Introduction

Catalysis lies at the heart of transformative innovations in sustainable chemical synthesis and energy conversion technologies. Single-atom catalysts (SACs) have emerged as a powerful platform due to their maximal atom utilization efficiency and highly tunable electronic properties1,2,3,4. However, the intrinsic single-site nature limits their efficacy in complex chemical reactions that require cooperative multi-site interactions5,6,7,8,9,10. To address this challenge, heterogeneous catalytic pairs (CPs) comprising two adjacent active atoms offer a promising solution, which can enable cooperative multiple substrate adsorption/activation, and thereby unlock unconventional reaction pathways11,12,13,14. For example, heteronuclear Ir1-P1 CPs supported on carbon exhibit outstanding hydrogen oxidation activity, arising from the spontaneous H2 dissociation on Ir and the cooperative *H desorption by *OH generated on oxophilic P via spontaneous water dissociation15. Similarly, heteronuclear Rb-Ir CPs embedded in a poly(heptazine imide) framework achieve selective photooxidation of glycerol to hydroxypyruvic acid16. Despite these advances, homonuclear CPs remain largely unexplored, particularly regarding the impact of inter-CP spacing and coordination asymmetry on their catalytic performance. A fundamental understanding of these factors is critical to unlock the full potential of CPs, enabling enhanced catalytic efficiencies and novel reaction pathways.

In this work, we prepare atomically precise homonuclear Pt1-Pt1 CPs with asymmetric Pt1C3-Pt1O1C3 coordination, anchored on reduced graphene oxide (termed as Pt1C3-Pt1O1C3/rGO). By finely tuning the inter-distance between the Pt1C3-Pt1O1C3 CPs to an optimal ~5.3 Å, we achieve a turnover frequency (TOF) of 27,218 h-1 for transfer hydrogenation of azobenzene, exceeding benchmarking catalysts by more than an order of magnitude. First-principles calculations reveal that the asymmetric Pt1C3-Pt1O1C3 CPs separated by a distance of 5.3 Å modulate the electronic structure of the Pt sites, facilitate the co-adsorption of sterically hindered intermediates, and enable an energetically favorable hydrogen shuttling and hydrogenation pathway. These synergistic effects collectively enhance the overall transfer hydrogenation efficiency.

Results and discussion

Synthesis and characterization of Pt1C3-Pt1O1C3/rGO

The Pt1C3-Pt1O1C3/rGO catalysts with tunable Pt loadings (0.5–7.5 wt.%) were synthesized by an interface confinement method, leveraging the diatomic structure of the Pt precursor and SiO2-induced physical confinement17. Specifically, the (ethylenediamine)iodoplatinum(II) dimer dinitrate precursor was adsorbed onto graphene oxide (GO), followed by amorphous SiO2 layer deposition via tetraethyl orthosilicate hydrolysis and thermal treatment. NaOH etching then removed the SiO2 layer, yielding well-dispersed Pt1-Pt1 CPs on rGO. X-ray diffraction (XRD) analysis (Supplementary Fig. 1) confirmed the reduction of GO to rGO, while atomic force microscopy (Supplementary Fig. 2) demonstrated that Pt1C3-Pt1O1C3/rGO retained a two-dimensional (2D) morphology with atomic thickness.

Aberration-corrected high-angle annular dark-field scanning transmission electron microscope (HAADF-STEM) images (Fig. 1a–j) revealed atomic dispersion of Pt atoms (bright spots) on rGO surface at varying Pt loadings (0.5 wt.%, 1.0 wt.%, 2.5 wt.%, 5.0 wt.%, and 7.5 wt.%, as confirmed by inductively coupled plasma-optical emission spectroscopy), with no observable formation of Pt clusters or nanoparticles (NPs). Magnified HAADF-STEM images (Fig. 1k–o) clearly confirmed the presence of Pt dimers even at low Pt loadings (Fig. 1k), highlighted by solid orange boxes. As the Pt loading increased, the density of Pt dimers correspondingly rose, accompanied by a noticeable reduction in the inter-distance between Pt dimers (the inter-distance is defined as the shortest distance between two Pt atoms from adjacent Pt dimers, namely each Pt atom belongs to a different Pt dimer unit). Notably, the minimal inter-distance between Pt dimers significantly decreased from ~18.1 Å at 0.5 wt.% Pt loading to ~4.6 Å at 7.5 wt.% Pt loading (Fig. 1p–t).

