Highly efficient ammonia synthesis at low temperature over a Ru–Co catalyst with dual atomically dispersed active centers

The desire for a carbon-free society and the continuously increasing demand for clean energy make it valuable to exploit green ammonia (NH3) synthesis that proceeds via the electrolysis driven Haber–Bosch (eHB) process. The key for successful operation is to develop advanced catalysts that can operate under mild conditions with efficacy. The main bottleneck of NH3 synthesis under mild conditions is the known scaling relation in which the feasibility of N2 dissociative adsorption of a catalyst is inversely related to that of the desorption of surface N-containing intermediate species, which leads to the dilemma that NH3 synthesis could not be catalyzed effectively under mild conditions. The present work offers a new strategy via introducing atomically dispersed Ru onto a single Co atom coordinated with pyrrolic N, which forms RuCo dual single-atom active sites. In this system the d-band centers of Ru and Co were both regulated to decouple the scaling relation. Detailed experimental and theoretical investigations demonstrate that the d-bands of Ru and Co both become narrow, and there is a significant overlapping of t2g and eg orbitals as well as the formation of a nearly uniform Co 3d ligand field, making the electronic structure of the Co atom resemble that of a “free-atom”. The “free-Co-atom” acts as a bridge to facilitate electron transfer from pyrrolic N to surface Ru single atoms, which enables the Ru atom to donate electrons to the antibonding π* orbitals of N2, thus resulting in promoted N2 adsorption and activation. Meanwhile, H2 adsorbs dissociatively on the Co center to form a hydride, which can transfer to the Ru site to cause the hydrogenation of the activated N2 to generate N2Hx (x = 1–4) intermediates. The narrow d-band centers of this RuCo catalyst facilitate desorption of surface *NH3 intermediates even at 50 °C. The cooperativity of the RuCo system decouples the sites for the activation of N2 from those for the desorption of *NH3 and *N2Hx intermediates, giving rise to a favorable pathway for efficient NH3 synthesis under mild conditions.


Introduction
The N 2 -to-NH 3 conversion is one of the most important reactions for human society.2][3][4][5][6][7] The potential utilization of NH 3 in the coming chemical industry revolution is highly promising. 8][12] Traditionally, the H 2 supply for the HB process is mainly from coal or natural gas through the water-gas shi (WGS) or methane reforming reaction, 13 and these processes account for the energy requirement and carbon dioxide production.Therefore, it is imperative to develop new HB technology that is both environmentally friendly and energy saving.Recently, with the green generation of electricity using renewable sources (such as hydro, wind, solar and tidal), it becomes economically acceptable to use the H 2 produced by water electrolysis, and NH 3 can be synthesized from renewable H 2 and N 2 via the electrolysis driven Haber-Bosch (eHB) process.
Nowadays, the subsequent NH 3 re-conversion to H 2 and the handling as well as shipping infrastructure including regulations for transportation are already in place. 64][15] However, the present bottleneck is that NH 3 cannot be synthesized under mild conditions.Currently, huge plants are required, and for a plant with an annual capacity of 200 000 tons, the investment is at the scale of US$ 1000-2000 per ton of ammonia. 7The aim is to match the pressure of the NH 3 synthetic system with that of the electrolysis system employed for H 2 production (<5 MPa, mostly lying in the range of 1.0-3.2MPa) so as to avoid expensive pressure ramping. 15Therefore, advanced catalysts that are adoptable to an efficient eHB process are urgently needed, which can activate N 2 for NH 3 synthesis at lower pressure.
9][30][31] The d-band center of such metals results in strong binding of Ncontaining intermediates, which require high temperature to desorb.Upon regulating the surface d-band centers of TMs, it is possible to simultaneously tune the binding energy of N 2 adsorption and that of surface N-containing intermediates.By so doing, the scaling relation for TMs can be decoupled.Indeed, the surface d-band of a metal-based catalyst can be effectively tuned through the strategy of single-atom catalysts (SACs) [32][33][34] or single-cluster catalysts (SCCs), 23,35 leading to d-band adsorption modes different from those of pure metals and conventional alloy compositions. 36,37The unusual electronic behavior of SACs and SCCs could also change the adsorption properties and bonding abilities of each metal atom, 30,38 resulting in signicant deviations from the scaling relationship. 36,39,40Among the SCCs, dual single-atom catalysts (DSACs) offer the simplest active sites for catalytic reactions.
For the well-known active Ru and Co metals in NH 3 synthesis, the Co d-band is narrower than that of Ru because of the quantum primogenic effect, 41 resulting in the N 2 dissociative adsorption energy of Co being much lower than that of Ru.Despite the fact that the state-of-the-art Ru metal catalysts are efficient for NH 3 synthesis, the loading and high cost limit the large-scale use of the noble metal.It is hence advantageous to achieve efficient utilization of Ru by having a small amount of Ru atomically dispersed on atomistic Co to generate a RuCo DSAC.3][44] As there is so far no experimental demonstration on the use of DSACs for NH 3 synthesis, we are interested in exploring whether the RuCo DSAC is capable of decoupling the scaling relationship and which mechanism this DSAC system might adopt.Especially, the following issues deserve investigation: (1) the existing form of dual active centers, (2) the state of the surface dband, (3) the pathway of N 2 activation and dissociation, and (4) how to decouple the scaling relation.
Herein, we report the preparation of a RuCo DSAC by introducing a Ru atom on a surface layer of Co-N-C material with Co atoms coordinated by a pyrrolic N of g-C 3 N 4 .It is found that with Ru atomically anchored on the surface of single-site Co, there is signicant overlapping of Co t 2g and e g orbitals, resulting in an electronic structure resembling that of a "free" Co atom.Our experimental and theoretical studies demonstrate that the cooperativity of Ru and Co dual single-atom active centers could decouple the scaling relationship by separating N 2 adsorption and activation (on the Ru sites) from H 2 dissociative adsorption as well as N-containing intermediate species (mainly *N 2 H 4 and *NH 3 ) desorption (on the Co sites).By so doing, the scaling relation over RuCo DSAC in NH 3 synthesis can be decoupled, and the developed RuCo DSAC could efficiently promote NH 3 synthesis at 200 C, giving an NH 3 synthesis rate of up to 1.24 mmol NH 3 g cat À1 h À1 .We believe the acquired understanding can help to accomplish rational design of transition-metal-based DSACs for the decoupling of the scaling relation to achieve high catalytic efficiency under mild conditions as well as to develop advanced catalysts to achieve pressure matching between the electrolysis and catalytic systems in eHB technology.

