Unprecedentedly high activity and selectivity for hydrogenation of nitroarenes with single atomic Co1-N3P1 sites

Transition metal single atom catalysts (SACs) with M1-Nx coordination configuration have shown outstanding activity and selectivity for hydrogenation of nitroarenes. Modulating the atomic coordination structure has emerged as a promising strategy to further improve the catalytic performance. Herein, we report an atomic Co1/NPC catalyst with unsymmetrical single Co1-N3P1 sites that displays unprecedentedly high activity and chemoselectivity for hydrogenation of functionalized nitroarenes. Compared to the most popular Co1-N4 coordination, the electron density of Co atom in Co1-N3P1 is increased, which is more favorable for H2 dissociation as verified by kinetic isotope effect and density functional theory calculation results. In nitrobenzene hydrogenation reaction, the as-synthesized Co1-N3P1 SAC exhibits a turnover frequency of 6560 h−1, which is 60-fold higher than that of Co1-N4 SAC and one order of magnitude higher than the state-of-the-art M1-Nx-C SACs in literatures. Furthermore, Co1-N3P1 SAC shows superior selectivity (>99%) toward many substituted nitroarenes with co-existence of other sensitive reducible groups. This work is an excellent example of relationship between catalytic performance and the coordination environment of SACs, and offers a potential practical catalyst for aromatic amine synthesis by hydrogenation of nitroarenes. Modulating the atomic coordination structure has emerged as a promising strategy to further improve catalytic performance. Here, the authors report an atomic Co1/NPC catalyst with unsymmetrical single Co1N3P1 sites that displays high activity and chemoselectivity for hydrogenation of functionalized nitroarenes.

C hemoselective hydrogenation of nitroarenes is a key reaction in the fine chemical industry and has wide applications in the synthesis of pigments and pharmaceuticals [1][2][3][4] . Noble metal nanocatalysts (e.g., Pt, Au, and Pd) are usually used for this reaction [5][6][7] . However, noble metal catalysts are costly and their high activities usually come with unsatisfactory selectivity against many substituted nitroarenes [8][9][10][11] . Since Beller et al. reported highly selective traditional metal catalysts based on Co 3 O 4 @N/C and Fe 3 O 4 @N/C, it has sparked intensive research interest in this type of catalysts 1,12 . Among them, transition metal single-atom catalysts (SACs) with M 1 -N x -C (M = Fe, Co, Ni, x = 2-6) coordination configuration exhibited much better activities than their counterpart nanoparticles while maintaining high selectivity, owing to their maximum atom efficiency and particular electronic structure [13][14][15][16][17][18] . Recently, Wang et al. found that the electron density of Ni single atoms increased with the decrease of Ni-N coordination numbers (CN), and the capability of Ni single sites to dissociate H 2 was greatly enhanced, leading to higher catalytic activity in chemoselective hydrogenation of functionalized nitroarenes 19 . This result suggests that the catalytic activity of M 1 -N x -C can also be enhanced by adjusting the coordination structure of transition metal SACs.
In most SACs, the central metal atoms were stabilized by coordination bonds with N, S, O, etc. atoms within support matrix [20][21][22][23][24][25][26][27][28][29] . The electronic and geometric structures of central metal atoms can be adjusted by tailoring the coordination environment, which would change the absorption energy of reactants on metal atoms and thus influence the catalytic process 20,23,28,30 . For SACs with M 1 -N x -C sites, the symmetric electronic distribution may limit the activation of reactants, thereby leading to hampered catalytic kinetics and performances 20,31 . Recent studies have found that introducing heteroatom P for an unsymmetrical N/P mixed-coordination can further modulate the electronic properties of center metal atoms. The unsymmetrical geometric structure can evoke the distortion of electronic density and alter the d-band center 20,23,28,31,32 . For example, Yuan et al. prepared an N/P dual-coordinated Fe single-atom catalyst, which was more favorable for the adsorption of oxygen intermediates for ORR in fuel cell 23 . Li et al. reported that a Fe 1 -N 3 P 1 single-atom nanozyme exhibited peroxidase-like catalytic activity, and the high activity was ascribed to the less positive charge on Fe atoms as P atoms are electron donors 32 . Thus, we anticipated that constructing the unsymmetrical N/P dual-coordinated transition metal SACs would improve the catalytic performance for the hydrogenation of nitroarenes.