Fig. 1: Structural characterization of Pt1C3-Pt1O1C3/rGO samples.
Fig. 1: Structural characterization of Pt1C3-Pt1O1C3/rGO samples.
Full size image

Low-magnification HAADF-STEM images of Pt1C3-Pt1O1C3/rGO with Pt loadings of 0.5 wt.% (a), 1.0 wt.% (b), 2.5 wt.% (c), 5.0 wt.% (d), and 7.5 wt.% (e). f–j High-magnification HAADF-STEM images highlighting atomically dispersed Pt sites. k–o Magnified HAADF-STEM images with orange boxes indicating representative Pt1C3-Pt1O1C3 pairs. p–t Statistical distribution of the shortest distances (d) between adjacent Pt1C3-Pt1O1C3 pairs for each sample. u–w A representative Pt1-Pt1 CP on rGO imaged by HAADF-STEM with corresponding EELS spectra. x Atomic-scale charge-density mapping by 4D-STEM. Left: HAADF-STEM image reconstructed from the 4D-STEM dataset, resolving a Pt1-Pt1 pair on rGO. Right: Projected charge-density distribution derived from the inverted divergence of the center of mass (dCoM), highlighting the spatial positions of C, O, and Pt atoms. y EDS mapping of Pt1C3-Pt1O1C3/rGO catalysts.

To elucidate the coordination environment of Pt1-Pt1 CPs, atomic-resolution electron energy-loss spectroscopy (EELS) and 4D-STEM were performed. EELS measurements of an individual Pt1-Pt1 CP simultaneously detected the C-K, O-K, and Pt-M signals, indicating that Pt atoms are coordinated with both C and O (Fig. 1u–w). Complementary 4D-STEM charge-density mapping revealed two adjacent positively charged Pt centers, each coordinated by negatively charged atoms (Fig. 1x). Strikingly, the coordinating atom adjacent to the lower-left Pt displayed a markedly stronger negative charge density than other Pt-coordinating atoms, consistent with the Pt1C3-Pt1O1C3 pair owing to the higher electronegativity of O relative to C. In addition, HAADF-STEM images reconstructed from 4D-STEM datasets consistently resolved abundant Pt1-Pt1 CPs (Supplementary Fig. 3). A small fraction of isolated single Pt atoms was also observed in the HAADF-STEM images, likely due to pair dissociation or projection overlap in the 2D images18. At high Pt loadings (≥8.5 wt.%), small crystalline Pt NPs began to appear (Supplementary Fig. 4)19. Energy-dispersive X-ray spectroscopy (EDS) elemental mapping confirmed the homogeneous distribution of C, O, and Pt in the Pt1C3-Pt1O1C3/rGO catalysts (Fig. 1y).

To examine the electronic property of Pt, X-ray photoelectron spectroscopy (XPS) was performed20,21. All Pt1C3-Pt1O1C3/rGO samples exhibited similar Pt 4f7/2 and 4f5/2 peaks at ~73.0 eV and ~76.3 eV, indicating a stable Pt2+ oxidation state (Fig. 2a and Supplementary Fig. 5)22,23,24. In contrast, Pt NPs on rGO (PtNPs/rGO), synthesized via wet-chemical reduction of H2PtCl4 (Supplementary Figs. 6 and 7), displayed zero-valent Pt states with Pt 4f7/2 and 4f5/2 peaks at 71.4 eV and 74.7 eV, respectively. To further determine the Pt atomic coordination in Pt1C3-Pt1O1C3/rGO, X-ray absorption spectroscopy (XAS) measurements were conducted. The normalized Pt L3-edge X-ray absorption near-edge structure (XANES) spectra showed that the white-line of all Pt1C3-Pt1O1C3/rGO samples was positioned between that of Pt foil and PtO2 (Fig. 2b), indicating the oxidative state of Pt in Pt1C3-Pt1O1C3/rGO samples, in good alignment with the XPS results25. Fourier-transformed (FT) k3χ(k) extended X-ray absorption fine structure (EXAFS) spectra for Pt foil, PtO2, and Pt2 precursor revealed dominant peaks at 2.48 Å, 1.68 Å, and 2.45 Å, corresponding to Pt-Pt, Pt-O, and Pt-I first-shell scattering, respectively (Fig. 2c)26,27,28. In contrast, Pt1C3-Pt1O1C3/rGO samples showed only one prominent peak at around 1.56 Å, which could be attributed to the Pt-C/O first-shell scattering, suggesting that Pt atoms were atomically dispersed on the rGO support, matching well with the HAADF-STEM observations.