Conrmation of Ru and Co atomic dispersion
To highlight the effect of the DSAC active centers, monometallic Ru/N-C and Co/N-C (denoted hereinaer as Co SAC) were also prepared for comparison purposes.Representative scanning electron microscopy (SEM) images demonstrate that the morphology of RuCo DSAC (Fig. 1a) is similar to that of Ru/N-C (Fig. S1a †) and Co SAC (Fig. S1b  According to the AC-STEM results, there is distribution of Ru single atoms as well as small clusters on Ru/N-C (Fig. 1c and d, and S4a-d †); the latter is a result of Ru aggregation as indicated by the individual bright dots of different intensity levels.Moreover, the AC-STEM images (Fig. 1e and f, and S5a-d †) taken at different regions of Co SAC show the predominant presence of Co single atoms (Fig. 1e and f, and S5a-d, related descriptions are provided in the ESI †).Interestingly, RuCo DSAC exhibits many individual bright dots (Fig. 1g and h).The dots over RuCo DSAC are repeatedly found at different magni-cations (Fig. S6a-c †).Most of these dots show no obvious difference (Fig. 1g and h), and the results show that Ru and Co are uniformly dispersed.The result agrees with the fact that there is formation of dual single-atom Ru and Co as conrmed by extended X-ray absorption ne structure (EXAFS) measurements over RuCo DSAC.The conclusion is in accordance with the fact that there is a Ru-Ru bond attributable to Ru clusters as revealed by EXAFS measurements over Ru/N-C (Fig. S7 †).
The EXAFS Ru K-edge (Fig. 2a) and Co K-edge (Fig. 2b and S8 †) spectra show no obvious Ru-Ru or Co-Co interaction over RuCo DSAC and Co SAC (Fig. S9 †), providing direct evidence for the presence of individually isolated Ru and/or Co atoms in the samples.Assignment of signals from 1 to 4 Å in RoCo DSAC is further conrmed by detailed WT-EXAFS wavelet transform plots (Fig. S10 †).A WT intensity maximum near 4.5 ÅÀ1 can be assigned to the Co-N contribution.Moreover, the EXAFS tting curves (Fig. S11a-d †) show that Ru (Fig. S11b †) and Co (Fig. S11d †) atoms over RuCo DSAC are coordinated with nitrogen atoms, and the coordination numbers (CN) of Ru-N and Co-N over RuCo DSAC are 2.9 AE 0.5 and 2.9 AE 0.9 (Table S2 †), respectively, while the CN for Co-Ru is around 2.1.
We then performed XPS and XANES measurements to investigate the chemical state of Ru and Co species.XPS study results show that the surface Ru content in the case of RuCo DSAC is close to the total Ru content (total Ru content is determined by ICP-AES measurements).Notably, there is no detection of the Ru 3p signal aer deep etching by Ar + (Table S3 †).These results indicate that Ru mainly disperses on the surface of DSAC.Moreover, the energy absorption edge of Ru Kedge XANES over RuCo DSAC (Fig. 2c) is higher than that of Ru foil but lower than that of RuO 2 , suggesting an oxidation state of Ru n+ (0 < n < 4) rather than Ru 4+ or Ru 0 . 45Similarly, the Co Kedge XANES spectra (Fig. 2d) show that the absorption edge position of RuCo DSAC locates between that of Co foil and CoO bulk, suggesting that the single Co atom carries positive charge, which is in accord with the XPS Co 2p study (Fig. S12a †).The XPS binding energy (BE) of the Co 2p 3/2 peak (Fig. S12a †) over RuCo DSAC is higher than that of Co 2+ (779.2 eV), and lower than that of Co 3+ (781.0 eV), 46 suggesting that the surface Co atom has a hetero-valence state of Co 2+ and Co 3+ .Additionally, the absorption edge of Ru K-edge over RuCo DSAC (Fig. S13 †) is lower than that of Ru/N-C, while the BE of the XPS Co 2p 3/2 peak for RuCo DSAC is higher than that for Co SAC (Fig. S12a †), indicating that there is electron transfer from Co to Ru via the Ru-Co bond, 47 plausibly due to the higher electronegativity of the Ru atom. 48Meanwhile, the XPS Co 2p peaks of the RuCo DSAC sample signicantly shi to lower BE values (À1.58 AE 0.7 eV, Fig. S12b †) when it is subject to Ar + etching at different depths.These observations further suggest electron redistribution via electron transfer from Co to the surface Ru atom.
"Free-Co-atom" and electron transfer Additional evidence of electron transfer between Co and Ru can be obtained by electron paramagnetic resonance (EPR) and NEXAFS analyses.According to the EPR spectra of Fig. 3a, the g value (3.0) of RuCo DSAC can be ascribed to an unpaired electron in the 3d x 2 Ày 2 orbital of Co II and Ru III . 49In comparison with Co SAC, the value shi in g value and the shape broaden asymmetrically for RuCo DSAC, indicative of a dipolar broadening due to electron-electron interaction between atomically dispersed Ru and Co, and the phenomenon supports the idea of charge transfer, 50 in agreement with the observation of Co Kedge NEXAFS results (Fig. 3b).It is to be noted that there is no EPR signal of Ru/N-C because of no obvious electron transfers between the Ru and N atoms.The Co L-edge NEXAFS over RuCo DSAC and Co SAC can be tted into L2 and L3 subbands (Fig. 3b), matching well with the transition from 2p 1/2 and 2p 3/2 levels to the vacant d band, 51,52 respectively.Compared with Co SAC, the absorption Co L-edge of RuCo DSAC shis towards lower excitation energy by 0.6 eV.Meanwhile, the Co Ledge intensity of RuCo DSAC is lower than that of Co SAC (Fig. 3b), implying that the former has higher occupancy of Co 3d electrons.It is to be noted that the extent of electron transfer is dependent on the nature of surface unoccupied Co 3d charge, which could be semi-quantitatively calculated on the basis of the Co L-edge NEXAFS spectra (Fig. S14 †), using the calculation method of Mattheiss and Dietz (see the ESI †). 53The relative number of unoccupied Co 3d charge (Kh T ) in RuCo DSAC (14.7) is higher than that of Co SAC (11.2).The much higher unpaired Co 3d charge of RuCo DSAC indicates the higher feasibility of d electron donation from Co to the unoccupied Ru 3d t 2g (feature L2) and e g (feature L3) states in the RuCo DSAC case. 36wing to the fact that the total electron yield (TEY) is highly surface sensitive, we collected the N K-edge NEXAFS spectra of RuCo DSAC, Co SAC and Ru/N-C to gain insight into the direction of electron transfer (Fig. 3c).Over RuCo DSAC in the range of 397-402 eV, the characteristic p* peaks at 397.5 (peak a), 398.6 (peak b) and 401.2 (peak c) eV can be assigned to pyridinic, pyrrolic and graphitic nitrogen species, respectively. 54,55The presence of the p* peak is attributed to N coordination that involves electron transfer from N to the coordinated metal atom.Based on DFT calculation results including the electron localization function (ELF) (Fig. 3d-f), charge density differences (Fig. 3g and h), and Bader charges (Fig. 3i), it can be conrmed that there is charge transfer from the N atom of g-C 3 N 4 to the Co atom, with loss of negative charge in the former while gain of negative charge in the latter.Then, the Co atom enriched with an electron acts as a bridge to facilitate the transfer of electrons from Co 3d orbitals to the Ru atom.
The extent of electron transfer from Co to Ru is dependent on the nature of the N species involved in the interaction.The chunk in ELF mapping (Fig. 3d-f) demonstrates the electron interaction between Co and Ru, and that the involvement of pyrrolic N species is the strongest (a greener chunk represents weaker electron interaction).The transferred electron density from the nitrogen species to Co and then to Ru follows the order pyrrolic N [ graphitic N > pyridinic N species (Table S4 †).Taking together the N K-edge NEXAFS results (Fig. 3c), because of the absence of the a peak over Ru/N-C, it is deduced that Ru does not coordinate with pyrrolic N. In a previous study of ours, it was conrmed that single cobalt atoms can anchor and get stabilized on pyrrolic N species. 56Then, Co acts as a bridge to facilitate electron transfer from g-C 3 N 4 through pyrrolic N and Co to the surface Ru atom.Consequently, there is a decrease of N 2 adsorption energy, and the N 2 molecule adsorbed on Ru shows a negative charge of À0.39 e À (Fig. 3i).As a site for N 2 adsorption, the electron-enriched Ru atom readily donates electrons to the antibonding p* orbitals of adsorbed N 2 , promoting N 2 activation via the weakening of the N^N triple bond.
The electron release and/or donation behaviors of RuCo could inuence the nature of N 2 adsorption and activation, while the binding and desorption energies of N-containing intermediates are closely related with the d-band centers.Therefore, we investigated the d-band center of the RuCo DSAC surface by ultraviolet photoemission spectroscopy (UPS) and projected density of states (PDOS) calculation.The UPS spectrum (Fig. 4a) shows that RuCo DSAC has a d-band width narrower than that of bulk Co and Ru foil. 57Also, the d-band width of RuCo DSAC is obviously narrower than that of traditional RuCo alloy, as revealed in Fig. S15 (for more details see the ESI †).These results strongly suggest that the d bands of Co and Ru in RuCo DSAC are narrow, which is further implied by the results of PDOS calculation (Fig. 4b-e).For RuCo DSAC, the PDOS of Co over RuCo DSAC (Fig. 4c) signicantly shis towards the Fermi level in comparison with that of bulk Co (Fig. 4b), and the corresponding d-band center of Co shis from À1.30 eV for bulk Co to À1.65 eV for RuCo DSAC (Table S5 †).Also, the shi of the d-band center of Co is obviously more than that of Ru.In comparison with bulk Co (Fig. 4g), the PDOS (Fig. 4h) of RuCo DSAC shows that the e g and t 2g states of Co in RuCo DSAC are almost degenerate, suggesting almost overlapping t 2g and e g PDOS, and there is a nearly uniform coordination eld for the Co 3d states.These results imply that the electronic structure of the Co 3d state in RuCo DSAC resembles that of a "free-Co atom". 39It is expected that such unique electronic structure could promote desorption of N-containing intermediates, and thus have a benecial inuence on lowtemperature NH 3 synthesis.

