Overall Oxygen Electrocatalysis on Nitrogen‐Modified Carbon Catalysts: Identification of Active Sites and In Situ Observation of Reactive Intermediates

Abstract The recent mechanistic understanding of active sites, adsorbed intermediate products, and rate‐determining steps (RDS) of nitrogen (N)‐modified carbon catalysts in electrocatalytic oxygen reduction (ORR) and oxygen evolution reaction (OER) are still rife with controversy because of the inevitable coexistence of diverse N configurations and the technical limitations for the observation of formed intermediates. Herein, seven kinds of aromatic molecules with designated single N species are used as model structures to investigate the explicit role of each common N group in both ORR and OER. Specifically, dynamic evolution of active sites and key adsorbed intermediate products including O2 (ads), superoxide anion O2 −*, and OOH* are monitored with in situ spectroscopy. We propose that the formation of *OOH species from O2 −* (O2 −*+H2O→OOH*+OH−) is a possible RDS during the ORR process, whereas the generation of O2 from OOH* species is the most likely RDS during the OER process.


Introduction
Overall oxygen electrocatalysis,i ncluding oxygen reduction reaction (ORR) and oxygen evolution reaction (OER), represents the cornerstone for aw ide range of renewable electrochemical energy conversion and energy storage technologies. [1][2][3] Nitrogen (N)-modified graphitic carbon materi-als,a st he most promising metal-free catalysts in these two reactions,h ave received tremendous attention over the last decade,o wing to their abundance and high sustainability compared with metal catalysts. [4,5] Some impressive experiments and theoretical predictions regarding the possible catalytic mechanisms and active sites have been conducted by regulating the Nc oncentration in N-modified model materials or introducing the simulation models of the active structures. [6][7][8] Although pyridinic Na nd graphitic Nh ave recently been pointed out to be the most likely species that could create active sites,the connection remains controversial due to the inevitable mixing with other Nconfigurations (e.g. pyrrolic,amine,o rlactam) in doped carbon catalysts synthesized by the recent, deficient doping methods.Specifically,the involved intermediate products (e.g.a dsorbed O 2 molecule, superoxide anion O 2 À *, peroxide HO 2 *) and rate-determining steps (RDS) during the ORR and OER processes have been rarely studied by experimental approaches.T he precise structure-function relationship between active sites and reactivities is still unclear. Thei nhomogeneities of Nspecies associated with the morphology of catalysts also hamper the exploration of active sites.D etermining the genuine active sites,m onitoring the important intermediate products,a nd revealing the possible reaction pathways is therefore highly important in order to understand the catalytic nature and to optimize the design and development of new carbon-based catalysts.
To disclose the genuine role of each Nspecies in oxygen electrocatalysis,p reparing N-containing graphitic carbon materials with one single Nspecies is ideal to provide direct inference;b ut the inductive influence of surface oxygen species and defect sites (heteroatom-free) of carbon materials on various resultant Nspecies is difficult to rule out. Alternatively,aromatic organic molecules have been considered as graphene molecules at the nanoscale due to their similar pconjugated structures and properties. [9] Ap romising strategy is to use functionalized aromatic organic molecules (Figure 1a)w ith isolated Nspecies as model active matrices in order to mimic the edge structure of N-doped carbon materials and to identify the active sites. [10,11] Thee lectronic and reactive properties of active components can be tuned by extending the p-conjugated domains of these organic molecules at ad iscrete molecular level. [10] Moreover,t he intrinsic low conductivities and dispersibilities of these organic molecules can be improved by using nanocarbon materials as supports to build electrically coupled catalytic systems through non-destructive p-p interactions. [12,13] Herein, three nanocarbon supports-onion-like carbon (OLC), high temperature-treated carbon nanofiber (HHT), and carbon nanotubes (CNTs)-were chosen to support seven different model aromatic organic molecules to form model catalytic systems via as olvothermal self-assembly process (Figure 1b,h ere,O LC is used as an example in the preparation process). Ther esulting model catalytic systems not only have ahigh concentration of desired single Nspecies, but also exhibit ac lear structure-activity relationship.T he pyridinic Nspecies are demonstrated to be advantageous for both ORR and OER processes over aw ide pH range.T he ORR is dominated by ap rocess similar to the four-electronlike pathway,w hich is associated with the local conjugated structure of Nspecies.T he activities of catalysts mainly originate from electrocatalytic reactions rather than from carbon corrosion during the OER process.T he possibly adsorbed intermediates and RDS are proposed for both ORR and OER processes.