In this work, we report an N/P dual-coordinated Co SAC (denoted as Co 1 /NPC) with Co 1 -N 3 P 1 coordination structure and investigate its catalytic performance for chemoselective hydrogenation of nitroarenes. The single atomic feature and coordination structure of the Co 1 -N 3 P 1 site are characterized through aberration-corrected high angle annular dark-field scanning transmission electron microscopy (AC HAADF-STEM), atomicresolution electron energy-loss spectroscopy (EELS), X-ray photoelectron spectroscopy (XPS), and X-ray absorption spectrum (XAS). In nitrobenzene hydrogenation reaction, the Co 1 -N 3 P 1 SAC exhibits a turnover frequency of 6560 h −1 , which is 60 times higher than that of Co 1 -N 4 SAC and 10 times higher than the state-of-the-art M 1 -N x -C SACs in literatures. Furthermore, Co 1 -N 3 P 1 SAC shows superior selectivity (>99%) toward many substituted nitroarenes with the co-existence of other sensitive reducible groups. The unprecedentedly high activity of Co 1 -N 3 P 1 SAC can be ascribed to the upshift d-band center of Co single atoms, which is more favorable for H 2 dissociation as verified by the kinetic isotope effect and density functional theory calculation results. This is an excellent example of such an unsymmetrical N/ P dual-coordinated structure of metal SACs in hydrogenation.

Results
Structural characterization. Supplementary Fig. 1 illustrates the synthesis procedures for preparing N/P dual-coordinated Co SAC (denoted as Co 1 /NPC) via a two-step process. First, tannic acid, (2-Aminoethyl)phosphonic acid (AePA), and cobalt ion precursors were co-adsorbed on the surface of graphitic carbon nitride (g-C 3 N 4 ) nanosheets; then the resultant powder was subjected to pyrolysis under flowing Ar gas at 900°C to obtain Co 1 /NPC, where the AePA was absent and the introduced-P species served as the donors for anchoring Co atoms (Supplementary Fig. 2). For comparison, N-coordinated Co SAC (denoted as Co 1 /NC) was also prepared via the same procedure without the addition of AePA. As exhibited in Raman spectra, the carbon matrices in both Co 1 /NPC and Co 1 /NC were disordered with a large number of defects ( Supplementary Fig. 3). Only two broad peaks at~24.3°and 42.6°could be observed from their X-ray diffraction (XRD) patterns ( Supplementary Fig. 4), corresponding to (002) and (101) planes of carbon, suggesting highly dispersed states of Co species in both of two samples. Further increasing the pyrolysis temperature of Co 1 /NPC to 1000°C led to the formation of Co 2 P nanoparticles (Supplementary Figs. 4,5). Scanning electron microscopy (SEM) and transmission electron microscopy (TEM) images show that both catalysts retain a two-dimensional layered structure and no obvious nanoparticles are observed (Supplementary Figs. 6,7). Energy-dispersive spectroscopy (EDS) mappings reveal Co elements are distributed uniformly over the entire samples (Supplementary Figs. 8,9). Additionally, Co single-atom feature in Co 1 /NPC and Co 1 /NC is directly observed by AC HAADF-STEM, as reflected by the highly dispersed bright dots due to the heavy Z-contrast ( Fig. 1a and Supplementary Fig. 10). The Co loading in Co 1 /NPC was 0.45 wt% as determined by the inductively coupled plasma mass spectroscopy (ICP-MS) analysis (Supplementary Table 1). The porosity features of Co 1 /NC and Co 1 /NPC were investigated using nitrogen physisorption measurements. Both catalysts displayed characteristics of IV-type N 2 adsorption-desorption isotherms, suggesting the existence of mesopores, which would be beneficial for the exposure of active sites and mass transportation ( Supplementary Fig. 11). The calculated BET-specific surface area of Co 1 /NC and Co 1 /NPC were 512 and 471 m 2 g −1 , respectively.