NH 3 synthesis performance
The catalytic performances of the C 3 N 4 support and assynthesized samples for NH 3 synthesis were evaluated in a 25% N 2 -75% H 2 feed at a WHSV of 60 000 mL g À1 h À1 as a function of reaction temperature.Obviously, there is only insignicant NH 3 production over the C 3 N 4 support within the test temperature range, which is less than 0.04 mmol NH 3 g cat À1 h À1 even at 400 C.Meanwhile, the NH 3 synthesis rates over the three samples in the range of 200-400 C differ signicantly (Fig. 5a).The addition of Ru to Co SAC signicantly enhances the NH 3 synthesis rate (Fig. 5a), and the rate of RuCo DSAC is much higher than that of the monometallic Ru and Co catalysts.For example, the NH 3 synthesis rate over RuCo DSAC at 200 C is 1. 24    originates from the catalytic synthesis of N 2 gas rather than the nitrogen source of N-doped carbon support, the NH 3 synthesis rate of RuCo DSAC has also been measured at 400 C and 1 MPa using the feed gas of 75% H gases.The H 2 -TPD-MS experiment was also carried out (Fig. S21 †).The signal of m/z ¼ 17 is very close to the baseline, showing that the production of ammonia can be ignored.However, a very weak signal of m/z ¼ 32 can be discerned, indicating that there are only tiny amounts of dynamic N in the catalyst that can be reacted with hydrogen.
In addition, the NH 3 synthesis rate over RuCo DSAC at 400 C exhibits an approximately linear increase from 11.20 to 20.39 mmol NH 3 g cat À1 h À1 when the pressure is raised from 1.0 to 5.0 MPa (Fig. S22 †).The catalytic performance of RuCo DSAC at such pressure range should permit convenient eHB operation on a large scale.To further reveal the unique intrinsic catalytic activity, we calculated the turnover frequency (TOF).It is well known that Ru and Co entities are both active for NH 3 synthesis, and therefore it is difficult to differentiate the catalytic contribution of the two.Herein, TOF M was calculated (more details are provided in the Experimental section) to express catalytic activity on a per-M-active-site basis.Interestingly, TOF Ru and TOF Co over RuCo DSAC reach 0.016 s À1 and 6.7 Â 10 À3 s À1 (Fig. S23 †), respectively, at 400 C, which is 10.7-fold that of monometallic Ru/N-C and 5.8-fold that of Co SAC.Therefore, it is reasonable to attribute the effective formation of NH 3 under mild conditions to strong synergism between the dual atomically dispersed Ru and Co active centers.
Previous studies have shown that the addition of a proper promoter to Ru or Co-based catalysts could promote NH 3 synthesis. 58,59In the present work, we studied the effect of adding Ba into the best-performing RuCo DSAC (5 wt% against RuCo DSAC) on the NH 3 production rate.As displayed in Fig. 5a, the results reveal that the NH 3 synthesis rate of Ba/RuCo DSAC is (1.8-3)-fold that of the non-promoted one, depending on the reaction temperature.Surprisingly, when the temperature is 400 C, the NH 3 synthesis rate over Ba/RuCo DSAC is 23.90 mmol NH 3 g cat À1 h À1 .Moreover, we compared the NH 3 synthesis rates of Ba/RuCo DSAC (Table S6 †) and RuCo DSAC (Fig. 5b) with those of selected Ru-and Co-based catalysts, and they are superior to those of the conventional Ru-and Co-based catalysts.
The long-term stability test of RuCo DSAC was conducted at 350 C for 100 h, and its NH 3 synthesis rate remains almost constant aer a slight decrease at the initial stage (Fig. 5c).The used RuCo DSAC was subject to TEM, HR-TEM, AC-STEM and XRD analyses.The TEM (Fig. S24a and b †) and HR-TEM images (Fig. S24c and d  Meanwhile, the H 2 -TPR (Fig. S27 †) prole of the C 3 N 4 support shows no reduction peak below 550 C. Also, the H 2 -TPR prole of RuCo DSAC exhibits no reduction peaks below 500 C. In addition, the Ar-TPD-MS analysis of the as-synthesized fresh catalysts does not show any desorption or decomposition peak below 600 C (Fig. S28 †).These results undoubtedly conrm that the C 3 N 4 support and RuCo DSAC are stable under the reaction for NH 3 synthesis.Evidently, the RuCo DSAC is not only active but also stable under the adopted mild conditions for NH 3 synthesis.

