to the atomic number of the atoms in the sample and the sample thickness. For Pt atoms supported on well faceted CeO 2 , it is possible to distinguish a Pt atom from columns of Ce atoms, because the image contrast can be considered to be proportional to where N Ce represents the number of Ce atoms along the electron beam direction. Therefore, even if a single Pt atom is located on top of a 30-Ce-atom-thick CeO 2 column, the image contrast would still be greater than 5%-and reliably detectable in digital HAADF images. However, in this special case, even when a thin layer of carbon coating was used to reduce electron-beam-induced effects and sample charging, low-dose STEM imaging conditions and fast image acquisition were needed to preserve the crystallinity of the electron-beam-sensitive CeO x nanoclusters. Such a low-dose approach greatly reduced the visibility of single Pt atoms. Furthermore, because the CeO x nanoclusters were extremely small and their surfaces were characterized by many steps or other types of surface defects, it became extremely difficult to reliably distinguish the contrast between a supported Pt atom from that of a supported Ce atom (or two Ce atoms). Depending on the specific location of the Pt atom, distinguishing the image contrast of a Pt atom from that of two or more Ce atoms is an impossible task with the currently available technology. All these complications made it impossible to unambiguously identify the presence of Pt single atoms on the CeO x clusters by the STEM-HAADF imaging technique. Nor did EDXS (energy-dispersive X-ray spectroscopy) and EELS (electron energy-loss spectroscopy) techniques provide useful results, primarily owing to the electron-beam-induced effects.
CO DRIFTS. In a typical experiment, the infrared cell was partly filled with inert KBr powder, followed by catalyst packed onto the KBr support. All the gases that flowed through the cell flowed downward through the catalyst bed. For fresh catalysts, a pretreatment at 300 °C with 10% O 2 /Ar was conducted for 30 min. The flowing gas was then switched to He, and the background spectra were recorded at 50 °C characterizing the as-synthesized 0.4 wt% Pt/CeO x /SiO 2 catalyst. CO adsorption was conducted by flowing 15 ml min −1 of 10% CO/Ar and 30 ml min −1 of He for 20 min. The CO desorption spectra were recorded following a He purge at a flow rate of 30 ml min −1 to remove gas-phase CO from the cell. When the sample to be characterized was Pt x O y /CeO x / SiO 2 , no further oxidative pretreatment was carried out. The CO adsorption experiment was conducted at 100 °C by flowing 10 ml min −1 of 10% CO/Ar, 10 ml min −1 of 10% O 2 /Ar, and 20 ml min −1 of He. Then the O 2 flow was stopped to allow investigation of the CO adsorption behaviour on Pt x O y clusters. The in situ H 2 reduction treatments were conducted with the sample in the cell-fresh catalyst samples were loaded. The catalysts were heated to the target temperatures in flowing 10% O 2 /Ar, and then the gas was switched to 10% H 2 /Ar. During the reduction treatment at 300-600 °C, 20 ml min −1 of 10% H 2 /Ar flowed through the cell for various periods. The cell was then cooled down to the desired temperatures with the sample in the H 2 environment. The flowing gas was then switched to He and backgrounds of the Pt-containing, Pd-containing and Rh-containing catalysts were recorded at 100 °C, 25 °C and 180 °C, respectively. CO adsorption was investigated as the sample was exposed to 10% CO/Ar flowing at 15 ml min −1 and He flowing at 30 ml min −1 for 20 min. The CO desorption spectra were recorded following a He purge at a flow rate of 30 ml min −1 to remove gas-phase CO from the cell. The measurement process was conducted in such a way to avoid oxidation of Pt species.
Simulation of CO IR spectra. To predict possible adsorption configurations and vibrational frequencies of CO on harshly reduced (above 500 °C) Pt/CeO x /SiO 2 , spin-polarized density functional theory (DFT) calculations were performed using the Vienna Ab-Initio Simulation Package (VASP) code 34 . The projector augmented-wave method (PAW) and the Perdew-Burke-Ernzerhof (PBE) functional were used to deal with the electron-ion interactions and exchange correlation, respectively 35,36 . DFT+U correction (U eff = 5 eV for Ce) was considered to treat the strongly correlated 4f electron 37 . The van der Waals interactions were described using the empirical DFT+D3 method to improve the accuracy of representation of CO adsorption 38 . The gamma point was set for the total energy calculations together with a cut-off energy of 400 eV for the plane-wave basis. A vacuum space larger than 10 Å was added in each direction to avoid the interactions between each cluster model. During the structure optimization and frequency analysis, all the atoms were relaxed until the force on each ion was less than 0.02 eV Å −1 .
