Synergizing metal–support interactions and spatial confinement boosts dynamics of atomic nickel for hydrogenations

Atomically dispersed metal catalysts maximize atom efficiency and display unique catalytic properties compared with regular metal nanoparticles. However, achieving high reactivity while preserving high stability at appreciable loadings remains challenging. Here we solve the challenge by synergizing metal–support interactions and spatial confinement, which enables the fabrication of highly loaded atomic nickel (3.1 wt%) along with dense atomic copper grippers (8.1 wt%) on a graphitic carbon nitride support. For the semi-hydrogenation of acetylene in excess ethylene, the fabricated catalyst shows extraordinary catalytic performance in terms of activity, selectivity and stability—far superior to supported atomic nickel alone in the absence of a synergizing effect. Comprehensive characterization and theoretical calculations reveal that the active nickel site confined in two stable hydroxylated copper grippers dynamically changes by breaking the interfacial nickel–support bonds on reactant adsorption and making these bonds on product desorption. Such a dynamic effect confers high catalytic performance, providing an avenue to rationally design efficient, stable and highly loaded, yet atomically dispersed, catalysts. Synergizing metal–support interactions and spatial confinement through atomic copper grippers boost the dynamics of highly loaded atomic nickel for high activity, high thermal/chemical stability and unprecedented coke inhibition in hydrogenation reactions.

S upported atomically dispersed catalysts (SADCs) have recently drawn great attention in numerous reactions for their maximized atom efficiency 1,2 . Exclusive site homogeneity, distinct energetics and spatial site confinement make SADCs very attractive in high selectivity and alleviated coke formation in the conversion of hydrocarbons [3][4][5] , sharply distinct from the low selectivity and heavy coke formation on metal nanoparticle (NP) catalysts 6 . However, owing to drastically increased surface free energies, the stability of atomically dispersed metal species under reaction conditions, especially at high metal loadings, has been the key issue to substantially limit the overall performance, restraining their practical applications 7 .
Realization of a highly loaded and stable SADC for hydrogenation reactions is particularly challenging, because atomically dispersed metal species often show higher mobility in reducing environments 8 . For instance, in the semi-hydrogenation of acetylene in excess ethylene, a crucial industrial process operated at a scale of billions of tons per annum to purify alkene streams 9,10 , severe metal aggregations occurred even for low-loaded SADCs, which sequentially generated heavy coke formation-an even worse scenario to further accelerate catalyst deactivation to a shorter lifetime 11 . Increasing metal-support interactions (MSIs) 12,13 or confining active metal species in restrained spaces (for example, micropores) 14,15 are two widely adopted methodologies to achieve higher stability. However, excessively strong MSIs might over-stabilize the active sites and cause low activity 13,16 ; confinement of metal species in micropores might also limit mass transfers 14,17 . Consequently, a seesaw relation between stability and activity has been often observed, obscuring efforts for catalytic performance optimizations [18][19][20] .
In this work, we report that the challenge can be solved by synergizing MSIs and spatial confinement to break the activitystability seesaw of SADCs at high metal loadings. Specifically, strong Ni-N/C interaction and spatial confinement by dense atomic Cu grippers (8.1 wt%) on a graphitic carbon nitride (g-C 3 N 4 ) support are synergized to endow great structural flexibility of highly loaded (3.1 wt%) atomic Ni with reversible bond breakings and makings on reactant adsorption and product desorption, respectively. Such a dynamic effect not only facilitates acetylene adsorption for high activity but also ensures high thermal/chemical stability and unprecedented inhibition of coke formation.

Results
Synthesis and catalytic performance. As it contains periodic triangular cavities with six coordinately unsaturated pyridinic nitrogen (N py ) atoms, g-C 3 N 4 provides a perfect platform for anchoring active and dense metal species 21,22 . Screening of metals among groups VIII and IB by density functional theory (DFT) calculations revealed that Ni and Cu atoms exhibit strong binding to g-C 3

N 4 underneath
Synergizing metal-support interactions and spatial confinement boosts dynamics of atomic nickel for hydrogenations along with low formation energy (Supplementary Table 1); therefore, they are the most promising candidates for highly loaded and stable metals on g-C 3 N 4 with atomic dispersion.
Inspired by these results, atomic layer deposition (ALD), a gas-phase technique relying on sequential self-limiting molecular-level surface reactions 23,24 , was employed to deposit Cu and Ni with near-atomic precision 25,26 . Atomic Cu grippers were first fabricated on g-C 3 N 4 by Cu ALD (denoted as Cu 1 /g-C 3 N 4 ), as confirmed by X-ray absorption spectroscopy and aberration-corrected high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM), where the saturated loading was ~11.2 wt% (Supplementary Figs. 1a and 2-5 and Supplementary Table 2). Ni y Cu 1 /g-C 3 N 4 catalysts (y is the atomic ratio of Ni to Cu) were synthesized by depositing Ni on subsaturated Cu 1 /g-C 3 N 4 using NiO x ALD, as shown in the schematic and morphology in Supplementary Figs. 6 and 7, respectively. Therein, the Ni loadings decreased with an increase in Cu loadings ( Supplementary  Fig. 1b), which implies that at Cu subsaturation coverages, the open N py sites along with neighbouring Cu atoms provide anchoring sites for guest atoms, analogous to pincer ligands in organometallic chemistry 27 . For comparison, Ni 1 /g-C 3 N 4 SADCs were also fabricated by NiO x ALD (Supplementary Figs. 8-12 and Supplementary  Table 3). HAADF-STEM measurements showed that Cu 1 /g-C 3 N 4 , Ni 1 /g-C 3 N 4 and Ni 1 Cu x /g-C 3 N 4 were thermally stable at 350 °C under H 2 , in line with the above theoretical prediction.