Kinetic studies
The kinetic parameters of NH 3 synthesis such as activation energies and reaction orders over the as-synthesized catalysts are depicted in Fig. 5d-f.The apparent activation energy (E a ) of NH 3 synthesis over the as-synthesized catalysts was derived from the Arrhenius plots depicted in Fig. 5d.The E a is determined to be 58 kJ mol À1 for RuCo DSAC, which is lower than that of Ru/N-C (79 kJ mol À1 ) or Co SAC (65 kJ mol À1 ).The E a value of RuCo DSAC is similar to that of previously reported Ruloaded electrides (50-60 kJ mol À1 ) 11,26,60 and hydrides (49-60 kJ mol À1 ). 6,25Moreover, experiments to determine the reaction orders of N 2 and H 2 over the as-prepared catalysts were performed.For traditional Ru-based catalysts in NH 3 synthesis, the N 2 reaction order is between 0.8 and 1.0 while that of H 2 is negative in value.The former suggests that N 2 dissociation is the rate-limiting step, while the latter indicates the poisoning effect of H 2 on Ru. 61,62 It was observed that the reaction orders of N 2 (Fig. 5e), H 2 (Fig. 5f) and NH 3 (Fig. S29 †) over RuCo DSAC are 0.37, 0.61 and À1.07, respectively.The results of the positive H 2 reaction order indicate that the effect of H 2 poisoning on the Ru sites is circumvented over RuCo DSAC.The reaction orders for the RuCo DSAC are different from those for Ru/N-C and Co SAC, consistent with the presence of the dual single-atom active centers that may create the cooperativity that favors N 2 adsorption/activation and H 2 dissociative adsorption occurring separately on different sites.In this regard, we performed surface-science characterization and theoretical modeling to acquire insights into the roles of the dual single-atom active centers at the atomic level for NH 3 synthesis.