To obtain plausible structures of harshly reduced Pt/CeO x /SiO 2 , ab initio molecular dynamics (AIMD) simulations were performed. First, a Pt-doped model cluster (Pt 1 /Ce 15 O 30 ) with local coordination environment consistent with the EXAFS results of the as-synthesized Pt 1 /CeO x (crystalline) was established. Then the canonical ensemble (NVT) with the Nosé-Hoover thermostat was implemented in the AIMD simulations for a period of 2 ps at 873 K to give relatively stable structures. The generated configurations were further optimized using standard DFT calculations. The reduced Pt 1 /CeO x was represented by a model (

XAS measurements.
Ex situ EXAFS spectra were used to characterize the Pt-containing sample supported on CeO x /SiO 2 at the Pt L III edge. We used a highly sensitive fluorescence detection technique 39 , carried out with a 30-element solid-state Ge detector at SSRL beamline 4-1 and with a 100-element solid-state Ge detector at beamline 9-3 with each sample pressed into a pellet at 25 °C. In situ XANES spectra were collected at beamline 9-3 at SSRL by using the 100-element solid-state Ge detector at the Pt L III edge. Approximately 50 mg of catalyst sample was loaded into a flow-through cell and held in place with quartz wool; a Kapton capillary cell (inside diameter = 2.8 mm), connected to a treatment gas line 40 was used to collect in situ XANES spectra during treatment of the as-synthesized Pt/CeO x /SiO 2 in 10% H 2 /He flowing at a rate of 20 ml min −1 as the cell temperature was ramped from 25 to 300 °C at a rate of 5 °C min −1 followed by a dwell of 60 min at 300 °C. The Pt x O y /CeO x /SiO 2 catalyst was also investigated at the TPS 44A beamline of the NSRRC. The data were collected in fluorescence mode by using a seven-element silicon drift detector, and a standard Pt foil was used as reference for the energy calibration.
Analysis of the EXAFS data was carried out with Athena and XDAP. Athena, part of the Demeter package, was used to merge and deglitch the data, and XDAP was used for background subtraction, normalization and conversion of the data into an EXAFS function file. Reference backscattering phase shifts were calculated from crystallographic data determined with FEFF7. The Pt-O, Pt-O long , and Pt-O-Pt contributions were calculated on the basis of the structural parameters of PtO 2 ; the Pt-Ce contribution was calculated on the basis of the structural parameters of the PtCe alloy 41 . The number of parameters used in the fitting was less than the statistically justified number, computed with the Nyquist theorem, n = 2ΔkΔr/π + 2 (where Δk and Δr, respectively, are the wavevector range and distance interval in the real space range used in the fitting).
Data fitting was based on an iterative process with a difference-file technique to determine a model comparing the overall fits and fits of individual shells as well. The model was chosen as the best-fitting model when the k 1 -and k 3 -weighted EXAFS data, Fourier-transformed data, and Fourier-transformed data characterizing each shell contribution were overall in best agreement with the calculated fits. Quality of fits was evaluated by the value of 'goodness of fit', defined below: where χ exp and χ model are the experimental and calculated EXAFS functions, respectively; σ exp is the error in the experimental results; ν is the number of independent data points in the fit range; N free is the number of free parameters, and NPTS is the number of data points in the fit range. We emphasize that we examined the data characterizing the as-synthesized 0.4 wt% Pt/CeO x /SiO 2 catalyst to check for Pt-Pt contributions, but found no evidence of any; thus, the reported fit includes no contributions of Pt-Pt shells from those that might have been present characterizing either metallic platinum (platinum nanoparticles) (with a Pt-Pt absorber-backscatterer distance of 2.81 Å) or PtO 2 (which would give a Pt-O-Pt contribution at a distance of about 3.19 Å), consistent with the presence of atomically dispersed platinum. The coordination number of the Pt-O shell in the model without a Pt-Pt contribution was found to be 4.5 ± 0.5 with a length of 1.97 ± 0.02 Å (which is a bonding length), consistent with the EXAFS analyses reported elsewhere for site-isolated platinum anchored on ceria and on iron oxide 1,42 . A metal-O long shell has been observed for numerous samples comparable to ours, for example, for atomically dispersed platinum supported on the potassium form of zeolite LTL 43 . Such a shell is not easily identified with any particular metal-oxygen contribution; rather, because of the complexity of the surface structures, it is an average representing the metal atom near support surface oxygen atoms to which it is not chemically bonded. Our observation of a Pt-backscatterer contribution associated with the Ce atoms of the support is as expected for an atomically dispersed platinum species on the ceria support. The result is consistent, for example, with results characterizing atomically dispersed platinum species anchored on FeO x 1 . We emphasize, however, that the fit of the Pt-Ce shell characterizing our data does not provide precise information, instead representing an average of various absorber-backscatterer contributions characteristic of isolated Pt atoms on a heterogeneous surface with various Pt-Ce distances; it is beyond the capabilities of EXAFS spectroscopy to resolve these interactions at distances from 0.5 to 2.3 Å. We thus infer that there are some backscatterer contributions associated with the complexity of the CeO x surface that we could not resolve in our fitting technique.  Fig. 9k-p, respectively. The fits were found to be the best of the candidate models in terms of quality of fit evaluated by the values of 'goodness of fit' and also the parameters that make good chemical sense.