XPS was then applied to reveal the chemical structures of both Co SACs. In N 1s spectra, besides pyridinic N, pyrrolic N, graphitic N, and oxidized N species, a peak at 399.1 eV corresponding to Co-N can be distinguished 33 , indicating the existence of N coordination environment with Co single atoms in both catalysts ( Fig. 1b and Supplementary Fig. 12). Note that from P 2p spectra in Fig. 1c, an obvious peak at~129.3 eV corresponding to Co-P bond was presented in Co 1 /NPC 23,31 , which can also be observed in the comparison Co 2 P NPs/C ( Supplementary Fig. 13); while it is absent in Co 1 /NC sample. These results suggest that the atomically dispersed Co atoms possess N/P dual-coordinated configuration in Co 1 /NPC, while only N-coordinated configuration in Co 1 /NC.
To further determine the coordination environment of Co single atoms, X-ray absorption fine structure (XAFS) measurements were conducted. Figure 1d shows the Co K-edge X-ray absorption near-edge structure (XANES) curves of Co 1 /NC and Co 1 /NPC, with Co foil, CoO, and cobalt phthalocyanine (CoPc) as reference samples. It can be seen that the absorption threshold positions for Co 1 /NC and Co 1 /NPC are located between Co foil and CoO, suggesting that the valence states of Co species are between 0 and +2 in both two catalysts. Moreover, the Co K-edge position and white line of Co 1 /NPC are lower than that of Co 1 / NC (inset of Fig. 1d), which indicates that Co atoms in Co 1 /NPC possess more negative charges than Co 1 /NC. Such difference could be attributed to the less electron transfer from Co to P because of the weaker electronegativity of P than N atoms 20 . The Fourier-transformed k 3 -weighted EXAFS (FT-EXAFS) spectra demonstrated that both Co 1 /NC and Co 1 /NPC only exhibited a prominent peak at 1.38 Å (without phase shift), no Co-Co peaks at 2.17 Å or larger bond distances were detected, confirming atomically dispersed Co species in Co 1 /NC and Co 1 /NPC (Supplementary Figs. 14, 15). The coordination configuration of Co moieties was further surveyed using quantitative least-squares EXAFS curve-fitting. The EXAFS spectrum of Co 1 /NPC was investigated by utilizing Co-N and Co-P backscattering pathways. The best-fitting analysis displays that the main peak at 1.38 Å could be satisfactorily interpreted as Co-N first-shell coordination with CN = 3.2 ± 0.1 and the shoulder peak at 1.77 Å originated from Co-P contribution with CN = 0.9 ± 0.1 (Fig. 1e and  Supplementary Table 2 Table 2). In order to better confirm the proposed configurations, the theoretical XANES spectrum were simulated based on the Co 1 -N 3 P 1 model as well as Co 1 -N 4 , which presented a good agreement with the experimental data, indicating the rationality of the two structures ( Fig. 1f and Supplementary Fig. 16).
Besides, we performed atomic-resolution electron energy-loss spectroscopy (EELS) analysis at a relatively low beam current to minimize electron-beam perturbations to provide strong evidence of Co 1 -N 3 P 1 structure (Fig. 2a, b). The extracted Co L-edge EELS spectrum from Fig. 2d presents a clear Co signal (Fig. 2g), providing direct evidence for the presence of atomically dispersed Co species. More importantly, the existence of N, P dualcoordination vicinal to Co site is revealed by identifying the surrounding heteroatoms. From the N, P, and overlap maps ( Fig. 2e, f, c), three N atoms (green) and one P atom (red) exist around the Co site. Atomic-scale N K-edge and P L-edge EELS spectra collected at the corresponding positions from Fig. 2e, f are further demonstrated by the N and P signals (Fig. 2h, i). This forcefully confirms the Co 1 -N 3 P 1 configuration in Co 1 /NPC sample. Moreover, the formation energy of the Co 1 -N 3 P 1 structure in the Co 1 /NPC sample was estimated to be about −0.864 eV by DFT calculation, indicating the high stability of the proposed configuration ( Supplementary Fig. 17). These results revealed that the Co single sites in Co 1 /NPC were stabilized with N/P dual-coordinated structure, forming an unsymmetrical Co 1 -N 3 P 1 geometric configuration (as depicted in Fig. 1f), which is different from the Co site in Co 1 /NC with traditional in-plane Co 1 -N 4 configuration.