Discussion
A suite of elaborate characterization and DFT calculations were employed to explore the adsorption and activation behavior of N 2 and H 2 .In H 2 -TPD (Fig. S30a †) and N 2 -TPD (Fig. S30b †) studies, the desorption temperatures of H 2 and N 2 over RuCo DSAC and monometallic Ru/N-C as well as Co SAC are slightly different, but they are all below 200 C, suggesting that the adsorption and activation of H 2 and N 2 are not difficult on these catalysts.For Ru/N-C and Co SAC, there is detection of mass signals of m/z ¼ 32 (Fig. S31a †) and m/z ¼ 17 (Fig. S31b †), respectively, corresponding to the desorption of *N 2 H 4 and *NH 3 species accumulated on the catalyst surface during NH 3 synthesis, and desorption of these species can still be observed even at 700 C. It is to be noted that the signal intensity of m/z ¼ 32 and m/z ¼ 17 over the used RuCo DSAC is much lower, the former getting close to the baseline at ca. 200 C, while the latter can be ignored even at 50 C.This observation is in good agreement with the UV-vis DRS experiment (for more details see the ESI †) and DFT calculation reported below, displaying that the energies needed for *N 2 H 4 and *NH 3 desorption over RuCo DSAC are only 0.2 eV and 0.1 eV, respectively.These results strongly suggest facile desorption of surface *N 2 H 4 and *NH 3 intermediates from RuCo DSAC.Moreover, an in situ DRIFTS deuterium labeling investigation was performed to determine the reactivity of intermediate species for NH 3 synthesis.For the fresh C 3 N 4 sample, there is detection of bands at 1637, 1406, 1316, and 1236 cm À1 which are characteristic of the CN heterocycle. 63Aer C 3 N 4 was exposed to a mixture of 25% N 2 -75% D 2 or 25% N 2 -75% H 2 at 400 C for 30 min, there is no detection of any additional IR peaks in comparison with fresh C 3 N 4 (Fig. S32 †).The observation is in accord with the fact that C 3 N 4 shows almost no catalytic activity.Aer Ru/N-C and Co SAC were exposed to a mixture of 25% N 2 -75% D 2 at 400 C, there is detection of bands at 2394 cm À1 that are related to the v 5 (B u )N 2 D 2 transition or N-D torsion modes within the ND 2 or N 2 D 4 fragments, [64][65][66][67][68] and the band intensity signicantly increases with prolonged exposure from 1 to 30 min, further indicating the accumulation of *N 2 D x species on the surface of Ru/N-C (Fig. S31c †) and Co SAC (Fig. S31d †).These results indicate that the surface Ru or Co active sites are partially covered by the intermediate species, which can only desorb at high temperature (>400 C).Interestingly, such a troublesome scenario is virtually eliminated in the case of RuCo DSAC (Fig. S33 †).Specically, the band at 2394 cm À1 is attributed to the v 5 (B u )N 2 D 2 transition or N-D torsion modes within the ND 2 or N 2 D 4 fragments.Two bands located at 2574 cm À1 and 1545 cm À1 can also be discerned, which can be attributed to trans-HNND and NN stretching in N 2 D x species, respectively. 66he formation of trans-HNND may be a result of NND interaction with the surface H le behind in the pretreatment of the catalyst in 10% H 2 /Ar at 400 C for 2 h.These results further conrm the existence of N 2 D x species.The UV-vis DRS spectra (Fig. S34 †) show that the main intermediate species of RuCo DSAC in NH 3 synthesis is N 2 H 4 , and the peak intensity of the N 2 H 4 compound decreases with the increase of reaction temperature.
From these observations and analyses, it follows that *N 2 H 4 is likely the main intermediates under the reaction conditions, and the presence of *N 2 H 4 also suggests that the activation of N 2 is more facile via hydrogenation to NNH than via direct dissociation of N^N triple bonds.Moreover, the accumulation of species such as *N 2 H 4 and *NH 3 on Ru/N-C and Co SAC will block the sites for further activation of N 2 and H 2 , thus hindering NH 3 formation at low temperature over both samples.For RuCo DSAC, the immediate implications of the involvement of the dual single-atom active sites with narrow dband centers are the interruption of the scaling correlation between the adsorption/activation of N 2 and desorption of surface N-containing species, and hence the superior lowtemperature NH 3 synthesis performance upon the decoupling of scaling correlation.
To illustrate the role of dual single-atom active centers and to deduce the possible reaction pathway over RuCo DSAC, DFT calculations were performed and the calculation results are shown in Fig. 6a-e and S35-S39.† From these results, the Ru atom is the dominant site for N 2 adsorption, due to the higher adsorption energy (À0.56 eV, Fig. S36a †) on Ru than on the Co atom (À0.43 eV, Fig. S36c †).Notably, the adsorption of a N 2 molecule on a single Ru atom tends to adopt the side-on coordination (Fig. S36a and c †), due to symmetry matching between the p* orbitals of N 2 and 3d orbitals of Ru.Then, the adsorbed N 2 on the Ru site is more facile to be hydrogenated than direct N 2 dissociation.The energy for directly breaking the N^N bond under the active state on the Ru atom is still up to 2.53 eV, indicating difficult direct dissociation.In general, the cleavage of N^N triple bond needs at least two adjacent Ru atoms. 69,70he absence of Ru-Ru ensembles as revealed in EXAFS and AC-STEM analyses over RuCo DSAC implies that the direct dissociation of N 2 is unlikely, which is also consistent with the determination of *N 2 H 4 desorption in Ar-TPD measurements (Fig. S31a †).Then, the N 2 molecule on the Ru site needs to be activated before the occurrence of hydrogenation (Fig. 6a).The Co site as a bridge accelerates the electron transfer from g-C 3 N 4 to the Ru site, which promotes the donation of 3d electrons from the Ru atom to the p* orbitals of N 2 and weakening of N^N bonds as mentioned earlier. 69n the other hand, the dissociative adsorption of H 2 over the Co active center of RuCo DSAC is efficient (Fig. S37a-d †), where H 2 adsorption is thermodynamically exothermic throughout the entire reaction pathway.Also, the Co atom plays an important role in gathering and activating H 2 for N 2 hydrogenation at the Ru atom.The above analysis indicates that the adsorption and dissociation of N 2 and that of H 2 occur at different active sites.The signicant advantage of RuCo DSAC is that the competitive adsorption of N 2 and H 2 can be effectively avoided, as reected by the positive reaction order of H 2 (Fig. 5f).Comparing the charge density differences (Fig. 6c-e) of N 2 and H 2 coadsorption on RuCo DSAC (Fig. 6c) and Co or Ru SAC (Fig. 6d  and e) sites, electron lling of p* orbitals decreases with the drop of electron density on N 2 , clearly showing that the existence of H 2 poses a negative effect on N 2 activation, and it is more obvious in the latter (charge of N 2 shi more than 0.7 e À ).H 2 -TPD results (Fig. S30a †) show that the amount of H 2 desorption from RuCo DSAC (1.40 mmol g À1 ) is much higher than the Co coverage, therefore it is reasonable to deduce that there is a spillover of H atoms from the Co atom to Ru active centers.
What follows is the attack of activated N 2 on Ru by spillover H atoms to generate N 2 H x species.Note that when the N-N bond becomes weak enough the cleavage of the N-N bond and the generation of *NH 2 species may take place.Fig. 6a displays the energy proles of the full pathway for NH 3 production (from state i to xxi), and the pathways of N-N bond dissociation (states iii, iv, ix, and xi) and hydrogenation (from state ii to xiii) for comparison.The rst hydrogenation step for the generation of *N 2 H (from state ii to iii) needs to overcome an energy barrier of 1.56 eV, much lower than that of direct N 2 breakage (2.53 eV).Upon the formation of *N 2 H, the N]N bonds are further weakened as indicated by the charge density differences shown in Fig. 6b.Both the hydrogenation of *N 2 H species and the cleavage of the N-N bond of *N 2 H x species are much easier than the direct dissociation of N 2 , demonstrating the unique function of H atoms for N 2 activation (Fig. 6a).Notably, the entire pathway of N 2 hydrogenation is in the presence of excessive adsorbed H atoms, and the Ru site with spillover H atoms can keep the adsorption of N 2 activated throughout as illustrated in Fig. 6a.
Subsequently, the *N 2 H species is further hydrogenated to *N 2 H 2 , *N 2 H 3 and *N 2 H 4 intermediates with energy barriers in the range of 0.11-1.20 eV.With the stepwise hydrogenation of *N 2 H x species, the energy barrier for breaking the N-N bond keeps falling (2.53 eV for *N 2 , 2.08 eV for *N 2 H, 2.06 eV for *N 2 H 2 , 0.94 eV for *N 2 H 3 and 0.83 eV for *N 2 H 4 ), showing that N 2 hydrogenation is an effective way to sharply weaken the N-N bond.Finally, the indirect cleavage of the *H 2 N-NH 2 bond takes place together with the formation of two *NH 2 species.It is noteworthy that the activation energy for *N 2 H 3 dissociation is also lower than that for *N 2 H 3 hydrogenation, thus the cleavage of the N-N bond of *N 2 H 3 is also possible for the generation of the *NH 2 intermediate.Then, the transfer of the *NH 2 species from the Ru atom to the Co atom is favorable because of the crowding at the Ru site.With active H atoms available on the Co site, further hydrogenation of *NH 2 to *NH 3 on the Co site is ready to occur.Compared to the Ru site, the Co site is more favorable for NH 3 desorption, releasing the rst NH 3 molecule to the gas phase.It is remarkable that the reaction energies for the generation of the second NH 3 on the Co sites and Ru sites are similar (À1.32 eV and À1.34 eV, Fig. 6f), while the desorption energy of *NH 3 on the Co atom (0.10 eV) is much lower than that on the Ru atom (0.66 eV), indicating that NH 3 desorption from the Co sites is more preferred in comparison to that from the Ru site.
Overall, two signicant ndings have been acquired in the present study.First, with the Ru and Co dual single-atom active site, a new route for NH 3 synthesis is made possible (Fig. 6g), which is essentially different from those of previously reported monometallic Co or Ru catalysts as well as traditional alloy clusters. 71,72According to the results of experimental and theoretical investigations, electron release and/or donation between Ru and Co atoms could promote N 2 adsorption and activation as well as hydrogenation, while the narrow d-band centers of Ru and Co allow desorption of surface intermediate species such as *N 2 H 4 and *NH 3 at low temperature.Therefore, the superior performance of RuCo DSAC in NH 3 synthesis under mild conditions is attributed to the cooperativity of dual single-atom Ru and Co centers.Second, NH 3 synthesis over RuCo DSAC proceeds via the associative pathway, similar to that occurring in the cases of metal single atoms, 24,57 clusters, 23,36 Li-promoted Ru catalysts and Co 3 Mo 3 N. 73,74 Over the RuCo DSAC in N 2 activation to NH 3 , the step with the highest kinetic barrier is the hydrogenation of N 2 to generate *N 2 H on the Ru site, which is the rate-determining step.Aer considering the entropy contribution, the entropy contribution in the RDS step is only 0.007 eV.Notably, it was experimentally observed that *N 2 H 4 is the major detected species, which can be ascribed to the fact that the stepwise hydrogenation of *N