Fig. 2: XPS, Pt L3-edge XANES, and EXAFS analysis of Pt1C3-Pt1O1C3/rGO samples.
Fig. 2: XPS, Pt L3-edge XANES, and EXAFS analysis of Pt1C3-Pt1O1C3/rGO samples.
Full size image

a XPS spectra of PtNPs/rGO and Pt1C3-Pt1O1C3/rGO samples with varying Pt loadings. b Pt L3-edge XANES spectra of Pt foil, PtO2, and Pt1C3-Pt1O1C3/rGO samples at 0.5 wt.%, 2.5 wt.%, and 7.5 wt.% Pt loading. c FT-EXAFS spectra of Pt foil, PtO2, and Pt1C3-Pt1O1C3/rGO samples at 0.5 wt.%, 2.5 wt.%, and 7.5 wt.% Pt loading. d Comparison of experimental and simulated XANES spectrum of Pt1C3-Pt1O1C3/rGO. e k3-weighted FT-EXAFS spectrum of Pt1C3-Pt1O1C3/rGO with fitting result. f The most plausible atomic configuration of Pt1C3-Pt1O1C3/rGO (black: C, red: O, and blue: Pt).

To determine the exact atomic structure of Pt1C3-Pt1O1C3/rGO, we proposed several possible models informed by the above results (Supplementary Fig. 8). The most plausible structure consists of two Pt atoms anchored on adjacent benzene rings in the rGO network (Fig. 2d–f). Specifically, one Pt atom coordinates with three carbon atoms (Pt1C3 moiety), while the other coordinates with three carbon atoms and one oxygen atom (Pt1O1C3 moiety), forming an asymmetric paired Pt1C3-Pt1O1C3 configuration. This assignment is corroborated by the charge-density distribution observed in 4D-STEM (Fig. 1x). The average Pt-C/O bond length (approximately 2.11 Å) and the coordination number (around 3.5) of this model agree well with the EXAFS fitting results (Supplementary Table 1). Notably, the average distance between adjacent Pt atoms in a Pt1C3-Pt1O1C3 pair is approximately 5.0 Å, aligning closely with the statistical data derived from HAADF-STEM imaging. All other proposed structural models failed to replicate the key experimental features, as evidenced by discrepancies in their XANES simulation spectra (Supplementary Fig. 9) and variations in bond lengths and coordination numbers (Supplementary Table 2).

In summary, converging evidence from precursor-directed synthesis, HAADF-STEM, atomic-resolution EELS, 4D-STEM charge-density mapping, electron localization function (ELF) calculation (Supplementary Fig. 10), and XAS, together with XANES simulations and EXAFS fittings, unequivocally establishes the asymmetric Pt1C3-Pt1O1C3 pair. For clarity, we also present a merged comparison of the atomic structure, 4D-STEM charge-density mapping, and ELF analysis (Supplementary Fig. 11), which supports the proposed Pt1C3-Pt1O1C3 configuration. Notably, charge-transfer analysis reveals a charge difference of only ~0.07 e- between the two Pt sites, below the threshold for a detectable XPS shift and consistent with prior reports15,29. This is corroborated by ELF calculation and 4D-STEM charge-density mapping results, both of which demonstrate two adjacent Pt centers in comparable charge environments, confirming the persistence of a uniform Pt2+ oxidation state in Pt1C3-Pt1O1C3/rGO.

Hydrogenation of azobenzene

The transfer hydrogenation of azobenzene (C6H5N=NC6H5) to hydrazobenzene (C6H5NH-HNC6H5) via in-situ ammonia borane (NH3BH3) hydrolysis was conducted to evaluate the catalytic performance of Pt1C3-Pt1O1C3/rGO in complex transformations (Fig. 3a). The overall stoichiometric reaction involves one NH3BH3 molecule, three C6H5N=NC6H5 molecules, and three H2O molecules, yielding three C6H5NH-HNC6H5 products30. First of all, we optimized the reaction conditions, including solvent, temperature, and reaction time, and revealed methanol as the optimal solvent at a reaction temperature of 80 °C (Supplementary Figs. 12 and 13). As shown in Fig. 3b, at a constant total Pt mass, increasing the Pt loading of Pt1C3-Pt1O1C3/rGO from 0.5 wt.% to 5.0 wt.% leads to a pronounced enhancement in the yield of C6H5NH-HNC6H5, rising from 26.6% to as high as 99.8%. However, at a Pt loading of 7.5 wt.%, the yield decreased to 33.6%. It is found that the TOF value strongly depends on the inter-distance between the adjacent Pt1C3-Pt1O1C3 pairs, peaking at ~5.3 Å with a maximum TOF of 27.2 × 103 h–1 (Fig. 3c). On the other hand, both rGO and conventional PtNPs/rGO (5.0 wt.%) catalysts displayed poor performance in hydrogenation of azobenzene (Fig. 3d). Quantitative comparison of TOF and yield with previously reported catalysts further highlights the superior activity of the Pt1C3-Pt1O1C3/rGO (Supplementary Table S3). Besides excellent catalytic activity, the Pt1C3-Pt1O1C3/rGO catalyst also showed outstanding catalytic stability, which could maintain high catalytic activity throughout 10 hydrogenation cycles with negligible activity decay (Fig. 3e). The Pt1C3-Pt1O1C3/rGO morphology was well preserved after 10 cycles of hydrogenation of azobenzene (Supplementary Figs. 14 and 15), highlighting the structural stability of the catalyst.