Catalytic performance. To evaluate the catalytic performances of the as-prepared Co SACs for the hydrogenation of nitroarenes, nitrobenzene is first chosen as a probe molecule. The reaction kinetics with Co 1 /NC and Co 1 /NPC were obtained at 110°C with 3 MPa H 2 in a Teflon-lined stainless steel autoclave. As shown in Fig. 3a, Co 1 /NPC exhibits significantly higher activity than that of Co 1 /NC. Nitrobenzene was completely converted with >99% amine selectivity in 210 min with Co 1 /NPC, while less than 20% conversion was observed with Co 1 /NC under the same reaction condition. In addition, no conversion was observed with NC and NPC supports, suggesting that atomic Co site was active species in both Co 1 /NC and Co 1 /NPC catalysts ( Supplementary Fig. 18). The reaction rate (k) for hydrogenation of nitrobenzene over Co 1 / NPC could reach as high as 35.9 mol mol −1 min −1 , which is ten times higher than that with Co 1 /NC (3.1 mol mol −1 min −1 ).
The turnover frequency value (TOF) (based on the substrate conversion at about 20%) of Co 1 /NPC is calculated to be 6560 h −1 , which is over 60 times higher than that with Co 1 /NC (108 h −1 ) (Fig. 3b). Besides nitrobenzene, Co 1 /NPC also exhibited superior high activity and excellent selectivity (>99.7%) for hydrogenation of 3-nitrostyrene with TOF of 4499 h −1 . The impressive activity of Co 1 / NPC is ten times higher than the state-of-the-art M 1 -N x -C SACs in  Table 3) [34][35][36][37][38][39][40][41][42][43] . Such excellent catalytic performance of Co 1 /NPC sample inspired us to carry out the reaction under milder conditions (e.g., 40°C, 1 bar H 2 ). A high nitrobenzene conversion of 97.2% was achieved within 5 h ( Supplementary Fig. 19). To further compare the catalytic performance between Co 1 /NC and Co 1 /NPC, the apparent activation energies of these two catalysts were measured (Supplementary Fig. 20 and Supplementary Table 4). As shown in Fig. 3d, the calculated activation energy of Co 1 /NPC catalyst is about 21.8 kJ mol −1 , which is much lower than that of Co 1 /NC (51.3 kJ mol −1 ).
In order to clarify the intrinsic higher activity of Co 1 -N 3 P 1 , the electronic properties of the central metal sites over Co 1 /NC and Co 1 / NPC are examined by electron-density isosurface and partial density of states from DFT calculations. Different charge distributions of the two models are observed (Fig. 3e). Compare to the Co 1 -N 4 configuration, the symmetric electron structure is broken by introducing heteroatom P in Co 1 -N 3 P 1 . The Bader charge of the Co 1 -N 3 P 1 site is estimated to be +0.81 e, while the Co 1 -N 4 site is +0.97 e, indicating the Co atom in the Co 1 -N 3 P 1 site carries more charge since the P element in the Co 1 /NPC transfers 2.308 e to support, which is consistent with the XAFS results. Moreover, the Co d-band center of Co 1 /NPC is up-shifted, much closer to the Fermi level (Fig. 3f). As a result, the antibonding state of Co atoms and adsorbed H 2 species are more occupied, then such change enhances the capabilities of H 2 dissociation 6,44 . Thus, Co 1 /NPC catalyst with Co 1 -N 3 P 1 configuration exhibits much higher activity than that of Co 1 /NC with Co 1 -N 4 configuration.
Catalytic hydrogenation mechanism. Such a large activity difference between Co 1 /NC and Co 1 /NPC implies that the hydrogenation activity is closely correlated with local coordination structure, which influences the electronic structure of Co single atoms. In order to elucidate the reaction mechanisms on both catalysts, we carried out a kinetic isotope effect (KIE) study to examine the H 2 dissociation step. Using D 2 for nitrobenzene hydrogenation, the reaction rate was slowed down by a factor of 3.25 for Co 1 /NC (Fig. 4a). For comparison, a larger KIE (k H / k D = 5.54) was observed on Co 1 /NPC catalyst (Fig. 4b). These results suggest that H 2 dissociation undergoes heterolytic cleavage on both Co 1 /NC and Co 1 /NPC [45][46][47] .