Conclusions
To summarize, we have successfully synthesized a RuCo DSAC, in which there is integration of Ru atoms onto an atomically dispersed cobalt surface in the form of RuCo dual single-atom sites on g-C 3 N 4 .We found that the RuCo DSAC structure could effectively facilitate electron transfer from pyrrolic N to the surface Ru atom, which acts as an efficient site for N 2 adsorption and activation as well as hydrogenation.

Chemicals and materials
Melamine and ruthenium nitrosyl nitrate solution were purchased from Shanghai Aladdin Biochemical Technology Co., Ltd.Cobalt phthalocyanine (CoPc) and cyanuric acid were from Shanghai Macklin Biochemical Co., Ltd.Dimethyl sulfoxide (DMSO) and ethanol were from Sinopharm Chemical Reagent Co., Ltd.High purity argon (99.9999%) and nitrogen (99.9999%) gases were supplied by Linde Industrial Gases.The N 2 -H 2 , H 2 -Ar and CO-He mixed gases of designated proportions were also from Linde Industrial Gases.D 2 (99.999%) was purchased from Cambridge Isotope Laboratories, Inc.

Catalyst preparation
Preparation of RuCo DSAC.Typically, 0.35 mL of ruthenium nitrosyl nitrate solution, 0.201 g of CoPc and 0.50 g of melamine were mixed and dissolved in 40 mL of DMSO under ultrasonic treatment for 10 min to obtain a blue solution.Meanwhile, 0.51 g of cyanuric acid was dissolved in 10 mL of DMSO under ultrasonic treatment for 10 min to give a transparent solution.Then, the transparent solution was slowly added into the blue solution and the resulting mixture was stirred at room temperature for 10 min.Aer ltration and washing with 150 mL deionized water and 100 mL ethanol, the solid precursor was obtained.Finally, the obtained precursor was dried at 60 C for 12 h, followed by thermal polymerization at 600 C under an Ar atmosphere for 8 h in a tube furnace at a ramp rate of 1 C min À1 .
The synthetic procedure of Ru/N-C and Co/N-C SAC was similar to that of RuCo DSAC, except for the absence of CoPc and ruthenium nitrosyl nitrate solution, respectively.