Fig. 3: Catalytic performance.
Fig. 3: Catalytic performance.
Full size image

a Schematic representation of the azobenzene hydrogenation reaction. b Catalytic yield of hydrazobenzene as a function of Pt loading in Pt1C3-Pt1O1C3/rGO. c Comparison of TOF values for different Pt1C3-Pt1O1C3/rGO catalysts with varying inter-pair distances. d Yield comparison of Pt1C3-Pt1O1C3/rGO catalyst (5.0 wt.% Pt) with reference catalysts, including PtNPs/rGO and rGO, for azobenzene hydrogenation. The error bars in (b, d) represent the standard deviation of three independent technical replicates (n = 3). e Stability of Pt1C3-Pt1O1C3/rGO catalyst (5.0 wt.% Pt) over 10 catalytic cycles, measured under identical reaction conditions. f Substrate scope for Pt1C3-Pt1O1C3/rGO catalyst (5.0 wt.% Pt).

Besides azobenzene, a broad range of azobenzene derivatives could be hydrogenated via in situ NH3BH3 hydrolysis, producing the corresponding products with high efficiency (Fig. 3f). The reaction exhibited excellent tolerance for various substituents, including halogen and alkyl groups at different positions, yielding hydrazobenzene derivatives in remarkable yields (Supplementary Figs. 16–25). This broad applicability highlights the versatility of the Pt1C3-Pt1O1C3/rGO catalyst for catalyzing the hydrogenation reactions.

Theoretical studies

To provide a deep understanding of the transfer hydrogenation of azobenzene taking place on Pt1C3-Pt1O1C3, we performed ab initio density functional theory calculations. The Pt1C3-Pt1O1C3 motif was embedded into four defective graphene supercells to simulate different Pt1C3-Pt1O1C3/rGO catalysts with varying Pt1C3-Pt1O1C3 densities. These models, labeled as Models 1-4 (Supplementary Fig. 26), were assigned to the Pt1C3-Pt1O1C3/rGO catalysts with Pt loading of 1.0 wt.%, 2.5 wt.%, 5.0 wt.%, and 7.5 wt.%, respectively, based on the close agreement between the theoretical inter Pt1C3-Pt1O1C3 pair spacings and the experimental values obtained from HAADF-STEM (Fig. 1q–t).

We first computed the projected density of states and the d-band center for each Pt in Pt1C3-Pt1O1C3 across different structures (Fig. 4a and Supplementary Fig. 27). At an inter-spacing between adjacent Pt1C3-Pt1O1C3 pairs of 9.38 Å, both Pt sites show similar d-band centers at around −1.11 eV. As the inter-pair distance decreases from 7.24 Å to 5.45 Å, the d-band center of Pt1C3 increases to −0.04 eV, while that of Pt1O1C3 remains at −1.37 eV. At an inter-pair distance of 5.23 Å, this trend reverses, with Pt1O1C3 showing a slightly higher d-band center (−0.33 eV) than Pt1C3 (−0.62 eV). Notably, Model 3 shows a highest d-band center of −0.04 eV at the Pt1C3 site, indicative of an optimal electronic structure for reactant adsorption31,32,33, as confirmed by adsorption energy calculations (Fig. 4b and Supplementary Fig. 28). Specifically, the Pt site with a Pt1C3 configuration for Model 3 displays an optimal adsorption energy of −1.53 eV for NH3BH3 and −0.43 eV for C6H5N=NC6H5. H2O adsorption is moderate at both sites, indicating that water does not severely compete with the main reactants for Pt sites (Supplementary Table S4).