Remarkably, in addition to high thermal stability, Ni 1 Cu 2 /g-C 3 N 4 showed unprecedented chemical stability under hydrogenation conditions, without any visible decline in either activity or selectivity for at least 350 h at 160 °C, and the used sample remained structurally intact ( Fig. 1d and Supplementary Figs. [17][18][19]. In sharp contrast, Ni 1 /g-C 3 N 4 showed a slight activity increase approximately in the first six hours, and then rapidly deactivated with an activity loss of ~50% in about 50 h. X-ray diffraction (XRD) and HAADF-STEM measurements also revealed severe agglomeration of Ni atoms into Ni particles (Supplementary Figs. 19 and 20).
State-of-the-art in situ synchrotron vacuum ultraviolet photoionization mass spectrometry (SVUV-PIMS) revealed that besides C 2 products, considerable gaseous C 3 -C 8 species were produced from Ni 1 /g-C 3 N 4 ( Fig. 1e and Supplementary Fig. 21), indicating the facile formation of green oils. In contrast, no higher hydrocarbons formed on Ni 1 Cu 2 /g-C 3 N 4 , except a trace amount of C 4 H 8 . Temperature-programmed desorption (TPD) SVUV-PIMS measurements of the two samples after 2 h of reaction further showed negligible accumulation of C 4+ species on Ni 1 Cu 2 /g-C 3 N 4 ( Fig. 1f and Supplementary Figs. 22 and 23), unambiguously confirming the complete inhibition of the formation of green oils on Ni 1 Cu 2 /g-C 3 N 4 .
Coke formation was further quantitatively monitored by in situ thermogravimetric analysis (TGA) under hydrogenation conditions. After 12 h, the weight gain of Ni 1 Cu 2 /g-C 3 N 4 was negligible, close to the thermal noise. However, Ni 1 /g-C 3 N 4 showed a considerable weight increase by 7.6% (Fig. 1g), in line with the above SVUV-PIMS measurements. Meanwhile, the XRD data revealed the formation of graphitic coke on Ni 1 /g-C 3 N 4 ( Supplementary Fig. 19), but not for Ni 1 Cu 2 /g-C 3 N 4 . Clearly, the capability of coke inhibition and long durability achieved on Ni 1 Cu 2 /g-C 3 N 4 is unprecedented, far superior to the benchmark catalysts reported in the literature (Table 1 and Supplementary Tables 4-6). Besides the above, Ni 1 Cu 2 /g-C 3 N 4 also exhibited extraordinary performance in the semi-hydrogenation of 1,3-butadiene in excess propene ( Supplementary Fig. 24), rendering this material promising for wide applications.
Structural characterization. Statistical analysis of atomic-resolution STEM images of these samples found that there were mainly three structural motifs, namely, single atoms, linear trimers and triangular trimers ( Fig. 2a and Supplementary Figs. 5, 7, and 12). For Cu 1 /g-C 3 N 4 , single atoms were the most populated (68%), followed by linear trimers (20%) and triangular trimers (12%) ( Supplementary  Fig. 25). Subsequent Ni deposition forms Ni 1 Cu 2 /g-C 3 N 4 , and linear trimers increased dramatically (71%) at the expense of single atoms (18%), while triangular trimers remained at ~11% (Fig. 2b). This suggests that most of the predeposited single Cu atoms were likely bridged by Ni atoms to form new linear trimers ( Supplementary  Fig. 6). Atomic-resolution electron energy-loss spectroscopy (EELS) elemental mapping of one linear trimer at a relatively low beam current to minimize electron-beam perturbations (Supplementary Fig. 26) further provided strong evidence of the linear Cu-Ni-Cu trimer structure, by identifying a Ni atom in the middle and two Cu atoms on the adjacent sides (Fig. 2c,d). The overall lengths of the linear trimers were 5.66 ± 0.20 Å comprising two unequal atom-atom distances of 2.61 ± 0.14 and 3.04 ± 0.13 Å on average ( Supplementary Fig. 7).