Results and Discussion
Determination of aM odel Catalytic System with aSingle NGroup As shown in Figure 1c,the results of X-ray photoelectron spectroscopy (XPS) confirm main four Nspecies:p yridinic Ngroups (398.3 eV), lactam or amide Ngroups (399.2 eV), or pyrrolic Ngroups (400.3 eV) exist in these model catalytic systems. [14] Due to the good symmetry and the narrow full width at half-maximum (FWHM, ca. 1.6 eV) of all Npeaks, we can conclude that each model catalytic system contains only one type of Nfunctional group.The content of Nspecies on the model systems ranges from 1.2 to 2.3 at %. The structural features of these model molecules on OLC support were further studied by attenuated total reflectance infrared (ATR-IR) spectroscopy ( Figure S1-S5). Them ain characteristic peaks of CB,PTD,PTA,P AD,and AD molecules can be clearly observed, meaning that the structures of various model molecules are maintained during the preparation process.T he net contents of these model molecules on OLC support are quantified to be 0.72-1.31 wt %b yu sing thermogravimetric (TG) measurements ( Figure S6).

Model Catalysts Reveal ORR Active Sites
TheO RR activity of these model catalytic systems in alkaline medium is shown in Figure 2a.O nly supported AD (AD + OLC) and PA D( PA D + OLC) model catalytic systems exhibit higher current densities and more positive onset potentials (E onset )c ompared with pure OLC and other supported model systems with respective pyrrolic N(CB + OLC), O=CÀNH (PTD + OLC) and ÀNH 2 (PTA + OLC) groups,while unsupported model molecules do not show the relevant activities ( Figure S7), indicating that pyridinic Nspecies are the possible active species in the ORR. To further investigate the role of pyridinic Nspecies,t wo other concentrations of AD molecules in AD + OLC are deliberately regulated ( Figure S6f). It can be found that the current density of AD + OLC is declining with decreased concentrations of AD molecules ( Figure S8a,b), reflecting the important role of pyridinic Nspecies.F urthermore,t he similar behavior of AD + OLC and PA D + OLC catalysts indicates that the position of the pyridinic N, noted as the edge zigzag or armchair position, negligibly affects catalytic performance.A nthracene (AT) was additionally used to mimic the zigzag configuration of pure carbon materials ( Figure S9). Thel ower performance of supported AT (AT + OLC) relative to AD + OLC further indicates the activity enhancement of pyridinic Nspecies as well as the non-critical role of edge configuration in the ORR process in alkaline medium.
To investigate the influence of the size of the p-conjugated structure for the pyridinic Nspecies on the ORR activity,two other molecules,B AD and DBAD,w ere chosen. Then et concentrations of these two molecules are determined to be 1.18 and 1.05 wt %, respectively ( Figure S10). Their molecular structures exhibit good stability as confirmed by ATR-IR (Figures S11 and S12). In the CKedge X-ray absorption nearedge structure (XANES) spectra ( Figure S13a), compared with pristine OLC,an additional peak of supported DBAD (DBAD + OLC), observed at 287.2 eV,can be assigned to the s*bond of C À N. [15] Another weak peak appeared at about 286.6 eV for DBAD + OLC which might be attributed to p* (ring) resonance of the DBAD molecule itself. [16] Other carbon-nitrogen species could not be observed. In the NK-edge XANES spectra ( Figure S13b), compared with pristine OLC, DBAD + OLC exhibits an obvious peak at 398.3 eV which can be assigned to a p*feature of ap yridinelike Nstructure, [17] suggesting that the dominant Nspecies on the surface of OLC is the expected pyridinic Ngroup.T his is further supported by XPS results of DBAD + OLC (Figure S16b), which shows that only pyridinic Nspecies (at 398.4 eV,1 .8 at %) can be observed. TheL SV results displayed in Figures 2b and S14a show that the current densities are directly proportional to the sizes of the p-conjugated structures with pyridinic Nspecies.F urther electrocatalytic results summarized in Figure 2c,e demonstrate that the size of the pconjugated system is ac ritical parameter to accelerate the ORR process;t his is indicated by the gradually enhanced intrinsic current densities and the gradual positive shift in onset potential (E onset ), which might be ascribed to the enhanced electron delocalization effect from al arger pconjugated system of model molecules.I na ddition, the reasonable catalytic stability of DBAD + OLC at two applied potentials was demonstrated by the absence of as ignificant current change for 2h( Figure S16a). To further confirm the role of pyridinic Nspecies,D BADw as removed from DBAD + OLC ( Figure S15) by ultrasonication. Thec orre- sponding activity of the residual OLC sample returns to the level of pristine OLC,which evidences the advantageous role of pyridinic Nspecies for ORR activity.