NH 3 synthesis performance
Before the evaluation of catalytic performance for NH 3 synthesis, the samples (0.15 g, diluted with quartz powder in a 1 : 8 volumetric ratio) were reduced in a ow of 25% N 2 -75% H 2 at 400 C for 4 h.Under the conditions for NH 3 synthesis in a 25% N 2 -75% H 2 mixture at a WHSV of 60 000 mL g À1 h À1 and a given pressure, the outlet NH 3 concentrations were measured using a known amount of diluted H 2 SO 4 solution (1 mol L À1 ) and analyzed by ion chromatography (Thermo Scientic, DIO-NEX, ICS-600).Finally, the NH 3 synthesis rates were acquired based on the outlet NH 3 concentrations.Turnover frequency (TOF M , M ¼ Co or Ru) was acquired by dividing the NH 3 synthesis rate by the total number of Co or Ru atoms.

Experiments for methanation determination
For the determination of the possibility of methanation, 0.2 g of RuCo DSAC was exposed to a ow of 25% N 2 -75% H 2 at 400 C at a WHSV of 60 000 mL g À1 h À1 under a pressure of 1 MPa.The outlet CH 4 concentration was detected using an online GC-mass spectrometer (GCMS-QP2010 SE).

Materials characterization
Powder X-ray diffraction (XRD) was performed (at 40 kV and 40 mA) on a Panalytical X'Pert Pro diffractometer using Cu-Ka radiation (l ¼ 0.1789 nm).The Brunauer-Emmett-Teller (BET) surface area and Barrett-Joyner-Halenda (BJH) pore size distribution were measured by N 2 adsorption-desorption on a Micromeritics ASAP 2020 instrument at À196 C aer the sample was degassed at 120 C for 2 h in a vacuum.Inductively coupled plasma atomic emission spectroscopy (ICP-AES) analysis was conducted using an Ultima 2 spectrometer.Scanning electron microscopy (SEM) was performed on a Hitachi Model S-4800 microscope operated at 5 kV.Transmission electron microscopy (TEM) and high-resolution transmission electron microscopy (HR-TEM) were conducted on a JEM-2010 microscope.
Aberration-corrected high-angle annular dark-eld scanning transmission electron microscopy (HAADF-STEM) was conducted on a JEOL JEM-ARM 200 F instrument equipped with a CEOS probe corrector, with a guaranteed resolution of 0.08 nm.

Ultraviolet photoelectron spectroscopy (UPS)
UPS measurements were conducted using a helium resonance lamp which provided He I (hn ¼ 21.2 eV) and He II (hn ¼ 40.8 eV) photons (1 eV ¼ 0.16 aJ).
H 2 temperature-programmed reduction (H 2 -TPR) A H 2 -TPR experiment was performed on an AutoChem II 2920 equipped with a thermal conductivity detector (TCD), in which samples were rst pretreated under Ar ow (30 mL min À1 ) at 400 C for 0.5 h.Aer cooling to room temperature, the temperature was increased from RT to 800 C at 5 C min À1 in a gas ow of 10 vol% H 2 /Ar (30 mL min À1 ).

H 2 temperature-programmed desorption (H 2 -TPD-MS) experiment
First, the fresh RuCo DSAC was pretreated with H 2 at 400 C for 2 h and then cooled to room temperature, and then temperature-programmed desorption was performed.In this process, the signals of m/z ¼ 32 and 17 were collected using the mass spectrometer.

NEXAFS measurements
The N K-edge and Co L-edge near-edge X-ray absorption ne structure spectroscopy (NEXAFS) measurements were performed at the BL12B beamline of the Beijing Synchrotron Radiation Facility.The bending magnet was connected to the beamline, which is equipped with three gratings covering photon energies from 100 to 1000 eV with an energy resolution of ca.0.2 eV.The NEXAFS signal was detected at room temperature using the surface-sensitive total electron yield (TEY) mode by recording the sample drain current.The resolving power of the grating was typically E/DE ¼ 1000, and the photon ux was 1 Â 10 À10 photons per second.

XANES and EXAFS measurements
X-ray absorption near-edge structure (XANES) and extended Xray absorption ne structure (EXAFS) analyses were conducted at the 1W2B beamline of the Beijing Synchrotron Radiation Facility.Before the test, the sample was rstly treated with 25% N 2 -75% H 2 at 400 C for 2 h.The Co and Ru K-edge spectra of the samples and reference samples in transmission mode were measured at room temperature.A Si(111) double-crystal monochromator was used to abate the harmonic content of the monochromatic beam.
XPS measurements and Ar + etching X-ray photoelectron spectroscopy (XPS) measurements were performed on an ESCALAB 250Xi photoelectron spectrometer (Thermo Fisher Scientic) equipped with a monochromatic Al-Ka source (K a ¼ 1486.6 eV) and a charge neutralizer.The XPS binding energy was calibrated against the C 1s peak at 284.6 eV of adventitious carbon.Prior to in situ measurements, XPS spectra of the fresh sample were acquired.Argon ion etching was carried out with the MAGCIS dual mode ion source, which can be operated as a monatomic argon ion source, and the monatomic mode at an energy of 1000 eV was selected.

In situ DRIFTS deuterium labeling experiments
An in situ DRIFTS D 2 -isotopic labeling experiment was performed using a Nicolet Nexus FT-IR spectrometer.The sample was rst reduced in a 10% H 2 /Ar mixture at 400 C for 4 h.Aer collecting the background spectrum, the catalyst was exposed to a 25% N 2 -75% D 2 mixture at 400 C for different periods.

Temperature-programmed Ar desorption
An Ar-TPD-MS experiment was conducted by mass spectrometry using an Autochem 2920 instrument.Aer the activity test of NH 3 synthesis, 50 mg of the used catalyst was ushed with Ar before being heated to 700 C at a rate of 10 C min À1 .The m/z ¼ 32 (N 2 H 4 ) and 17 (NH 3 ) signals during desorption were recorded.

UV-vis absorption spectra
A UV-vis DRS experiment was performed using a PerkinElmer Lambda 750s UV-visible spectrometer.We installed a ask trap containing sulfuric acid solution and para-(dimethylamino) benzaldehyde at the exit of the reactor during the NH 3 synthesis reaction.The collected solution was then used for the UV-vis measurements.

Electron-paramagnetic resonance (EPR) measurements
EPR measurements were carried out on an E500 spectrometer (Bruker-BioSpin) with a 100 kHz magnetic eld in the X band at RT.