The fitting of the data characterizing the Pt x O y /CeO x /SiO 2 (4 wt% Pt relative to CeO x ) catalyst indicates the following absorber-backscat-

Determination of physical property parameters of catalysts
Estimated surface area of CeO x . According to the average sizes of CeO x nanoclusters determined from the HAADF images, loadings of Pt and Ce from ICP-MS results, and the BET surface area of Pt/CeO x / SiO 2 , we calculated various parameters characterizing the nanostructured CeO x /SiO 2 . We use a cuboidal model here for CeO x nanoclusters. The average dimension of these CeO x clusters was taken to be 2 nm with a volume (V CeO x ) of 8 nm 3 . The loading (X Ce ) of Ce atoms was taken to be 10 wt%. The surface area (SA) of Pt/CeO x /SiO 2 was measured to be 267 m 2 g −1 . Each unit cell of the CeO x contains four cerium atoms. The mass of one CeO x cluster is estimated by Extended results and discussions Stabilization strategies. The most critical challenge to practical applications of single-atom catalysts is to prevent sintering of metal atoms (that is, formation of clusters and nanoparticles) during a desired catalytic reaction, especially under reducing environments at elevated temperatures 3,4,45,46 . Substantial efforts have been made to anchor supported single metal atoms, for example, by filling cation vacancies to form chemical bonds 1,11 , step trapping 5,22 , edge-anchoring 47 , engineering of support defects to form strong bonds 48,49 , use of surface functional groups [50][51][52] and introduction of strong metal-support interactions 53 . All of these approaches have focused on manipulating the support structure or chemistry to anchor isolated single metal atoms through strong metal-support interactions or interactions between the supported single metal atoms and surface functional groups. These anchoring strategies, however, may not be effective for reactions taking place under strongly reducing and/or oxidizing environments at elevated temperatures. For example, H 2 reduction of metal oxide-supported metal atoms at temperatures >200 °C may cause sintering of metal species regardless of the nature of the metal oxide supports (whether they are reducible or nonreducible) 4 . Although H 2 activation is frequently used for supported metal particle catalysts, the heretofore unavoidable sintering of supported metal atoms during H 2 activation makes it difficult to evaluate the true catalytic performance of supported metal atoms.
The strategies for stabilization of supported metal particles and clusters, illustrated, for example, by formation of chemical bonds at metal-support interfaces 54 , encapsulation by porous structures 55 , and incorporation of additives on supports with strong metal-additive interactions 56,57 , are not readily transferred to stabilizing isolated single metal atoms. Alternative approaches are needed to mitigate or confine the movement of metal atoms during a targeted catalytic reaction.
Extended Data Fig. 1c-e shows HAADF images of the Ce-SiO 2 precursor. The Ce-containing species conformally and uniformly coated the mesoporous SiO 2 surfaces, and there was no evidence of large Ce-containing particles after the SEA process. Owing to the 3D mesoporous structure of the SiO 2 support, it needs several images at different electron beam focus values to reveal the full picture of the conformal Ce coating as demonstrated in the images of the same region (Extended Data Fig. 1d-e), which were obtained with defocus value differences of ~30 nm. Prior to high-temperature calcination, the Ce atoms were atomically dispersed on the SiO 2 . Extended Data Fig. 1f-i shows that the CeO x nanoclusters were uniformly and conformally 'glued' onto the mesoporous SiO 2 surfaces. The CeO x nanoclusters were isolated from each other. The average dimensions of the as-synthesized CeO x nanoclusters were measured to be 1.8 nm × 2.1 nm.