To confirm direct Cu-Ni interactions, in situ diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) of CO chemisorption measurements were carried out. As shown in Fig. 2e and Supplementary Fig. 27, Cu 1 /g-C 3 N 4 showed one dominant peak at 2,103 cm −1 and a weak shoulder at 2,123 cm −1 , assigned to linear CO on isolated Cu + ions and trimers, respectively 29 . CO chemisorption on Ni 1 /g-C 3 N 4 was negligible. Deposition of Ni atoms on Cu 1 /g-C 3 N 4 gradually diminished the intensity of the CO peak, strongly verifying the direct Cu-Ni interactions. On Ni 1 (3.1%)Cu 1 (8.1%)/g-C 3 N 4 (the optimized Ni 1 Cu 2 /g-C 3 N 4 sample), when considering the Cu:Ni mole ratio of 2.4:1.0, the trivial CO peak strongly confirms that one Ni is approximately coordinated with two Cu atoms but without forming Ni particles/clusters, in line with the formation of Ni 1 Cu 2 linear trimers ( Supplementary Fig. 28). Analysis of the CO peak areas before and after Ni deposition reveals that Ni 1 Cu 2 linear trimers might be about 87% in Ni 1 Cu 2 /g-C 3 N 4 , agreeing well with the STEM statistics (Fig. 2b). A control experiment of depositing Ni on a Cu 1 /C 3 N 4 sample at low Cu loading revealed that a high density of Cu 1 atoms is essential to form Ni 1 Cu 2 linear trimers ( Supplementary Fig. 29). C 2 H 2 and C 2 H 4 TPD were also applied to probe direct Ni-Cu interactions. We found that the main C 2 H 2 desorption peak of Ni 1 Cu 2 /g-C 3 N 4 was located at 107 °C, sharply different from that of Cu 1 /g-C 3 N 4 (146 °C) and Ni 1 /g-C 3 N 4 (82 °C), further supporting the formation of Ni 1 Cu 2 linear trimers ( Supplementary  Fig. 30). Meanwhile, the tailing C 2 H 2 desorption peak of Ni 1 Cu 2 /g-C 3 N 4 (146 °C) coincided with that of Cu 1 /g-C 3 N 4 , suggesting the coexistence of isolated Cu atoms. Quantitative analysis of the TPD spectrum reveals that the percentage of Ni 1 Cu 2 linear trimers is about 74% (Fig. 2b), close to the STEM statistics (71%). Analysis of C 2 H 4 TPD showed the same result. Considering the negligible activity of Cu 1 /g-C 3 N 4 , Ni 1 /g-C 3 N 4 and Ni 1 (1%)Cu 1 (0.5%)/g-C 3 N 4 below 170 °C ( Fig. 1a and Supplementary Fig. 29), the exceptional performance of Ni 1 Cu 2 /g-C 3 N 4 should come solely from the prevailing linear trimers.
In situ X-ray photoemission spectroscopy (XPS) showed that the Cu2p 3/2 binding energy was invariant at 933.2 eV, irrespective of the treatments-an indication of the +1 oxidation state 27,30 (Fig. 2f and Supplementary Fig. 31), which is in line with the DRIFTS CO chemisorption data (Fig. 2e). The Ni2p 3/2 binding energy located at 855.8 eV, close to that of Ni(OH) 2 (ref. 31 ). Along with this, considerable hydroxyl-type oxygen, with O1s at 531.4 eV, was found. Quantitative XPS analysis showed that the stoichiometric ratio between O and Ni was about 2:1. Taken together, these results suggest that the active linear trimers are Ni 1 (OH) 2 Cu 2 complexes formed under the reaction conditions.
To reveal the structure of the Ni 1 (OH) 2 Cu 2 complexes in Ni 1 Cu 2 /g-C 3 N 4 , in situ X-ray absorption spectroscopy measurements were further conducted ( Supplementary Fig. 15). At the Ni K edge, Ni 1 Cu 2 /g-C 3 N 4 Ni 1 /g-C 3 N 4 Ni 1 Cu 2 /g-C 3 N 4 Ni 1 /g-C 3 N 4 @160 °C @200 °C Ni 1 Cu 2 /g-C 3 N 4 Ni 1 /g-C 3 N 4 Fig. 1 | Catalytic performance of Ni-Cu pincer complex catalyst. a,b, Acetylene conversion as a function of temperature (a) and ethylene selectivity as a function of acetylene conversion (b) over Ni 1 Cu 2 /g-C 3 N 4 , Cu 1 /g-C 3 N 4 and Ni 1 /g-C 3 N 4 catalysts. c, Activity of these three samples at 160 °C. m, metal (Cu or Ni). d, Time-on-stream (TOS) stability with the same initial conversion of 70% at different temperatures. The inset shows the expanded conversion changes in the first 20 h, with the same x-and y-axis labels as the main figure. Reaction conditions: 0.5% C 2 H 2 , 5% H 2 and 25% C 2 H 4 balanced with Ar; the space velocity is 12,000 ml g −1 h −1 and the pressure is 0.1 MPa. e, Distributions of volatile C n (n = 3, 4, 5…) gas-phase products at the reactor outlet during reactions on Ni 1 Cu 2 /g-C 3 N 4 and Ni 1 /g-C 3 N 4 catalysts from in situ SVUV-PIMS measurements in the stable region. f, Distributions of the desorption of accumulated C n (n = 3, 4, 5…) green oils on these two samples in 2 h of reaction. g, In situ TGA measurements of coke deposition on Ni 1 Cu 2 /g-C 3 N 4 and Ni 1 /g-C 3 N 4 catalysts under reaction conditions at 160 and 200 °C, respectively.
the extended X-ray absorption fine-structure (EXAFS) curve of Ni 1 Cu 2 /g-C 3 N 4 after 350 °C hydrogen reduction (Ni 1 Cu 2 /g-C 3 N 4 -R) was nearly identical to that under the hydrogenation condition (Ni 1 Cu 2 /g-C 3 N 4 -H) (Fig. 2g). The main peak at 1.41 Å is assigned to Ni-O/N/C scattering, and the weak peak at 2.21 Å is attributed to Ni-O/N/C and Ni-Cu coordination according to wavelet transform ( Supplementary Fig. 32). Quantitative EXAFS fitting revealed that the coordination numbers (CNs) for the nearest Ni-O/N/C and Ni-Cu were 3.3 and 0.8 with bond distances of 1.96 and 2.90 Å, respectively (Supplementary Fig. 33 and Supplementary Table 7). Similarly, at the Cu K edge, Ni 1 Cu 2 /g-C 3 N 4 -R and Ni 1 Cu 2 /g-C 3 N 4 -H appeared at the same structure. The main peak at 1.47 Å was attributed to Cu-O/NC coordination with a CN of 1.7, and the weak peak at 2.33 Å was assigned to Cu-Cu/Ni coordination with a CN of 0.5 ( Fig. 2h, Supplementary Fig. 34 and Supplementary Table 8).