Fig. 4: Theoretical analysis.
Fig. 4: Theoretical analysis.
Full size image

a Variation of d-band center for Pt sites as a function of inter-pair distance between adjacent Pt1C3-Pt1O1C3 CPs. b Calculated adsorption energies of NH3BH3 and C6H5N=NC6H5 on Pt1C3 as a function of inter-pair distance between adjacent Pt1C3-Pt1O1C3 CPs. c Charge density difference mapping. Orange and purple region indicates charge depletion and charge accumulation, respectively. d Reaction pathway and energy profile for hydrogenation of C6H5N=NC6H5 to C6H5NHHNC6H5 on Pt1C3-Pt1O1C3/rGO catalyst with Model 3 configuration. e Volcano plot correlating azobenzene adsorption with the reaction energy of the first N=N hydrogenation on various Pt-based catalysts.

Further analysis reveals that local structural distortion is the primary cause of varying electronic properties among different models. As shown in Supplementary Fig. 26, the Pt atom in Pt1C3 for Model 3 exhibits the greatest protrusion, driven by the balance between Pt density and internal energy. Bader charge and charge differential analyses reveal that 0.3 e- can be transferred from NH3BH3 to Pt1C3 (Supplementary Fig. 29), while 0.1 e- can be transferred from C6H5N=NC6H5 to Pt1C3 site (Supplementary Fig. 30).

These findings demonstrate that the inter-distance between adjacent Pt1C3-Pt1O1C3 pairs and the asymmetric PtC3-PtO1C3 configuration can synergistically modulate the electronic properties of Pt sites. The Pt1C3-Pt1O1C3/rGO catalyst with Model 3 configuration not only offers spatially optimized sites for co-adsorption of sterically hindered C6H5N=NC6H5 and NH3BH3 (Fig. 4c and Supplementary Fig. 31), but also ensures optimal adsorption strength for these reactants. The azobenzene hydrogenation pathway on Pt1C3-Pt1O1C3/rGO catalyst with Model 3 configuration was further studied by density functional theory (DFT), which involved the following three key steps (Supplementary Fig. 32): 1. Transfer hydrogenation of C6H5N=NC6H5 by NH3BH3 hydrolysis, producing one C6H5NHHNC6H5 and leaving NH3BH1* at the Pt1C3 site (from co-adsorption to intermediate V). 2. Transfer hydrogenation with NH3BH1* and H2O to form NH3BOH* and a second C6H5NHHNC6H5 (from intermediate VI to XII). 3. Transfer hydrogenation with H2O as the proton donor to form NH3B(OH)3* and a third C6H5NHHNC6H5 (from intermediate XIII to final desorption). The overall pathway and associated energy barrier are clearly depicted in Fig. 4d, as well as in Supplementary Figs. 33 and 34. Additionally, an alternative pathway was also considered, in which the azo group in azobenzene bonded directly to the Pt1C3 site (Supplementary Fig. 35). However, this pathway was energetically unfavorable due to the strong co-adsorption strength and a high energy barrier of 1.08 eV for step III.

The reaction coordinate simulations identified the first H abstraction as the rate-determining step with the highest energy barrier of 1.73 eV (step I). Climbing nudged elastic band (cNEB) simulations further suggested that the first H abstraction process could be broken into two steps by introducing a new H* intermediate at a neighboring C atom (Supplementary Fig. 36)34. By adopting this two-step H abstraction mechanism, initialized with co-adsorption of NH3BH3 and C6H5N=NC6H5, the first H atom from BH3 group of NH3BH3 transferred to a neighboring C atom to form a new H* intermediate with an energy barrier of 0.83 eV. Subsequently, the H* intermediate at the C site was transferred to the Pt atom at the Pt1O1C3 site with an energy barrier of 0.90 eV. The reaction barrier heights were predicted to be 1.98 eV and 0.93 eV for TS3 and TS4, respectively. Importantly, NH3BH3 hydrolysis was also evaluated on Pt1C3-Pt1O1C3/rGO with different Pt loadings (Supplementary Table S5), and Model 3 was found to catalyze NH3BH3 hydrolysis and generate active H* intermediates with barriers comparable to those for the first hydrogenation step of azobenzene.