Extended Data Fig. 2 shows characterizations of various supports. In the case of the conventional impregnation method, SiO 2 might not interact strongly with the Ce species. After being dried, Ce-containing species formed large agglomerates. In pure CeO 2 powders, well crystalized CeO 2 NPs (~17 nm from XRD, determined by the Scherrer equation) connected together to form large agglomerates. The CeO 2 {111} lattice spacing was used as an internal calibration standard for measuring the CeO x {111} spacings (Extended Data Fig. 2h). To increase the measurement precision and to obtain statistically meaningful data, many sets of {111} lattice fringes were evaluated. Intensity line-scan profiles were used to calculate the corresponding lattice spacings. The average interplanar spacing of the {111} planes in the CeO x nanoclusters was measured to be 0.33 nm, a value markedly greater than the 0.31 nm characterizing the large CeO 2 nanoparticles-and indicating a considerable lattice expansion of the {111} planes in the CeO x nanoclusters. Because of their small sizes, strong lattice distortion, and high concentrations of oxygen vacancies, the ultrasmall (<2 nm) CeO x nanoclusters were much more easily reduced than the larger CeO 2 NPs (Extended Data Fig. 2j).
Extended Data Fig. 3a shows the ICP-MS results of Pt loadings in the final catalysts. The same SEA procedure and synthesis conditions were used for adsorbing Pt atoms onto CeO x /SiO 2 and pure SiO 2 supports. The Pt contents of the final catalysts suggest that the as-synthesized Pt/SiO 2 catalyst contained a negligible (0.0006 wt%) amount of Pt and that ~99% of all the Pt atoms adsorbed on the CeO x nanoclusters in the CeO x /SiO 2 catalysts. The ICP-MS data unambiguously demonstrate that there was a negligible loading of Pt atoms on the bare SiO 2 surfaces, corroborating the design strategy. Extended Data Fig. 3c shows estimates of the loading levels of Pt (relative to the CeO x nanoclusters) on the CeO x /SiO 2 . The average volume of one CeO x cluster is approximately 2 nm × 2 nm × 2 nm = 8 nm 3 . The lattice parameter of CeO 2 is 0. 54 Fig. 3e,f show plots of attainable wt% Pt loading (with respect to SiO 2 ) versus the specific surface area of SiO 2 , assuming that each CeO x nanocluster hosts only one Pt atom. Small and densely populated CeO x nanoclusters and high-surface-area SiO 2 (or other types of supports) are critical to developing stable single-atom or cluster catalysts for practical applications. Extended Data Fig. 3g,i show XPS data characterizing the as-synthesized 0.4 wt% Pt/CeO x /SiO 2 catalyst. The survey spectra show all the elements present in the as-synthesized catalyst. Platinum was not detectable owing to the low loading level (~0.05 wt% with respect to the total weight of the catalyst). The Cl2p peak (~199 eV) was not detectable either in the survey or in the local scan spectrum, suggesting that the vigorous washing and high-temperature calcination eliminated the Cl residues. The selective Pt adsorption and the subsequent processes did not appreciably change the nature and amounts of defects in the CeO x nanoclusters.
Extended Data Fig. 4a-c show HAADF images of the as-synthesized 0.4 wt% Pt/CeO x /SiO 2 catalyst, confirming the absence of detectable Pt clusters. Considering the relatively high atomic number of Ce and rapid changes in thickness across ultrasmall CeO x nanoclusters, it is not possible to unambiguously distinguish Pt single atoms from Ce atoms in low-dose HAADF images. High-current electron probes damage the crystallinity of the CeO x nanoclusters. Extended Data Fig. 4d shows an HAADF image of a Pt/CeO x /SiO 2 sample by the impregnation method. The Pt loading level was controlled to be ~1.5 wt% with respect to the total support (CeO x plus SiO 2 ). This particular catalyst contained various Pt species: NPs (red square), clusters (white square) linked to the CeO x nanoclusters, and subnanometre clusters (white circle) attached to the CeO x nanoislands. Although Pt single atoms could not be reliably identified, subnanometre Pt clusters are easily distinguishable from the CeO x nanoislands. The insets show Pt clusters (≥0.4 nm) attached to CeO x nanoislands. By analysing numerous such atomic-resolution HAADF images, we concluded that Pt clusters with sizes of ~0.4 nm or larger could be unambiguously identified. Single Pt atoms or tiny, weakly bonded Pt clusters (for example, with sizes <0.3 nm) would not be reliably identified in low-dose atomic-resolution HAADF images of Pt/CeO x /SiO 2 catalysts.