Spectroscopic insight into the hydrogenation process.
In situ DRIFTS measurements were carried out to investigate the catalytic roles of hydroxyl groups. We found that as Ni 1 /g-C 3 N 4 was exposed to acetylene at 160 °C, two negative peaks appeared at 3,568 and 3,430 cm −1 , assigned to Ni-OH groups (  33 . Switching the gas to hydrogen, the peaks of all C-H stretching declined, indicating the formation of ethylene and ethane. However, the Ni-OH groups were not recovered. Repeating the above experiments for another acetylene-hydrogen cycle showed a similar result. In sharp contrast to Ni 1 /g-C 3 N 4 , the consumption and regeneration of Ni-OH groups at 3,581 and 3,440 cm −1 on Ni 1 Cu 2 /g-C 3 N 4 was totally reversible during sequential acetylene and hydrogen exposures, implying that the Ni-OH group directly participates in the hydrogenation reaction (Fig. 3b). Interestingly, there was a lack of any features in the range of 2,800-2,900 cm −1 , suggesting that much less ethane was formed on Ni 1 Cu 2 /g-C 3 N 4 . Meanwhile, exposing Ni 1 Cu 2 /g-C 3 N 4 to ethylene did not cause any consumption of OH groups ( Supplementary Fig. 38). These findings unambiguously support the high ethylene selectivity of Ni 1 Cu 2 /g-C 3 N 4 (Fig. 1b). Finally, in situ DRIFTS of C 2 H 2 hydrogenation on the  Supplementary Fig. 39), validating the excellent stability.
Theoretical insights into Ni 1 and Ni 1 Cu 2 catalysts. Based on the above structural characterization and extensive DFT calculations, the Ni-Cu complex is identified as a chain-like Cu-OH-Ni-OH-Cu structure (Fig. 4a), whereas the Ni species in Ni 1 /g-C 3 N 4 after reduction forms the Ni 1 (OH) 2 complex (Supplementary Fig. 40 and Supplementary Table 9). For the former one, the Ni atom bonds to N py and C of g-C 3 N 4 underneath, and is confined by Cu atoms from two sides via the bridging OH group; the Cu atoms are  Fig. 3 | In situ DRIFtS investigation of C 2 H 2 hydrogenation. a,b, In situ DRIFT spectra over Ni 1 /g-C 3 N 4 (a) and Ni 1 Cu 2 /g-C 3 N 4 (b) catalysts after alternatively exposing to C 2 H 2 and H 2 for different times with two consecutive C 2 H 2 and H 2 cycles. The shaded regions with different colours highlight the variations in the peak position and intensity. Conditions: 0.03 g catalyst; 160 °C; 0.5% C 2 H 2 or 5% H 2 in Ar; total flow, 6 ml min −1 .    Table 10). The overall length of the optimized complex (5.84 Å) and two unequal Ni-Cu distances (2.74 and 3.10 Å) also perfectly agree with the STEM data ( Supplementary Fig. 25). There is substantial charge accumulation between the Ni-C/N bonds and between the Cu-C/N bonds, while considerable charge transferring from Ni/Cu to high-lying OH groups (Fig. 4a and Supplementary Fig. 41); therefore, the characteristic covalent bond and ionic bond coexist between Ni 1 Cu 2 (OH) 2 and the g-C 3 N 4 support, as well as in Ni 1 (OH) 2 /g-C 3 N 4 . Impressively, the formation of Ni 1 Cu 2 linear trimer is energetically favourable, exothermic by 0.79 eV ( Supplementary  Fig. 42), manifesting the driving force for the linear trimer formation ( Supplementary Fig. 6). Different chemical stability of Ni 1 /g-C 3 N 4 and Ni 1 Cu 2 /g-C 3 N 4 stems from the distinct structure change induced by stronger C 2 H 2 adsorption, as found by DFT calculations and TPD measurements (Supplementary Fig. 30 and Supplementary Table 11). We found that acetylene adsorption on Ni 1 /g-C 3 N 4 has a modest barrier of 0.62 eV and exothermic by 0.69 eV (Fig. 4b), while simultaneously breaking one Ni-N/C support bond. The one Ni-N/C support bond left can be further broken only with a barrier of 0.66 eV. In other words, the Ni(OH) 2 -C 2 H 2 complex is rather mobile on g-C 3 N 4 , which would facilitate its aggregation to form Ni NPs, as found by STEM and XPS measurements (Supplementary Figs. 20 and 37). Obviously, strong Ni-support interaction alone is not sufficient to stabilize the Ni(OH) 2 complexes, although the bonding by the two Ni-N bonds (−2.93 eV) is considerably strong.