Computational method
First-principles calculations based on spin-polarized density functional theory (DFT) were performed using the Vienna Ab initio Simulation Package (VASP) 75 and the projected augmented wave (PAW) method. 76The generalized gradient approximation with the Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional was employed. 77The kinetic energy cutoff in planewave expansion was set as 400 eV with 2 Â 2 Â 1 Monkhorst-Pack grids involved in the Brillouin zone integration. 78The convergence thresholds of the energy change and the maximum force for the structural optimizations were set as 10 À5 eV and 0.02 eV ÅÀ1 , respectively.The energies of species were corrected with zero-point energies and the thermodynamic data of gasphase molecules were obtained from NIST (http:// cccbdb.nist.gov/).The catalyst models were constructed on C 3 N 4 with loaded metal atoms, adopting a vacuum space thickness of 15 Å.Four different structures were built and optimized as depicted in the ESI.† The energy of Co@C 3 N 4 and Ru@C 3 N 4 SACs was set as the reference and the most stable structure with the lowest energy was used for further calculations.

Fig. 3
Fig. 3 Evidence of charge transfer.(a) EPR spectra, (b and c) NEXAFS results: (b) Co L-edge and (c) N K-edge.(d-f) Electronic function location (EFL) for Ru and Co atoms on (d) graphitic N, (e) pyridinic N and (f) pyrrolic N species.(g and h) Charge density differences over (g) RuCo DSAC and (h) RuCo DSAC@N 2 .(i) Charge variation during N 2 adsorption on RuCo DSAC and Co SAC samples.
) and monometallic Ru/N-C (1.49 mmol NH 3 g cat À1 h À1 ) catalysts, respectively.It is noteworthy that the outlet CH 4 concentration over RuCo DSAC is negligibly low (Fig. S16 †) upon the NH 3 synthesis at 400 C for 25 h under 1 MPa, suggesting that under the adopted conditions the RuCo DSAC is highly stable.Moreover, because the Brunauer-Emmett-Teller (BET) surface areas of Ru/N-C (88 m 2 g À1 ) and RuCo DSAC (189 m 2 g À1 ) as determined by N 2 adsorption measurements at 77 K (Fig. S17 †) are signicantly different (Table S1 †), we obtained the surface-areanormalized NH 3 synthesis rates and made a comparison (Fig. S18 †).An NH 3 synthesis rate of 6.36 Â 10 À6 mmol m À2 s À1 was acquired at 350 C on RuCo DSAC, while that over monometallic Ru/N-C is negligible.To nd out whether the N species of the Ru-Co catalyst could be involved in NH 3 synthesis, we exposed the Ru-Co catalyst to 75% H 2 /Ar at 400 C and 1 MPa, and the cumulative amount of NH 3 as a function of time is provided in Fig. S19.† It can be seen that the NH 3 synthesis rate rst increases to peak at 60 min, and then decreases sharply with time prolonging under a 75% H 2 /Ar atmosphere.The NH 3 synthesis rate is lower than 0.1 mmol NH 3 g cat À1 h À1 aer 250 min.To further conrm that the NH 3 produced mainly
†) of the used RuCo DSAC show no Ru or Co NPs, and the AC-STEM images (Fig. S25a and b †) of the used RuCo DSAC still show a large number of individual bright dots, revealing the retention of the Ru and Co atomic dispersion even aer a stability test of 100 h.The XRD patterns (Fig. S26 †) of the used Ba/RuCo DSAC and RuCo DSAC samples indicate the absence of Ru or Co phases.One characteristic peak that is related to the g-C 3 N 4 phase can still be observed aer the NH 3 synthesis reaction, suggesting high thermal stability of g-C 3 N 4 .

Fig. 6
Fig. 6 DFT calculations of NH 3 synthesis on the RuCo DSAC.(a) Reaction pathway on the RuCo DSAC.(b-e) Charge density differences over nitrogen adsorbed on (b and c) Ru-Co DSAC, (d) Ru SAC and (e) Co SAC sites with the adsorption of H 2 .The red area shows an increase of electron density while the blue area indicates a decrease.(f) NH 3 generation on different adsorption sites of RuCo DSAC.(g) Schematic of the NH 3 synthesis reaction pathway over RuCo DSAC.
mmol NH 3 g cat À1 h À1 , which is 8.2-fold that of monometallic Co SAC (0.15 mmol NH 3 g cat À1 h À1 ).However, there is hardly any activity over Ru/N-C below 350 C. Also, the RuCo DSAC outperforms the Cs-promoted Ru/C catalyst, which is one of the most active NH 3 synthesis catalysts, by 13.6 times at 200 C (Table S6 †).At 400 C, the NH 3 synthesis rate over RuCo DSAC is 11.20 mmol NH 3 g cat À1 h À1 , which is 2.6-fold and 7.5-fold that of monometallic Co SAC (4.20 mmol NH 3 g cat À1 h À1 2 -25% 15N 2 , instead of 25% N 2 -75% H 2 .Our studies show that the NH 3 synthesis rate in the presence of 75% H 2 -25% 15N 2 (10.59 mmol NH 3 g cat À1 h À1 ) is slightly lower than that of 75% H 2 -25% N 2 (11.20 mmol NH 3 g cat À1 h À1 ) at 400 C and 1 MPa (Fig. S20 †), showing that the NH 3 produced mainly originates from the catalytic synthesis of N 2 -H 2 mixed 2 H to *N 2 H 2 , *N 2 H 3 and *N 2 H 4 has much lower energy barriers than that of the "*N 2 + H / *N 2 H" or "*N 2 H 4 / *2NH 2 " process.The net outcome is a steady state of *N 2 H 4 presence and hence easy detection of N 2 H 4 .
The RuCo DSAC with narrow d-band centers favors desorption of surface intermediate species such as *N 2 H 4 and *NH 3 at low temperature.The cooperativity of Ru and Co centers decouples the scaling relation by providing separated sites discretely for N 2 activation and *N 2 H/*NH 3 desorption, respectively.The NH 3 synthesis rate over the RuCo DSAC is about 2.6-8.2-fold that of monometallic Co at 200-400 C and 6.2-7.5-fold that of Ru at 350À400 C. It is anticipated that the cooperative roles of Ru and Co disclosed in the present study shed light on the design of dual single-atom active sites that could enable energyefficient NH 3 synthesis under mild conditions.