Extended Data Fig. 6 reports data characterizing the dynamic behaviour of Pt atoms on CeO 2 or SiO 2 during reduction in H 2 . A model coreshell structure consisting of Cu[(Pt/CeO 2 )@SiO 2 ] was fabricated to allow tracking of Pt atoms diffusing through the porous SiO 2 shell-the strategy involves incorporating Cu NPs on the outer shell surface to trap (scavenge) Pt atoms that might migrate from the CeO 2 in the interior to-and through-the SiO 2 shell to reach the Cu NPs, where they would form Cu-Pt alloy NPs (Extended Data Fig. 6a,c) 58 . The Cu-Pt/SiO 2 was synthesized by depositing 4 wt% Cu NPs onto 0.3 wt% Pt/SiO 2 . Prior to collecting CO adsorption DRIFTS, all samples were in situ reduced in 10% H 2 /Ar at 400 °C for 3 h. After reduction of the Cu-Pt/SiO 2 , the CO adsorption peak characterizing Pt clusters/NPs disappeared, and a new broad peak appeared at approximately 2,012 cm −1 , indicating formation of isolated Pt atoms in Pt-Cu alloy 58 . In a control sample, Cu[(Pt/SiO 2 ) @SiO 2 ], Pt atoms were observed to form Cu-Pt alloy NPs, having migrated during the H 2 -reduction process (400 °C for 3 h) through the porous SiO 2 shell (Extended Data Fig. 6d-h). By contrast, when the Cu[(Pt/ CeO 2 )@SiO 2 ] sample was reduced by the same treatment process, Pt atoms were not associated with the Cu NPs (Extended Data Fig. 6i-l), confirming that a much stronger interaction of Pt atoms with CeO 2 than with SiO 2 suppressed their diffusion through the porous SiO 2 shell. The results of this set of experiments demonstrate that, when the interaction strength of Pt atoms with CeO 2 (CeO x ) is much stronger than that with SiO 2 , the migration of Pt atoms onto the SiO 2 surfaces is greatly mitigated, even under conditions of H 2 activation processes, although the Pt atoms still sinter on the CeO 2 (CeO x ) surfaces.
Extended Data Fig. 7e Fig. 7k-p). Extended Data Fig. 7j shows that Pt clusters were observed on the reduced 0.02 wt% Pt/CeO 2 (synthesized by SEA method). The number density of Pt atoms in this 0.02 wt% Pt/CeO 2 sample was estimated to be ~0.007 nm −2 , much lower than that of the 0.4 wt% Pt/CeO x /SiO 2 catalyst (0.036 nm −2 on CeO x ). Even with such a low number density of Pt atoms, Pt clusters/NPs were still formed during the reduction process. These experimental results confirm that ultrasmall size and isolation of the CeO x nanoglue islands are critical to confining the movement of Pt atoms. Extended Data Fig. 8a-c show HAADF images of the 0.4 wt% Pt/CeO 2 NPs/SiO 2 treated in H 2 at 300 °C for 1 h. This catalyst had a Pt loading level similar to that of the 0.4 wt% Pt/CeO x /SiO 2 , but the sizes of the CeO 2 crystals were markedly greater (~8 nm from HAADF images). Extended Data Fig. 8a,b show the same sample region but was obtained with a different electron beam defocus value to illustrate that each CeO 2 NP is associated with at least one Pt cluster. In examinations of numerous HAADF images, no Pt clusters were found on the bare SiO 2 support surfaces, indicating that all the Pt atoms were associated with the CeO 2 crystals. The results of this set of experiments clearly demonstrate that during the H 2 -activation process Pt single atoms on each individual CeO 2 NP sintered to form small clusters but did not migrate away from their own CeO 2 NPs to form large Pt NPs.
Extended Data Fig. 8g-j show Pt clusters on CeO x /SiO 2 . When the Pt loading level was increased to 4 wt%, a majority of the CeO x nanoglue islands contained several Pt atoms, and during the H 2 -reduction process these Pt atoms sintered to form small clusters. The fact that the sizes of the Pt clusters are similar (average size ~0.9 nm) suggests that (1) the Pt atoms on each CeO x nanocluster sintered after treatment in H 2 at 400 °C for 5 h and (2) even after H 2 reduction at 500 °C for 12 h, the Pt atoms were still confined to their own CeO x nanoclusters. The results of this set of experiments unambiguously demonstrate that the CeO x nanoglue islands strongly localized the Pt atoms even under harsh reduction conditions. Our design strategy of localizing Pt atoms or clusters by confining their movement during a catalyst activation process or a desired catalytic reaction thus proves to be successful. Such a strategy enhances both the catalyst's stability and activity and/or selectivity. The selection of an appropriate functional nanoglue is critical to enhancing both the catalyst's performance and stability.