Acetylene adsorption on Ni 1 Cu 2 /g-C 3 N 4 is stronger with an exothermic energy of 1.00 eV, in line with the TPD measurement ( Supplementary Fig. 30). The optimized structure indicates that although acetylene adsorption breaks two interfacial Ni-N/C support bonds with practically no barrier, spatial confinement by two atomic Cu grippers from both sides through two bridging hydroxyls endows the breakages of the Cu-O support bond extremely endothermic by more than 4 eV, ensuring the high chemical stability of this catalyst during acetylene hydrogenation. Meanwhile, the two new Ni-C bonds formed with adsorbed C 2 H 2 retain the overall CN of the Ni atom before and after acetylene adsorption ( Fig. 4b and Supplementary Video 1), which is in good agreement with the X-ray absorption fine-structure (XAFS) data ( Fig. 2g and Supplementary  Table 10). Therefore, the resulting dynamics of Ni atoms via synergizing MSIs and spatial confinements ensures not only a facile structural transformation by breaking the Ni-N/C bonds to enhance acetylene adsorption but also a high chemical/thermal stability of SADCs free from aggregations ( Supplementary Fig. 43).
Atomically dispersed and stable Ni atoms are beneficial to suppress C-C bond breaking and creating the prerequisite for the formation of graphitic coke and green oils due to the limited sites available (Fig. 4c and Supplementary Table 12). For instance, the cleavage of adsorbed acetylene is highly endothermic because of the absence of sites accommodating the broken CH* fragments. Moreover, after the preferential adsorption of acetylene at the Ni atom, the additional adsorption of acetylene or ethylene is only possible at the Cu sites with a weak binding energy of −0.55 and −0.51 eV, respectively, but will desorb at temperatures above room temperature. As a result, the formation of graphitic coke and green oils is prohibited on Ni 1 Cu 2 /g-C 3 N 4 , which is in excellent agreement with the in situ SVUV-PIMS and TGA measurements (Fig. 1e-g).
Optimized potential energy surface for the hydrogenation of acetylene over the Ni 1 (OH) 2 Cu 2 complex is shown in Fig. 4d and Supplementary Figs. 44-46. Acetylene is expected to be dominant on the Ni sites during hydrogenation, owing to its much stronger adsorption (−1.00 eV) than H 2 (−0.07 eV) (Supplementary Table 11). After acetylene adsorption, the proton in one bridging OH tends to directly attack the adsorbed acetylene with a barrier of 1.23 eV to form C 2 H 3 * (here * denotes surface species; Fig. 4d(i),(ii)). Next, hydrogen adsorbs on the Ni sites and effortlessly dissociates with one H atom in addition to C 2 H 3 * to form C 2 H 4 *, and the other one bonds to the Ni site (Fig. 4d(iii),(iv)). After that, the C 2 H 4 * species rearranges and the H atom transfers to the bridging O from Ni to regenerate the bridging OH with a low barrier of 0.81 eV. Compared with the sequential high hydrogenation barrier of 1.38 eV from ethylene to ethane, weakly bonded ethylene with a value of −0.67 eV boosts the desorption rate of ethylene-a key step for the high selectivity of Ni 1 Cu 2 /g-C 3 N 4 , which is in line with the catalytic results shown in Fig. 1. Obviously, the participation of bridging hydroxyls in the hydrogenation reaction suggested by DFT is in good agreement with the DRIFTS results (Fig. 3).

Conclusions
In summary, we have reported that synergizing MSIs and spatial confinement through Cu atomic grippers boosts the dynamics of highly loaded atomic Ni for efficient hydrogenations. Such dynamic and synergetic effects enable not only exceptionally high stability against both sintering and coke formation in the hydrogenation of acetylene but also lead the active sites to become highly adaptable towards both reactant adsorption and product desorption, thus boosting the activity (11 times higher than the conventional Ni 1 /g-C 3 N 4 catalyst) and ethylene selectivity. Compared with the literature [34][35][36] , the synthesis of metal trimer catalysts using an atom-by-atom fabrication approach provides an avenue for the rational design of highly active and stable SADCs for practical application in selective chemical reactions.

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/ s41565-021-00951-y.
Synthesis of g-C 3 N 4 support. Urea was placed in a crucible and heated to 600 °C for 4 h in static air at a ramp rate of 5 °C min −1 . The resulting yellow material was milled into fine powders in a mortar. Next, the yellow powder was placed in an open ceramic container and heated at 550 °C for another 4 h. A light-yellow powder of g-C 3 N 4 support was finally obtained.

Synthesis of Cu 1 /g-C 3 N 4 catalysts.
Cu ALD was carried out on a viscous-flow stainless steel tube reactor system (ACME (Beijing) Technology) at 300 °C using Cu(hfac) 2 and formaldehyde as the precursors. Ultrahigh-purity N 2 (99.999%) was used as a carrier gas at a flow rate of 60 ml min −1 . The Cu(hfac) 2 precursor container was heated to 65 °C to reach a sufficient vapour pressure. The inlet manifold was held at 100 °C to avoid precursor condensation. The timing sequence was 300, 300, 40 and 180 s for Cu(hfac) 2 exposure, N 2 purge, formaldehyde exposure and N 2 purge, respectively. One, two, and ten cycles of Cu ALD were performed on g-C 3 N 4 to obtain Cu 1 (5.1%)/g-C 3 N 4 , Cu 1 (8.1%)/g-C 3 N 4 and Cu 1 (11.4%)/g-C 3 N 4 catalysts, respectively. When the Cu(hfac) 2 exposure time was reduced to 25 and 150 s for one cycle of Cu ALD on g-C 3 N 4 , Cu loadings of 0.5% and 2.7% were obtained, which were denoted as Cu 1 (0.5%)/g-C 3 N 4 and Cu 1 (2.7%)/g-C 3 N 4 , respectively.