Thei ntroduction of pyridinc Nspecies significantly increases the electron transfer (ET) number from 2.4-2.7 for OLC to 2.8-3.0 for AD + OLC and can be further extended to 3.3-3.5 for DBAD + OLC at 0.2-0.7 Vv s. RHE (V RHE ) (Figure 2d,e). Thecorresponding selectivity in favor of HO 2 À production gradually decreases from 70-78 %for OLC to 50-60 %for AD + OLC,and eventually reaches 19.0-38.0 %for DBAD + OLC at 0.2-0.7 V RHE ( Figure S14b). Moreover,a s ac omparison, an AT + OLC sample with single-edge configuration (zigzag) and without any pyridinic Nspecies displays higher HO 2 À selectivity than OLC and AD + OLC (Figure S9b). All these findings indicate that (i)pyridinic Nspecies are prone to contribute the ORR in af our-electron-like manner (as illustrated by the increase in ET number and decrease in HO 2 À production);( ii)pyridinic Nspecies could improve the E onset of ORR to more positive values;( iii)the structure size of p-conjugated pyridinic Nspecies is acritical factor to facilitate the ORR process;a nd (iv) the edge structure in the absence of pyridinic Nspecies (for example, only zigzag configuration) has an egative effect on the acceleration of the four-electron pathway of ORR. Therefore, it can be expected that carbon-based catalysts with as ingle pyridinic Nfunction group in al arger p-conjugated system (e.g. nanographene) could show ac omplete four-electron pathway.
Additionally,D BAD + OLC is selected as ar epresentative catalyst to study the role of pyridinic Ni nn eutral (0.1 m phosphate buffer solution, pH = 7.0) and acidic (0.1 m H 2 SO 4 ) ORR processes.T his catalyst delivered ah igher current density than pristine OLC in neutral and acidic media (Figure 2f), which suggests that the model catalyst containing pyridinic Nspecies can exhibit an elevated catalytic performance over ab road pH range.

Model Catalysts Reveal OER Active Sites
In addition to the identification of active sites for ORR, we also unveiled the catalytic function of each Nspecies for the OER activity by evaluating the model catalytic systems in 0.1 m KOH ( Figures 3a nd S17). Similar to the ORR results, AD + OLC and PA D + OLC catalysts show higher current densities than other catalysts.T his indicates that pyridinic Nspecies are also active for OER. In contrast, pyrrolic, amine,a nd lactam species are inactive for the OER process, as demonstrated by the insignificantly improved performance of CB + OLC,PTA + OLC and PTD + OLC relative to OLC (Figure 3a). Also,i nl ine with the ORR results,t he OER process is insensitive to the edge configurations as demonstrated by the similar current densities and Tafel slopes of AD + OLC and PA D + OLC catalysts (Figures 3a and S18). By extending the p-conjugated system of the model molecules,t he OER activity of the catalysts can be increased (Figure 3b). There is ad irectly proportional relationship between the number of benzene units and theoretical TOF values,a ss hown in Figure S19. Specifically,t he value increases from 0.153 s À1 for AD + OLC to 0.638 s À1 for DBAD + OLC at 1.60 V RHE .B oth values are significantly higher than that of the reported highly active Ni/Fe-based OER catalysts (0.028-0.075 s À1 at 1.63 V RHE ). [18,19] Hence,i t can be concluded that the p-conjugated size of the carbon matrix accelerates the OER process.Moreover,similar Tafel slopes (77-85 mV dec À1 )are obtained for AD + OLC,BAD + OLC,a nd DBAD + OLC catalysts ( Figure S20), which implies that the extended p-conjugation does not change the OER reaction pathway.B yu sing the rotating ring-disk electrode (RRDE) technique ( Figure S21), the OER process occurring on DBAD + OLC is dominated by adesirable fourelectron pathway (99.9 %) with negligible peroxide intermediate formation.