Comparatively, we have conducted theoretical calculations on other Pt-based catalysts for transfer hydrogenation of azobenzene, including Pt1C3-Pt1C3/rGO, Pt1O1C3-Pt1O1C3/rGO, Pt1/rGO, and Pt NPs (Supplementary Fig. 37). In Pt1C3-Pt1C3/rGO, where all Pt sites adopt the symmetric Pt1C3 configuration, and the dissociated H atoms bind strongly to the Pt sites, with an adsorption energy of −8.26 eV (Supplementary Fig. 38). This strong binding inhibits H-atom transfer, making Pt1C3-Pt1C3/rGO ineffective for hydrogenating C6H5N=NC6H5. Pt1O1C3-Pt1O1C3/rGO exhibits universally weak adsorption for azobenzene, NH3BH3, and H2O (Supplementary Table S6). Pt1/rGO with Pt1O1C3 configuration exhibits weak binding to NH3BH3, with an adsorption energy of −0.15 eV (Supplementary Fig. 39). Pt NPs are suboptimal due to the excessive adsorption strength of NH3BH3 (−2.16 eV) and C6H5N=NC6H5 (−2.20 eV) (Supplementary Fig. 40), leading to catalyst poisoning. We further established a volcano relationship by correlating ΔEads(azobenzene) with the reaction energy of the first N=N hydrogenation for various models (Fig. 4e). In this plot, Pt (111) resides on the over-binding branch, whereas Pt1C3-Pt1O1C3/rGO (Models 1, 2, and 4), Pt1O1C3-Pt1O1C3/rGO, Pt1C3/rGO, and Pt1O1C3/rGO fall on the weak-binding branch. The first hydrogenation of azobenzene becomes energetically disfavored for Pt1C3-Pt1C3/rGO. By contrast, the Pt1C3-Pt1O1C3/rGO (Model 3) sits at the apex of the volcano plot, exhibiting a balanced adsorption strength coupled with the lowest reaction energy for the first hydrogenation step. These calculation results highlight the crucial asymmetric paired PtC3-PtO1C3 configuration with an optimized inter-pair spacing, which can effectively promote the transfer hydrogenation of azobenzene.

Discussion

In summary, we have successfully prepared atomically dispersed asymmetric Pt1C3-Pt1O1C3 CPs anchored on rGO. By precisely tuning the inter-spacing between the adjacent Pt1C3-Pt1O1C3 CPs to approximately 5.3 Å, the Pt1C3-Pt1O1C3/rGO catalyst achieves an optimal performance in transfer hydrogenation of azobenzene via ammonia-borane hydrolysis. The Pt1C3-Pt1O1C3 CPs separated by 5.3 Å precisely modulate the local electronic environment of each Pt site and enable the co-adsorption of sterically hindered C6H5N=NC6H5 and NH3BH3 molecules. Furthermore, the unique asymmetric paired Pt1C3-Pt1O1C3 configuration facilitates hydrogen abstraction and transfer. Our work provides a deeper understanding of the structure-performance relationship in CPs and opens new avenues for rationally designing advanced catalysts for highly efficient chemical transformations.

Methods

Materials

(Ethylenediamine)iodoplatinum(II) dimer dinitrate (96%, Sigma-Aldrich), chloroplatinic acid hexahydrate (H2PtCl6·6H2O, 99.9%, Adamas), graphene oxide (GO, 0.5–5 µm, XFNANO), tetraethyl orthosilicate (TEOS, SiO2 > 28.4%, Sigma-Aldrich), ammonia solution (25–28%, Sigma-Aldrich), sodium hydroxide (NaOH, 98%, Aladdin), sodium borohydride (NaBH4, 98%, Energy chemical), azobenzene (97%, Macklin), borane-ammonia complex (NH3BH3, 97%, Macklin), and other reagents were purchased from commercial suppliers and used without further purification.

Synthesis of Pt1C3-Pt1O1C3/rGO catalysts

One hundred milligrams of GO was dispersed in 50 mL of deionized water through sonication, followed by the addition of 200 mL of ethanol. A specific amount of Pt2 precursor was dispersed in 10 mL of ethanol and subsequently added to the GO ethanol solution. The mixture was stirred at room temperature for 12 h. The TEOS (2.75 mL in 25 mL ethanol) was then added dropwise to the mixture and stirred for an additional 13 h. The resulting product was centrifuged, washed thoroughly with water and ethanol, and dried under vacuum. The dried powder was calcined at 400 °C under N2 atmosphere for 2 h. Finally, silicon species in the sample were removed by etching with 3 M NaOH solution at room temperature for 3 h, yielding the final Pt1C3-Pt1O1C3/rGO catalyst. The Pt loading in the Pt1C3-Pt1O1C3/rGO catalyst was controlled by varying the amount of Pt2 precursor used in the synthesis. The Pt loadings in catalysts were confirmed by inductively coupled plasma-optical emission spectroscopy.