Synthesis of Ni 1 /g-C 3 N 4 and Ni 1 Cu 2 /g-C 3 N 4 catalysts. NiO
x ALD was carried out on the g-C 3 N 4 support at 240 °C using NiCp 2 and ozone as the precursors. The NiCp 2 precursor container was heated to 90 °C to achieve a sufficient vapour pressure. The inlet manifold was held at 115 °C to avoid precursor condensation on the inner walls. The timing sequence was 1,260, 200, 1,260 and 200 s for NiCp 2 exposure, N 2 purge, O 3 exposure and N 2 purge, respectively. Different NiCp 2 exposure times (from 100 to 1,260 s) were used on g-C 3 N 4 to obtain a series of Ni single-atom catalysts with different Ni loadings (denoted as Ni 1 (x%)/g-C 3 N 4 , where x is the Ni loading). NiO x ALD was also carried on the Cu 1 (8.1%)/g-C 3 N 4 samples with different NiCp 2 exposure times (from 200 to 1,260 s). The resulting samples were denoted as Ni 1 (x%)Cu 1 (8.1%)/g-C 3 N 4 (the optimized sample, namely, Ni 1 (3.1%)Cu 1 (8.1%)/g-C 3 N 4 , was denoted as Ni 1 Cu 2 /g-C 3 N 4 ).

Morphology and compositions.
Aberration-corrected HAADF-STEM measurements were taken on a JEM-ARM200F instrument (University of Science and Technology of China) at 200 keV. Meanwhile, energy-dispersive X-ray elemental mapping was also collected on the same equipment. Transmission electron microscopy and STEM measurements were performed on a JEOL JEM-2100F instrument operated at 200 keV.
Atomic-resolution EELS analysis of Cu-Ni-Cu line structures was performed on a Nion HERMES-100 instrument equipped with a C3/C5 corrector, operated at 60 kV with a convergence angle of around 32 mrad and a collection angle of about 75 mrad (University of Chinese Academy of Sciences). High-angle annular dark-field Z-contrast images were also collected for the same area, using an annular detector with a range of 75-210 mrad. To map out the dispersion of Ni L 2,3 edge and Cu L 2,3 edge signals in the line structures, we performed the EELS mapping with a probe current of about 11 pA, sampling with a pixel size of about 0.03 nm and a dwell time of 150 ms per pixel. To further improve the signal-to-noise ratio in the EELS experimental data, principal component analysis-a commonly used mathematical method to de-noise multidimensional datasets-was applied to the experimental raw data using HyperSpy (ref. 51 ). The EELS data were recalibrated based on standard Cu and Ni spectra.
The Ni and Cu loadings of the catalysts were analysed by an inductively coupled plasma atomic emission spectrometer. The XRD spectra were recorded on a SmartLab X-ray diffractometer with a 9 kW rotating anode.

In situ SVUV-PIMS measurements.
To study the formation of green oils, in situ SVUV-PIMS measurements were conducted on the mass spectrometry end-station (BL04B) at the National Synchrotron Radiation Laboratory, Hefei, China 52 . In brief, 150 mg of Ni 1 Cu 2 /g-C 3 N 4 or Ni 1 /g-C 3 N 4 catalyst was loaded in a fixed-bed flow reactor, which was connected to the SVUV-PIMS system 53 . The catalyst was first reduced in 10% H 2 in Ar at a flow rate of 50 ml min −1 at 350 °C for 2 h and then cooled to 160 °C for Ni 1 Cu 2 /g-C 3 N 4 and 200 °C for Ni 1 /g-C 3 N 4 . After the sample temperature stabilized, a reaction gas consisting of 0.5% C 2 H 2 , 5% H 2 and 25% C 2 H 4 balanced with Ar at a flow rate of 50 ml min −1 was then introduced into the quartz-tube reactor at a total pressure of 2 torr. The volatile components at the reactor outlet were ionized by the synchrotron vacuum ultraviolet light, and the formed ions were analysed by a home-made time-of-flight mass spectrometer at a photon energy of 12 eV. After 2 h of continuous data collection, the reaction gas was switched to pure Ar; then, the the sample temperature was gradually increased at a heating rate of 5 °C min −1 to 300 °C, while the composition of accumulated green oils on the two samples was analysed by the online SVUV-PIMS spectrometer.
In situ TGA measurements. In situ TGA measurements were performed on a TGA550 instrument (TA Instruments) equipped with an evolved gas analysis furnace to quantify coke formation during the semi-hydrogenation of acetylene. After loading the sample into the furnace, the sample was heated to 350 °C at a heating rate of 3 °C min −1 in 10% H 2 /Ar (25 ml min −1 ) and then isothermally maintained for 2 h to reduce the sample. Next, the sample was cooled to a specified temperature in argon (25 ml min −1 ) and isothermally maintained for 1 h to stabilize the instrument. Next, acetylene hydrogenation gas consisting of 0.5% acetylene, 5% H 2 and 25% ethylene balanced with Ar was introduced into the sample at a flow rate of 25 ml min −1 for about 12 h at the specified temperature. The change in sample weight with time during acetylene hydrogenation was recorded. Here the temperatures for acetylene hydrogenation were 160 and 200 °C for Ni 1 Cu 2 /g-C 3 N 4 and Ni 1 /g-C 3 N 4 catalysts, respectively, to ensure a similar initial conversion for these samples.