Ah igh Faradaic efficiency (FE(O 2 )) of around 86 %a t two low potentials (e.g.1 .567 and 1.60 V RHE )w as obtained, which are comparable to reported metal-based catalysts to some degree. [20,21] Theloss of FE(O 2 )may be partly due to the inefficient oxygen collection by the Pt ring electrode or unavoidable carbon corrosion (e.g. CO or CO 2 formation) in alkaline medium. It should be emphasized that, when acquiring mass spectra, avery weak CO signal and an obvious O 2 signal were observed, as shown in Figure 3c,reflecting that the current density originates mainly from water oxidation rather than carbon corrosion. DBAD + OLC shows adecent stability during 2hat an applied potential ( Figure S16a). All the aforementioned results suggest that pyridinic Nspecies can efficiently accelerate both ORR and OER processes,and their catalytic performance could be regulated by the pconjugated size of active components at the molecular level. To further verify the catalytic role of pyridinic Nspecies,two other kinds of carbon (CNT and HHT) were tested as supports for DBAD.Indeed, dramatically enhanced catalytic behavior can be observed with both DBAD + HHT and DBAD + CNT catalysts for the ORR and OER processes ( Figures S22 and S23), which confirms the critical role of pyridinic Nspecies in facilitating the catalytic processes.

Doped Catalysts for OER and ORR
In general, pyridinic Nand graphitic Nare considered to be the most common species in oxygen electrocatalysis with N-doped carbon as catalysts. [8,22,23] After confirming the positive role of pyridinic Nspecies with model molecules, four kinds of N-doped OLC (NOLC) catalysts with enriched pyridinic Nbut without any graphitic Nspecies were synthesized and tested for both ORR and OER processes (Figure S24). Among them, NOLC-4 possesses the highest Ncontent (2.7 at %) and largest proportion (63.1 %) of pyridinic Nspecies in all Nspecies.A ss hown in Figure 4, by varying the contents of pyridinic Nspecies in different NOLC catalysts,t he current densities and onset potentials can be gradually manipulated for the ORR activity,w hile only the current densities are impacted for the OER activity.S ome linear relationships between current densities or ET numbers and the concentrations of pyridinic Nspecies at different applied potentials (Figure 4b,c,e) could be observed in the ORR and OER processes,r eflecting the fact that pyridinic Nspecies are the catalytically active sites in ORR and OER processes.M oreover,t he Tafel slope values (Figures 4f and  S25, 75-78 mV dec À1 )ofNOLC catalysts are analogous to the model catalytic systems (77-85 mV dec À1 ), suggesting that both model catalysts and doped catalysts have similar kinetic reaction processes.

Mechanistic ORR and OER Studies
Density functional theory calculations predict that the binding energy of some surface adsorbed intermediate oxygen species,s uch as O 2 (ads), O 2 À *, OOH*, O*, and OH*, governs the ORR activity, [24][25][26] whereas OH*, O*, and OOH* mainly dominate the OER process. [27] It is important to use effective experimental measurements to elucidate the evolution processes of surface-relevant species during electrochemical reactions.Asshown in the in situ ATR-IR spectra in Figure 5a,t he peak at 1400 cm À1 exhibits aq uasi-reversible behavior with ORR potential change,w hich is attributed to the OÀOvibration of the adsorbed oxygen molecule O 2 (ads) on the catalyst. [28] Moreover,e volution processes monitored by two other peaks at 1052 and 1019 cm À1 are observed (Figure 5b). Given that the infrared vibration peaks of O 2 À * and OOH* typically are located in the region 1100-980 cm À1 , [28][29][30][31][32][33] some control experiments were carefully introduced to further differentiate O 2 À *a nd OOH* species. Ther esults of Figure 5c reveal that (i)noo bvious peaks at 1052 and 1019 cm À1 positions can be observed in Ar-saturated KOH, suggesting these two peaks correspond to O-containing species,and (ii)ana pparent red-shift behavior (D % 26 cm À1 ) appears at 1019 cm À1 for O 2 -saturated KODelectrolyte,while the original vibration band at 1052 cm À1 is unaffected (Figures 5c and S26), indicating that the vibration band at 1019 cm À1 is most likely OOH* rather than O 2 