Synthesis of PtNPs/rGO catalysts

PtNPs/rGO catalysts were synthesized using a procedure similar to that of Pt1C3-Pt1O1C3/rGO, with H2PtCl6·6H2O as the Pt precursor and NaBH4 as the reducing agent. Specifically, an amount of H2PtCl6·6H2O was dispersed in 10 mL of ethanol and subsequently added to the ethanol solution containing GO. The mixture was stirred at room temperature for 13 h, followed by the addition of NaBH4 as the reducing agent. The subsequent steps, including washing, drying, calcination, and etching with 3 M NaOH to remove silicon species, were identical to those used for Pt1C3-Pt1O1C3/rGO synthesis. The final product was denoted as PtNPs/rGO.

The hydrogenation reactions

Reaction conditions for azobenzene and derivative hydrogenation

In an oven-dried 10 mL Schlenk tube equipped with a magnetic stir bar, azobenzene (0.2 mmol) and ammonia borane (0.4 mmol) were sequentially added. The tube was sealed with a rubber septum, evacuated, and refilled with N2 gas. Then, 0.01 mol.% Pt1C3-Pt1O1C3/rGO catalyst was introduced. For example, in the case of Pt1C3-Pt1O1C3/rGO (5 wt.%), 8 mg of catalyst was ultrasonically dispersed in 50 mL of deionized water, and 0.5 mL of this dispersion, together with 1 mL of solvent (typically methanol), was transferred into the reaction tube. A nitrogen balloon was attached as shown in Supplementary Fig. 41.

The mixture was stirred at 80 °C for 22 min, then cooled to room temperature. The catalyst was separated by centrifugation, washed twice with ethyl acetate (2 × 5 mL), and the combined filtrate was dried over anhydrous Na2SO4. After filtration, the solvent was removed under reduced pressure at 40 °C using a rotary evaporator. The residue was dissolved in deuterated chloroform (CDCl3), and 1,1,2,2-tetrachloroethane (TCE) was added as an internal standard for nuclear magnetic resonance (NMR) analysis. Optimization studies of solvent, temperature, and time identified methanol at 80 °C as the optimal condition (Supplementary Figs. 12 and 13). Reactions involving substituted azobenzene derivatives were performed under identical conditions except that the reaction time was extended to 120 min. The crude products were purified by flash column chromatography on silica gel using hexane/ethyl acetate as the eluent to yield the corresponding hydrogenated products.

Catalytic performance evaluation

Catalytic performance was quantified using azobenzene hydrogenation over the Pt1C3-Pt1O1C3/rGO (5 wt.%) catalyst as a representative case. Product yields were determined from NMR data according to Eq. (1):

$$\begin{array}{c}{{{\rm{Yield}}}}(\%)=\frac{n\left({{{\rm{obtained}}}}\; {{{\rm{products}}}}\right)}{n\left({{{\rm{initally}}}}\; {{{\rm{added}}}}\; {{{\rm{substrates}}}}\right)}\\=\frac{\frac{{m}_{{{{\rm{TCE}}}}}\times 2\times 2.6}{{M}_{{{{\rm{TCE}}}}}\times 4}}{{n}\left({{{\rm{initally}}}}\; {{{\rm{added}}}}\; {{{\rm{substrates}}}}\right)}\\=\frac{\frac{25.75{{{\rm{mg}}}}\times 2\times 2.60}{167.85{{{\rm{g}}}}/{{{\rm{mol}}}}\times 4}}{0.2{{{\rm{mmol}}}}}=99.8\%\end{array}$$
(1)

The conversion and selectivity were calculated according to Eqs. (2) and (3), respectively:

$$\begin{array}{c}{{{\rm{Conversion}}}}(\%)=\left[1-\frac{n\,\left({{{\rm{remained}}}}\; {{{\rm{substrate}}}}\right)}{n\,\left({{{\rm{initial}}}}\right. \; {{{\rm{substrate}}}}}\right]\times 100\%\\=\left(1-0\right)=100\%\end{array}$$
(2)
$${{{\rm{Selectivity}}}}\left(\%\right) =\frac{n\,\left({{{\rm{obtained}}}}\; {{{\rm{product}}}}\right)}{n\,\left({{{\rm{initial}}}}\; {{{\rm{substrate}}}}\right)-n\,\left({{{\rm{remained}}}}\; {{{\rm{substrate}}}}\right)}\times 100\%\\ =\frac{\frac{{m}_{{{{\rm{TCE}}}}}\times 2\times 2.6}{{M}_{{{{\rm{TCE}}}}}\times 4}}{n\,\left({{{\rm{initial}}}}\; {{{\rm{substrate}}}}\right)-n\,\left({{{\rm{remained}}}}\; {{{\rm{substrate}}}}\right)}\\ =\frac{\frac{25.75{{{\rm{mg}}}}\times 2\times 2.60}{167.85\frac{{{{\rm{g}}}}}{{{{\rm{mol}}}}}\times 4}}{0.2{{{\rm{mmol}}}}-0}=99.8\%$$
(3)