In situ XPS measurements. In situ XPS and Auger spectra measurements were conducted at the BL10B beamline photoemission end-station at the National Synchrotron Radiation Laboratory, Hefei, China. In brief, the beamline is connected to a bending magnet and covers photon energies from 100 to 1,000 eV. The end-station consists of four chambers, namely, analysis chamber, preparation chamber, quick sample load-lock chamber and high-pressure reactor. The analysis chamber, with a base pressure of <5 × 10 −10 torr, is connected to the beamline and equipped with a VG Scienta R3000 electron energy analyser and a twin-anode X-ray source. The high-pressure reactor houses a reaction cell where the samples can be treated with different gases up to 20 bar and simultaneously heated up to 650 °C. After sample treatment, the reactor can be pumped down to a high vacuum (<10 −8 torr) for sample transfer.
In this work, after loading the as-prepared Ni 1 Cu 2 /g-C 3 N 4 sample into the chamber, a set of XPS spectra in the Ni2p, Cu2p, N1s, C1s and O1s regions were first recorded without any treatment (denoted as Ni 1 Cu 2 /g-C 3 N 4 ). Next, the sample was transferred back into the high-pressure reactor and reduced in 10% H 2 in Ar at 350 °C for 2 h at atmosphere pressure. Another set of XPS spectra were collected after transferring back to the analysis chamber. The reduced sample was denoted as Ni 1 Cu 2 /g-C 3 N 4 -R. Finally, the sample was exposed to the acetylene hydrogenation stream consisting of 0.5% C 2 H 2 , 5% H 2 and 25% C 2 H 4 balanced with Ar in the high-pressure reactor at 160 °C for another 1 h. Another set of XPS spectra were then collected after the hydrogenation-reaction gas treatment without exposing to air, which was denoted as Ni 1 Cu 2 /g-C 3 N 4 -H. Similarly, XPS spectra of the Ni 1 /g-C 3 N 4 sample were also collected following the same procedure. The XPS spectra of the sample after reduction at 350 °C for 2 h and exposing the reaction gas at 200 °C for another 1 h were denoted as Ni 1 /g-C 3 N 4 -R and Ni 1 /g-C 3 N 4 -H, respectively. The C1s binding energy at 284.8 eV was used as an internal reference.
In situ XAFS. In situ XAFS measurements at the Ni K edge (8,333 eV) and Cu K edge (8,979 eV) were performed in the transmission mode with the Si(311) monochromator at the BL14W1 beamline at the Shanghai Synchrotron Radiation Facility (SSRF), China. The storage ring at the SSRF worked at 3.5 GeV with a maximum current of 210 mA (ref. 54 ). A home-made reaction cell was used for the operando experiments, which allows exposing the sample to different pretreatment gases or reactions gases, as well as sample heating up to 400 °C.
Typically, the as-prepared Ni 1 Cu 2 /g-C 3 N 4 sample was put into the reaction cell and purged with He for 10 min at room temperature; then, the XAFS spectra were recorded at both Ni and Cu K edges (denoted as Ni 1 Cu 2 /g-C 3 N 4 ). After that, the sample was reduced in 10% H 2 in He at 350 °C for 2 h. After cooling the sample to 160 °C in 10% H 2 in He, the XAFS spectrum was recorded at the Ni K edge and Cu K edge, which was denoted as Ni 1 Cu 2 /g-C 3 N 4 -R. Finally, the sample was exposed to the acetylene hydrogenation reaction stream consisting of 0.5% C 2 H 2 , 5% H 2 and 25% C 2 H 4 balanced with Ar at 160 °C for another 1 h, and the XAFS spectra were sequentially recorded at the Ni and Cu K edges, which were denoted as Ni 1 Cu 2 /g-C 3 N 4 -H. Similarly, in situ XAFS measurements of Ni 1 /g-C 3 N 4 were also collected following the same procedure but only collecting the XAFS spectra at the Ni K edge. The XAFS spectra of the sample after reduction at 350 °C for 2 h were denoted as Ni 1 /g-C 3 N 4 -R.
The acquired EXAFS data were processed according to standard procedures using the ARTEMIS module implemented in the IFEFFIT software package. The EXAFS oscillation functions χ(k) were obtained by subtracting the post-edge background from the overall absorption spectra and then normalized with respect to the edge-jump step.
In terms of the Ni K edge, fittings were done in the interatomic distance (R) space within the R range of 0.8-3.0 Å −1 for k 3 -weighted χ(k) functions with Hanning windows (dk = 1.0 Å −1 ) and a k range of 2.3-10.4 Å −1 was used; during curve fittings, the amplitude reduction factor S 0 2 was fixed at a value of 0.78 determined by fitting the data of Ni foil. For fitting the EXAFS data of the Ni 1 Cu 2 /g-C 3 N 4 sample, the energy shift (ΔE 0 ) values for the Ni-C/N bonds were fixed to reduce the number of adjustable parameters and to break the strong correlation between ΔE 0 and R. In terms of the Cu K edge, the R range of 1.0-3.0 Å −1 and k range of 2.9-12.4 Å −1 were used; during curve fittings, the amplitude reduction factor S 0 2 was fixed at 0.9 as determined by fitting the data of Cu foil. Debye-Waller factor (σ 2 ), N, R and ΔE 0 were treated as adjustable parameters during the fitting process.
The yielded R factors of all the samples are not larger than 0.008, indicating good fitting qualities. TPD measurements. The C 2 H 2 and C 2 H 4 TPD experiments were performed on a Micromeritics AutoChem II chemisorption analyser, which was connected to a Pfeiffer OmniStar mass spectrometer. In a typical experiment, 50 mg sample was first reduced in 10% H 2 /He at 350 °C for 2 h and then purged with He at 300 °C for 1 h. Next, the sample was cooled down to room temperature in He; then, C 2 H 2 or C 2 H 4 gas was introduced into the sample until saturation. Subsequently, the sample was purged with He at 25 °C for 1 h and then the sample temperature was gradually increased at a heating rate of 10 °C min −1 in He from 25 to 300 °C while recording the TPD data.