It is generally accepted that heterolytic cleavage of H 2 on metal single atoms occurs to form metal-H δand heteroatom-H δ+ 45,48 . Therefore, DFT calculations were further performed to understand the H 2 heterolytic cleavage on Co 1 /NC and Co 1 /NPC, respectively. As shown in Supplementary Fig. 21, the H 2 molecule is adsorbed on the Co atom of Co 1 /NPC with adsorption energy of −0.08 eV, and the bond length of H 2 is 0.814 Å, which is much longer than free molecular H 2 (0.752 Å) in the gas phase. While H 2 molecule is adsorbed on Co atom of Co 1 /NC with adsorption energy of 0.033 eV, and the bond length of H 2 is 0.783 Å. The much longer H-H bond length of adsorbed molecule H 2 indicates that Co 1 /NPC has a higher activation ability for H 2 dissociation than Co 1 /NC. Subsequently, one of the H atoms moves to a nearby heteroatom (N and P) to yield heteroatom-H δ+ , leaving another H atom on Co atom as Co-H δ-. Direct dissociation of H 2 on both Co 1 /NPC and Co 1 /NC in the absence of water, the transition state is almost the same (~1.21 eV), the only difference is that the dissociation of H 2 on Co 1 /NPC is exothermic by 0.20 eV, while on Co 1 /NPC is endothermic by 1.11 eV, suggesting such dissociation manner on Co 1 -N 4 site is thermodynamically unfavorable (Supplementary Fig. 22). The above difference in the DFT calculations confirms that Co 1 -N 3 P 1 exhibits much higher catalytic activity for heterolytic cleavage of H 2 . Ding et al. 18 . reported that the protic solvents play a dominant role in the case of Co-N-C-catalyzed hydrogenation of nitroarenes, where the solvent-mediated H-shuttling mechanism is crucial in the reaction pathway. Compared to the intrinsic hydrogen transfer, the protic solvent-mediated one usually possesses a lower activation barrier, leading to an enhancement of hydrogenation activity in the presence of water or alcohol 49 . Indeed, both Co 1 /NC and Co 1 /NPC show the best activities under ethanol/water solvent and significantly decreased activities in an aprotic solvent such as toluene, acetonitrile, THF, and n-hexane ( Supplementary Fig. 23). Further DFT calculations suggest that the activation energy barriers with water-mediated H-shuttling mechanism for the heterolytic cleavage of H 2 are lower by about 0.16 and 0.01 eV than that through direct dissociation on Co 1 -N 3 P 1 and Co 1 -N 4 sites, respectively (Fig. 4c, d). Both kinetic and thermodynamic results suggest that the dissociative activation of H 2 with help of H 2 O is more favorable to Co 1 /NPC catalyst.
According to the previous reports, the hydrogenation reduction of nitrobenzene to aniline follows the Haber mechanism 7 , namely, PhNO 2 * → PhNOOH* → PhNO* → PhNOH* → PhNHOH* → PhNH* → PhNH 2 *. Based on the above results and reported mechanism in literature, the reaction pathway for hydrogenation of nitrobenzene over Co 1 -N 3 P 1 catalyst is further proposed by virtue of DFT calculations, as shown in Fig. 5 and Supplementary  Fig. 24. One H 2 molecule first goes through heterolytic cleavage with the assistance of the H 2 O-mediated H-shuttling mechanism to form Co-H δand P-H δ+ at Co 1 -N 3 P 1 sites, which can serve as the initial state for the hydrogenation process ( Fig. 5a and Supplementary Fig. 25). Then, the target nitrobenzene molecule was adsorbed on the Co 1 -N 3 P 1 site with a free energy of −0.78 eV (Fig. 5b, I) Supplementary Fig. 26). After that, another H 2 molecule is dissociated to form an activated H atom, which attacks the oxygen atom of PhNO and reduces it to PhNHOH (III, IV). It is worth noting that the adsorption energies of intermediates PhNO and PhNHOH on the Co 1 -N 3 P 1 site are more favorable than Ph-NO 2 , ensuring the reaction progress of the targeted substrate (Supplementary Fig. 27). In the next step, the third H 2 molecule participates in and the formed H atom interacts with PhNHOH to generate the final PhNH 2 product (V, FS). It can be seen the whole process is highly endothermic, confirming the possibility of the proposed reaction path.