The turnover frequency (TOF) was calculated at the 22-min time point from NMR spectra according to Eq. (4):

$$\begin{array}{c}{{{\rm{TOF}}}}=\,\frac{{{{\rm{moles}}}}\; {{{\rm{of}}}}\; {{{\rm{reactant}}}}\; {{{\rm{converted}}}}}{{{{\rm{mole}}}}\; {{{\rm{of}}}}\; {{{\rm{total}}}}\; {{{\rm{active}}}}\; {{{\rm{sites}}}}\times {{{\rm{reaction}}}}\; {{{\rm{time}}}}}\\=\frac{{{{\rm{amount}}}}\; {{{\rm{of}}}}\; {{{\rm{azobenzene}}}}({{{\rm{mol}}}})\,\times {{{\rm{conversion}}}}(\%)\,\times {{{\rm{selectivity}}}}(\%)}{{{{\rm{amount}}}}\; {{{\rm{of}}}}\; {{{\rm{catalyst}}}}({{{\rm{mol}}}})\,\times {{{\rm{reaction}}}}\; {{{\rm{time}}}}({{{\rm{h}}}})}\\=\frac{0.2{{{\rm{mmol}}}}\times 100\%\times 99.8\%}{0.0002{{{\rm{mmol}}}}\times 0.367{{{\rm{h}}}}}=27218\,{{{\rm{h}}}}^{-1}\end{array}$$
(4)

Characterization

Wide-angle X-ray diffraction (XRD) patterns were collected on a Bruker D8 Focus Powder X-ray diffractometer using Cu Kα radiation (40 kV, 40 mA) at room temperature. Aberration-corrected high-annular dark-field scanning transmission electron microscopy (HAADF-STEM) and energy-dispersive X-ray spectroscopy (EDS) mappings were collected on an aberration-corrected Thermo Fisher Spectra 300 system equipped with a cold field emission gun and an ASCOR probe corrector at 300 kV. The metal loadings in all the samples were measured by inductively coupled plasma-optical emission spectroscopy (Perkin Elmer Avio 500, UK). Atomic force microscopy (AFM) topography measurements were performed with an XE-100 AFM (Park Systems) in noncontact mode. The X-ray absorption near-edge structure (XANES) and the extended X-ray absorption fine structure (EXAFS) measurements of Pt L3-edge were carried out at the beamline of the Singapore synchrotron light source. Pt foil was used for the energy calibration, and all samples were measured under transmission mode at room temperature. Si (111) double crystal monochromator was used to filter the X-ray beam. EXAFS oscillations χ(k) were extracted and analyzed using the Demeter software package. Energy-loss spectroscopy (EELS) spectrum imaging was conducted using a Gatan Quantum ER spectrometer with an exposure time of 2 ms per pixel. Four-dimensional (4D) STEM datasets were acquired on a JEOL ARM200F microscope equipped with an ASCOR aberration corrector and a Merlin Medipix3 1 R pixelated detector (256 × 256 pixels, 6-bit depth), with a convergence semi-angle of approximately 21 mrad. Charge-density mapping was carried out via Gauss’s law by computing the divergence of the inverted electric-field distribution reconstructed from the 4D-STEM datasets using the open-source LiberTEM software35. To minimize long-range background features, a Gaussian filter with a kernel size of two pixels was applied36.

Theoretical calculations

The spin-polarized density functional theory (DFT) calculations were performed utilizing the Vienna Ab Initio Simulation Package (VASP, version 5.4.4)37,38. The generalized gradient approximation in the Perdew-Burke-Ernzerh format and the plane wave basis kinetic energy cutoff of 500 eV were used in all calculations39,40. The structural relaxations were carried out until the Hellmann-Feynman force on each atom was less than 0.01 eV/Å, and the energy convergence criterion was set to 10–4 eV. The gamma point sampling in the Brillouin zone was employed in all calculations. Along the direction perpendicular to the 2D graphene substrate sheet, a vacuum of 15 Å is included in the supercell to avoid the interaction of neighboring images. The reaction barrier was determined by the climbing-image nudged elastic band method34.