Extended Data Fig. 9i-p show that the H 2 -reduction treatment at 300 °C not only reduced the oxidation state of the Pt atoms but also modified the local bonding environment of the platinum species. In CO oxidation under lean conditions (O 2 partial pressures in excess of the stoichiometric O 2 :CO ratio of 0.5), the probe reaction reported in this work, the H 2 -activated 0.4 wt% Pt 1 /CeO x /SiO 2 was active and stable at low temperatures (<200 °C), but at higher temperatures (for example, 300 °C) partial oxidation of the Pt 1 atoms can occur and reduce the catalyst's activity. This process is reversible: the activity is recovered by H 2 reduction. Nonetheless, it remains a considerable challenge to maintain the Pt 1 in a reduced state over a wider temperature window (>200 °C) for lean oxidation applications. Further work is needed to overcome this limitation.
Extended Data Fig. 10 shows results of a CO-DRIFTS investigation of Pd and Rh atoms supported on CeO x /SiO 2 and CeO 2 . Although the CO adsorption experiment was conducted at 25 °C, CeO 2 -supported Pd atoms were rapidly reduced by CO and aggregated into Pd clusters/ NPs (Extended Data Fig. 10b). The very small amount of bridge-bonded CO 60,61 (characterized by the band at ~1,978 cm −1 ) in the as-synthesized Pd/CeO x /SiO 2 might be evidence of sintering of Pd atoms on their own CeO x nanoglue islands (Extended Data Fig. 10a). Compared with the notable sintering of Pd atoms on the Pd/CeO 2 (Extended Data Fig. 10d), the Pd atoms on CeO x /SiO 2 were extremely stable after reduction in H 2 (Extended Data Fig. 10c). The CO-DRIFTS experiments characterizing the Rh-containing catalysts were conducted at 180 °C to avoid oxidative fragmentation in which exposure of Rh NPs/clusters to CO at temperatures lower than 150 °C leads to oxidation and fragmentation (assisted by support OH groups) that constitutes redispersion of Rh NPs/clusters into cationic single-atom Rh (ref. 62 ). After the reduction treatment, the Rh atoms remained atomically dispersed on the CeO x /SiO 2 support (Extended Data Fig. 10e). On the other hand, after the reduction treatment, the presence of the 2,040 cm −1 peak characterizing the reduced Rh/CeO 2 catalyst could be assigned to CO adsorbed on Rh NPs 63 (Extended Data Fig. 10f). Although more investigations of the intrinsic structure and catalytic properties of the Pd/ CeO x /SiO 2 and Rh/CeO x /SiO 2 catalysts are recommended, this set of experimental results clearly demonstrates that the use of functional nanoglues to localize metal atoms and/or clusters works equally well for atomically dispersed Pd and Rh as for Pt. Our metal-atom localization strategy involving grafted ultrasmall functional nanoglue islands on high-surface-area, robust supports is general and can be extended to a plethora of catalyst systems for chemical transformations of important molecules.

Data availability
All data that led us to understand the results presented here are available with the Article or from corresponding author J.L. upon reasonable request. Source data are provided with this paper. component. e, Plots of attainable wt% Pt loading (with respect to SiO 2 ) versus the specific surface area of SiO 2 for various sizes of CeO x nanoclusters and the distance between them, assuming that each CeO x nanocluster hosts only one Pt atom. D, average size (diameter) of CeO x nanoclusters; L, average distance between the edges of CeO x nanoclusters. Experimental parameters characterizing the 0.4 wt% Pt/CeO x /SiO 2 were used for the plot (red). f, Plots of attainable wt% Pt loading (with respect to SiO 2 ) versus specific surface area of SiO 2 for CeO x nanoclusters (D = 2 nm, L = 9 nm) reported in this work. N, number of Pt atoms on each CeO x nanocluster. The red, black and blue lines represent 1, 2 and 5 Pt atoms, respectively, on each CeO x nanocluster. g,h,i, XPS data characterizing the as-synthesized 0.4 wt% Pt/CeO x /SiO 2 catalyst.