Furthermore, Co 1 /NPC exhibited tolerable stability. As shown in Supplementary Fig. 28, a slight decrease of activity is observed after five cycles with Co 1 /NPC, which can be ascribed to the loss of some catalysts and active Co species during the recycling experiments (Supplementary Table 1). The AC HAADF-STEM image and Co K-edge EXAFS spectrum of spent Co 1 /NPC indicate that the atomically dispersed Co species is well preserved after five cycles (Supplementary Fig. 29). All these results demonstrate that the Co 1 /NPC catalyst with unsymmetrical Co 1 -N 3 P 1 configuration possesses unprecedented high activity, high selectivity, and good stability to a wide scope of substrates for hydrogenation of nitroarenes.

Discussion
In summary, we produced an atomically dispersed Co 1 /NPC catalyst with an unsymmetrically Co 1 -N 3 P 1 coordination structure. Due to the increased electron density and upshift d-band center of Co atoms in Co 1 -N 3 P 1 , H 2 dissociation was proved to be more favorable, resulting in much enhanced catalytic activity. In nitrobenzene hydrogenation reaction, the as-prepared Co 1 -N 3 P 1 SAC exhibited a 60-fold higher TOF value (6560 h −1 ) than that of Co 1 -N 4 SAC and more than tenfold higher than the state-of-theart M 1 -N x -C SACs in literature. In addition, Co 1 -N 3 P 1 SAC also displayed superior selectivity (>99%) towards the substituted nitroarenes with the co-existence of other sensitive reducible groups. This work provides new insight into rationally modulating the coordination structure of central metal atoms for boosting the catalytic performance of SACs in heterogeneous catalysis.

Methods
Synthesis of Co 1 /NPC and Co 1 /NC. In a typical procedure, Co(NO 3 ) 2 ·6H 2 O (6.5 mg), tannic acid (TA, 500 mg), and (2-Aminoethyl) phosphonic acid (AePA, 126 mg) were dissolved into 30 mL DI water at 100°C (marked as solution A). g-C 3 N 4 nanosheets (1 g) were dispersed well in 100 mL DI water with ultrasound (marked as solution B). Then, solution A was added dropwise into solution B with a strong stirring at 100°C until the mixed system was forced to yield a slurry. Subsequently, the obtained powder after freeze-dried was pyrolyzed at 900°C for 2 h under Ar atmosphere. Finally, the as-prepared material was directly used without further treatment, denoted as Co 1 /NPC. The synthesis process for Co 1 /NC is the same as that of Co 1 /NPC except without the addition AePA.  Scientific ESCALab220i-XL electron spectrometer using 300 W Al kα radiation. Inductively coupled plasma atomic emission spectroscopy (ICP-AES) was conducted on a Shimadzu ICPE-9000 to confirm the loading content of metal on the catalysts. The AC HAADF-STEM images were carried out in a JEOL ARM300F at 300 kV, equipped with a probe spherical aberration corrector. Brunauer-Emmett-Teller (BET) surface areas were measured by N 2 adsorptiondesorption isotherms at 77 K with a Micromeritics ASAP 2460 instrument. The HAADF imaging and EELS mapping were both performed using a Nikon HERMES-100 aberration-corrected scanning transmission electron microscope under 60 kV accelerating voltage with a~22 pA probe. The probe convergence semi-angle, HAADF collection semi-angle, and EELS collection semi-angle is 32 mrad, 75-210 mrad, and 0-75 mrad, respectively.
XAS measurements and analysis. The cobalt K-edges XAFS spectra of the standards and samples were collected at the beamline 1W1B of the Beijing Synchrotron Radiation Facility (BSRF). The typical energy of the storage ring was 2.5 GeV and the electron current was~250 mA in the top-up mode. The white light was monochromatized by a Si (111) double-crystal monochromator and calibrated with a Co foil (K-edge at 7709 eV). Samples were pressed into thin slices and positioned at 45°to the incident beam in the sample holder. The XAFS spectra were recorded in fluorescence mode with a Lytle detector oriented at 90°to the incoming beam.