À *, thus the vibration band at 1052 cm À1 is assigned to O 2 À *. All experiments demonstrate O 2 À *a nd OOH* species to be the important intermediates involved in the ORR process. Recently,t he formation of O 2 À *f rom O 2 (ads) coupled with the first electron transfer (that is,O 2 + e À !O 2 À *) or the generation of OOH* from O 2 À *( O 2 À * + H 2 O!OOH* + OH À )over carbon catalysts in the ORR has been considered to be the most common RDS. [34,35] To elucidate the RDS, isotopic electrochemical studies were performed, as displayed in Figure 5e.K inetic isotope effect (KIE) values of around 1.45-1.61 at 0.6-0.7 V RHE are obtained, suggesting that there is apotential primary isotope effect and the RDS of the ORR process involves ac leavage of the O À Hb ond of water molecules.Combined with the ATR-IR results,the following elementary step O 2 À * + H 2 O!OOH* + OH À is likely to be the potential RDS in the ORR. In the case of the OER, asimilar evolution behavior of OOH* species (at 1018 cm À1 ) is observed and no other oxygen species can be found in Figure 5d,i mplying that the RDS in the OER is mainly associated with the OOH* species.T ogether with the result of the primary isotope effect with the KIE values of around 1.51-1.68 at 1.567-1.6 V RHE (Figure 5e), this allows us to conclude that the generation of O 2 from OOH* species is the most possible RDS in the OER.
To gain more insight into the active sites,C ÀNvibrational bands were monitored with in situ ATR-IR spectroscopy.The red-shift of the C À Np eak at 1357 cm À1 (Figures 5f and S27) with apotential-dependent tendency suggests that active sites should originate from pyridinic Ni tself or adjacent carbon atoms during the ORR and OER. To determine the real active sites,s ome post-mortem measurements were further conducted after ORR and OER stability tests.T he XPS spectrum of DBAD + OLC after ORR ( Figure S16b) shows that the concentration of pyridinic Ndecreases and asignificant new N1s peak at 400.1 eV appears,w hich can be attributed to pyridonic Nr esulting from pyridinic Nt hrough areaction involving the carbon atoms next to pyridinic N-and O-containing species (possibly OH or OOH). [6] Therefore,i t seems that the active sites are the adjacent carbon atoms of the pyridinic Nspecies rather than pyridinic Ni tself.S imilar to the case of ORR, an impressive pyridonic Npeak appears in the XPS N1s spectrum after OER operation ( Figure S16b), which also implies that the active sites are likely the adjacent carbon atoms of the pyridinic Nspecies.A dditionally,w ed o not observe any characteristics of pyridine N-oxide species at approximately 402 eV in the N1s peak of XPS and at circa 1265 cm À1 in the ATR-IR spectra during the ORR and OER. [36] Thus,wepropose that the neighboring carbon atoms are strongly involved in the acceleration of both ORR and OER processes.

Conclusion
We have gained experimental insight into the catalytic mechanisms at am olecular level using aromatic organic molecules with designated Nspecies as models.P yridinic Nspecies play ac rucial role for the electrochemical ORR over awide pH range and alkaline OER. It can be concluded that pyridinic Nspecies are prone to facilitate the ORR process by af our-electron-like pathway,a nd they also improve the catalytic activity rather than carbon corrosion for the OER. Furthermore,t he location at edge zigzag or armchair positions of pyridinic Ndoes not obviously affect the catalytic performance.T he structure-function relationship of the active component can be established by delicately controlling their longitudinal extension (p-conjugated structures) and edge configurations.W eb elieve that OOH* (that is,H O 2 *) and/or O 2 À *s pecies are involved as the key intermediates in both ORR and OER processes,w hich was corroborated by in situ ATR-IR spectra and isotopic labeling studies.Neighboring carbon atoms of pyridinic Nspecies are likely the active sites,a sd emonstrated by the dynamic evolution of the vibration peaks.T ot he best of our knowledge,this work for the first time provides spectral evidence of the dynamic evolution of key intermediate products and the RDS of metal-free carbon catalysts in the overall oxygen electrocatalysis at amolecular level.