In situ DRIFTS. In situ DRIFTS was performed on a Thermo Nicolet iS10 FTIR spectrometer equipped with a mercury-cadmium-telluride detector at a resolution of 4 cm −1 and 128 scans. First, 30 mg of the sample was placed in a high-temperature cell (Harrick) with CaF 2 windows. For in situ DRIFTS of CO chemisorption, the sample was first reduced in 10% H 2 in Ar at 350 °C for 2 h. Then, the sample was cooled down to 25 °C in Ar. After that, the sample was exposed to 10% CO/Ar (30 ml min −1 ) for 30 min until saturation and the spectra were continuously collected during this process. The CO chemisorption spectra were obtained by subtracting the CO chemisorption DRIFT spectrum of g-C 3 N 4 to remove the gas-phase CO.
As for in situ DRIFTS of C 2 H 2 or C 2 H 4 hydrogenation, the sample was first reduced in 10% H 2 in Ar at 350 °C for 2 h. Then, the sample was cooled down to 160 °C in Ar and kept for 30 min while the spectra were collected. After that, the experiments were repeated by sequentially exposing to C 2 H 2 and H 2 for C 2 H 2 hydrogenation or sequentially exposing to C 2 H 4 and H 2 for C 2 H 4 hydrogenation. The total flow rate was 6 ml min −1 . It should be noted that all the gases were passed through a calcium chloride filter at −40 °C in the gas inlet line before the Harrick cell to remove trace water from the gas.
As for in situ DRIFTS of water adsorption, the sample was first reduced in 10% H 2 in Ar at 350 °C for 2 h. Then, the sample was cooled down to 25 °C in Ar. After that, the sample was exposed to 5% H 2 O/Ar (6 ml min −1 ) for 30 min until saturation, and the spectra were continuously collected during this process.
Catalytic activity test. Selective hydrogenation of acetylene in excess ethylene was conducted in a fixed-bed flow reactor. The total flow rate was kept at 30 ml min −1 . The amount of non-precious catalyst was 150 mg, which was diluted with 1 g of 200 mesh quartz chips. Before the reaction test, all the catalysts were reduced in 10% H 2 in Ar at 350 °C for 2 h. Then, the feed gas-consisting of 0.5% acetylene, 5% H 2 and 25% ethylene with Ar as the balance gas-was introduced into the reactor to start the reaction. Next, acetylene conversion of all these catalysts was increased by increasing the reaction temperature. The reaction products were analysed using an online gas chromatograph (A91, Panna Instrument) equipped with a flame ionization detector and a capillary column (ValcoPLOT VP-alumina-KCl; 50 m × 0.53 mm). The acetylene conversion and ethylene selectivity were calculated using the following equations, respectively: For the kinetic measurements, we reduced the amount of catalyst and increased the flow rate of the reaction gas to ensure that the acetylene conversion was below 15%. Turnover frequencies (TOFs) of the non-precious catalysts were calculated using the following equation: Mole of C2H2 converted per hour D × n metal Here n metal is the mole of Ni or Cu. For Ni 1 Cu 2 /g-C 3 N 4 , n metal is the mole of Ni. Also, D is the dispersion of metal NPs or clusters/atoms; D = 1 for Ni 1 /g-C 3 N 4 , Cu 1 /g-C 3 N 4 and Ni 1 Cu 2 /g-C 3 N 4 .
Semi-hydrogenation of 1,3-butadiene was also recorded on the same reaction test station and using the same gas chromatograph. The feed gas consists of 1.2% 1,3-butadiene, 12% H 2 with Ar as the balance gas. The same pretreatment was performed on all the catalysts before the reaction test. Here 1,3-butadiene conversion and butene selectivity were calculated using the following equations, respectively: Computational methods and models. All the spin-polarized calculations were performed using DFT calculations implemented in the Vienna ab initio simulation package 55 . We described the electron-ion interaction using the projector augmented wave method 56 . The exchange-correlation interaction was described by the optB86b-vdW functional (ref. 57 ). The Kohn-Sham equations were solved by using a plane-wave basis set with a kinetic energy cutoff of 400 eV. The convergence criterion for electronic self-consistent interactions is 10 −4 eV. The geometries of bulk and surface were optimized by the conjugate gradient algorithm until the maximum force on any ion was less than 0.02 eV Å -1 . All the transition states were determined by using the climbing image nudged elastic band method 58 with confirmation of vibrational analysis. The transition states are optimized until all the forces are below 0.05 eV Å -1 .
The g-C 3 N 4 was modelled by a (2 × 2) supercell of a single-layer g-C 3 N 4 , where all the atoms in the supercell were fully relaxed. A (3 × 3 × 1) Monkhorst-Pack k-point mesh was used to sample the surface Brillouin zone. A 20 Å vacuum was introduced to avoid interaction from adjacent cells. The reaction energy and barrier are calculated as ΔE = E FS -E IS and E a = E TS -E IS , where E IS , E FS and E TS are the energies of the corresponding initial state, final state and transition state, respectively.

Data availability
Source data are provided with this paper. The data that support the findings of this study are available from the corresponding authors upon reasonable request.