The XAFS data were analyzed using the software packages Demeter 50 . The spectra were normalized using Athena firstly, and then shell fittings were performed with Artemis. The χ(k) function was Fourier-transformed (FT) using k 3 weighting, and all fittings were done in R-space. The coordination parameters of samples were obtained by fitting the experimental peaks with theoretical amplitude. The quantitative curve-fittings were conducted with a Fourier transform k-space range of 2.7-11.8 Å −1 . The backscattering amplitude F(k) and phase shift Φ(k) were calculated by the FEFF7.0 code. While the curve-fitting, all the amplitude reduction factor S 0 2 was set to the best-fit value of 0.87 determined from fitting the data of cobalt foil by fixing coordination numbers as the known crystallographic value. In order to fit the curves in the R-range of 1.0-2.0 Å, we considered Co-N and Co-P paths as the central-peripheral. For each path, the structural parameters, like coordination number (CN), interatomic distance (R), Debye-Waller factors (σ 2 ), and inner potential correction (ΔE 0 ) were opened to be varied.
Catalytic performance evaluation. For the liquid phase hydrogenation of nitrobenzene, substrates (2 mmol), catalyst (5 mg), and solvent (EtOH/H 2 O, v-v = 4:1, 40 mL) were added into a 100 mL high-pressure autoclave. Then the autoclave was flushed three times with H 2 and charged to certain pressure (3 MPa H 2 ). The reaction was performed at the desired temperature. The product was collected at the reserved time and immediately analyzed using gas chromatography in combination with mass spectrometry (Shimadzu GCMS-QP2010S). The TOF values were determined within the substrate conversion below 20%, and the calculation of TOF was based on the total Co amount in catalysts. To evaluate the reusability of catalyst, the samples from the last reaction were separated by centrifugation, washing with ethyl acetate to remove the substrate, and drying under the vacuum. For the kinetic isotope effect test, D 2 and D 2 O are used instead of H 2 and H 2 O.
DFT calculations. All the spin-polarized first-principles calculations have used the code VASP [51][52][53] . Valence electrons of O(2s, 2p), N(2s, 2p), H(1s), C(2s, 2p), P(3s, 3p), and Co(3d, 4 s) were treated on a basis of plane waves explicitly 54 , while the core electrons were described with the projector-augmented wave method 55 . Spinpolarized calculations were carried out at the level of the generalized gradient approximation (GGA) adopting the Perdew, Burke, and Ernzerhof (PBE) exchange-correlation functional 56 . A kinetic energy cutoff of 400 eV was used for all calculations. The truncation criteria for the electronic and ionic loops were 10 −5 eV and 10 −2 eV/Å, respectively. Long-range dispersion was included according to the D 3 method introduced by Grimme 54 . The vacuum layer was set to 20 Å to avoid interaction from adjacent cells. All the transition states were determined by using the climbing image nudged elastic band (CINEB) method 57,58 and transition states were characterized via frequency analysis to ensure a single imaginary frequency in the desired reaction direction. The pure graphene is modeled by a (7 × 7) supercell with 49 carbon atoms, and the Co-N x P y -Gr model is modeled by one cobalt atom adsorption at vacancy site which is composed by getting rid of two carbon atoms and the N and P atoms substitute for four carbon atoms around cobalt atom, respectively. Monkhorst-Pack (5 × 5 × 1) Γ-centered grid sampling for the Brillouin zone was used for geometry optimization, and dipole corrections were included in the z-direction for each model surface. To study the stability of Co binding on graphene with N, P dual-coordination, the formation energy, and adsorption energy were defined as follows: E for ¼ EðCo-N x P y -GrÞ À EðN x P y Þ À EðCo-bulkÞ ð1Þ E ads ¼ EðCo-N x P y -GrÞ À EðN x P y Þ À EðCoÞ In the equations, E(Co-N x P y -Gr) is the total energy of adsorbed systems, E(N x P y ) is the energy of graphene with doped N and P, E(Co-bulk) is the energy of one atom in the most stable Co crystal, and E(Co) is the energy of cobalt in the gas phase.   Fig. 5 Reaction mechanism of the nitrobenzene reduction. a The proposed reaction pathway for the hydrogenation of nitrobenzene to aniline at Co-P interface site. b Energy profile of hydrogenation of nitrobenzene over Co 1 -N 3 P 1 site.

Data availability
The data supporting the findings of this study are available within the article and its Supplementary Information. Source data are provided with this paper. Additional data are available from the corresponding authors on reasonable request.